diff --git a/pytorch/batch.py b/pytorch/batch.py old mode 100755 new mode 100644 index db938cd..b35212d --- a/pytorch/batch.py +++ b/pytorch/batch.py @@ -1,5 +1,3 @@ -#! /bin/python3 - from data_stat import Cpu import argparse @@ -34,8 +32,20 @@ srun_args = { #'--exclusive', #'--output', '/dev/null', #'--error', '/dev/null' + ], + Cpu.EPYC_7313P: [ + '--account', 'nexus', + '--partition', 'tron', + '--qos', 'high', + '--cpus-per-task', '16', + '--ntasks-per-node', '1', + '--prefer', 'EPYC-7313P' ] } +python = { + Cpu.ALTRA: 'python3', + Cpu.EPYC_7313P: 'python3.11' +} def srun(srun_args_list: list, run_args, matrix_file: str) -> list: run_args_list = [ @@ -48,7 +58,7 @@ def srun(srun_args_list: list, run_args, matrix_file: str) -> list: run_args_list += [args.perf] if args.power is not None: run_args_list += [args.power] - return ['srun'] + srun_args_list + ['./run.py'] + run_args_list + return ['srun'] + srun_args_list + [python[args.cpu], 'run.py'] + run_args_list processes = list() diff --git a/pytorch/build.sh b/pytorch/build.sh index 159321f..b2615cf 100755 --- a/pytorch/build.sh +++ b/pytorch/build.sh @@ -1,13 +1,14 @@ #! /bin/bash -image_name="$1" +containerfile_name="$1" +image_name="${containerfile_name%.*}" if [[ -z "$1" ]]; then echo "Missing image name argument!" exit 1 fi -podman build . -t "$image_name":latest -f "$image_name".Containerfile && \ +podman build . -t "$image_name":latest -f "$containerfile_name" && \ podman save localhost/"$image_name":latest -o "$image_name".img && \ rm -fv "$image_name".sif && \ apptainer pull "$image_name".sif docker-archive:"$image_name".img diff --git a/pytorch/output_cpu/altra_10_10_ASIC_680k_10000.json b/pytorch/output_cpu/altra_10_10_ASIC_680k_10000.json new file mode 100644 index 0000000..b552439 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_ASIC_680k_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "ASIC_680k", "MATRIX_SHAPE": [682862, 682862], "MATRIX_SIZE": 466300511044, "MATRIX_NNZ": 3871773, "MATRIX_DENSITY": 8.303171256088674e-06, "TIME_S": 11.77456283569336, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.36, 20.44, 20.48, 20.72, 20.8, 21.0, 21.32, 21.32, 21.28, 21.08], "POWER": [92.0, 91.8, 78.72, 66.68, 51.2, 46.6, 53.36, 53.36, 70.48, 90.16, 100.04, 103.68, 98.2, 95.64, 97.16, 101.4], "JOULES": 938.4206715393068, "POWER_AFTER": [20.96, 20.76, 20.76, 21.08, 21.24, 21.16, 21.28, 21.2, 21.0, 21.08]} diff --git a/pytorch/output_cpu/altra_10_10_ASIC_680k_10000.output b/pytorch/output_cpu/altra_10_10_ASIC_680k_10000.output new file mode 100644 index 0000000..928b345 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_ASIC_680k_10000.output @@ -0,0 +1,26 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3471013 queued and waiting for resources +srun: job 3471013 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 3, 4, ..., 3871767, + 3871770, 3871773]), + col_indices=tensor([ 0, 11698, 11699, ..., 169456, 645874, + 682861]), + values=tensor([ 3.8333e-04, -3.3333e-04, -5.0000e-05, ..., + 0.0000e+00, 0.0000e+00, 7.9289e-02]), + size=(682862, 682862), nnz=3871773, layout=torch.sparse_csr) +tensor([0.6902, 0.5218, 0.8924, ..., 0.0864, 0.5539, 0.5194]) +Matrix: ASIC_680k +Shape: torch.Size([682862, 682862]) +Size: 466300511044 +NNZ: 3871773 +Density: 8.303171256088674e-06 +Time: 11.77456283569336 seconds + diff --git a/pytorch/output_cpu/altra_10_10_Oregon-2_10000.json b/pytorch/output_cpu/altra_10_10_Oregon-2_10000.json new file mode 100644 index 0000000..796afd3 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_Oregon-2_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "Oregon-2", "MATRIX_SHAPE": [11806, 11806], "MATRIX_SIZE": 139381636, "MATRIX_NNZ": 65460, "MATRIX_DENSITY": 0.0004696458003979807, "TIME_S": 0.9880795478820801, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [21.04, 21.12, 21.2, 21.12, 21.04, 20.96, 20.92, 20.88, 21.16, 21.08], "POWER": [25.92, 42.32, 42.32, 45.44, 45.4], "JOULES": 44.85881147384644, "POWER_AFTER": [20.72, 20.72, 20.84, 20.84, 20.84, 20.96, 20.92, 20.6, 20.68, 20.84]} diff --git a/pytorch/output_cpu/altra_10_10_Oregon-2_10000.output b/pytorch/output_cpu/altra_10_10_Oregon-2_10000.output new file mode 100644 index 0000000..d84fac7 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_Oregon-2_10000.output @@ -0,0 +1,23 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3471014 queued and waiting for resources +srun: job 3471014 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 583, 584, ..., 65459, 65460, 65460]), + col_indices=tensor([ 2, 23, 27, ..., 3324, 958, 841]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(11806, 11806), + nnz=65460, layout=torch.sparse_csr) +tensor([0.2158, 0.5422, 0.9585, ..., 0.6377, 0.8158, 0.5743]) +Matrix: Oregon-2 +Shape: torch.Size([11806, 11806]) +Size: 139381636 +NNZ: 65460 +Density: 0.0004696458003979807 +Time: 0.9880795478820801 seconds + diff --git a/pytorch/output_cpu/altra_10_10_as-caida_10000.json b/pytorch/output_cpu/altra_10_10_as-caida_10000.json new file mode 100644 index 0000000..ea58106 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_as-caida_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "as-caida", "MATRIX_SHAPE": [31379, 31379], "MATRIX_SIZE": 984641641, "MATRIX_NNZ": 106762, "MATRIX_DENSITY": 0.00010842726485909405, "TIME_S": 1.066300630569458, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.64, 20.48, 20.68, 20.64, 20.32, 20.32, 20.4, 20.2, 20.52, 20.52], "POWER": [26.32, 39.88, 50.16, 50.64, 50.24], "JOULES": 53.97094367980957, "POWER_AFTER": [20.28, 20.4, 20.2, 20.32, 20.32, 20.4, 20.48, 20.28, 20.28, 20.44]} diff --git a/pytorch/output_cpu/altra_10_10_as-caida_10000.output b/pytorch/output_cpu/altra_10_10_as-caida_10000.output new file mode 100644 index 0000000..aee2265 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_as-caida_10000.output @@ -0,0 +1,24 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3470988 queued and waiting for resources +srun: job 3470988 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 0, 0, ..., 106761, 106761, + 106762]), + col_indices=tensor([ 106, 329, 1040, ..., 155, 160, 12170]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(31379, 31379), + nnz=106762, layout=torch.sparse_csr) +tensor([0.8877, 0.6518, 0.0601, ..., 0.0372, 0.4806, 0.8853]) +Matrix: as-caida +Shape: torch.Size([31379, 31379]) +Size: 984641641 +NNZ: 106762 +Density: 0.00010842726485909405 +Time: 1.066300630569458 seconds + diff --git a/pytorch/output_cpu/altra_10_10_dc2_10000.json b/pytorch/output_cpu/altra_10_10_dc2_10000.json new file mode 100644 index 0000000..f445a04 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_dc2_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "dc2", "MATRIX_SHAPE": [116835, 116835], "MATRIX_SIZE": 13650417225, "MATRIX_NNZ": 766396, "MATRIX_DENSITY": 5.614451099680581e-05, "TIME_S": 3.0164122581481934, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.44, 20.72, 20.72, 21.0, 20.84, 21.08, 20.88, 20.8, 20.8, 20.88], "POWER": [64.4, 79.8, 83.24, 75.76, 58.2, 58.2, 56.64, 60.64, 75.88, 93.68], "JOULES": 194.69750034332276, "POWER_AFTER": [21.12, 21.0, 21.12, 20.88, 20.88, 20.84, 20.96, 20.92, 20.88, 20.8]} diff --git a/pytorch/output_cpu/altra_10_10_dc2_10000.output b/pytorch/output_cpu/altra_10_10_dc2_10000.output new file mode 100644 index 0000000..a0aa6f0 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_dc2_10000.output @@ -0,0 +1,26 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3470982 queued and waiting for resources +srun: job 3470982 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 1, 2, ..., 766390, 766394, + 766396]), + col_indices=tensor([ 0, 1, 2, ..., 116833, 89, + 116834]), + values=tensor([-1.0000e+00, -1.0000e+00, -1.0000e+00, ..., + 1.0331e+01, -1.0000e-03, 1.0000e-03]), + size=(116835, 116835), nnz=766396, layout=torch.sparse_csr) +tensor([0.3305, 0.9342, 0.6954, ..., 0.1999, 0.9064, 0.6304]) +Matrix: dc2 +Shape: torch.Size([116835, 116835]) +Size: 13650417225 +NNZ: 766396 +Density: 5.614451099680581e-05 +Time: 3.0164122581481934 seconds + diff --git a/pytorch/output_cpu/altra_10_10_de2010_10000.json b/pytorch/output_cpu/altra_10_10_de2010_10000.json new file mode 100644 index 0000000..1dfbbb2 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_de2010_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "de2010", "MATRIX_SHAPE": [24115, 24115], "MATRIX_SIZE": 581533225, "MATRIX_NNZ": 116056, "MATRIX_DENSITY": 0.0001995689928120616, "TIME_S": 1.1378686428070068, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [21.0, 20.88, 21.04, 20.8, 20.8, 20.44, 20.64, 20.48, 20.28, 20.16], "POWER": [22.84, 39.8, 49.48, 50.32, 50.28], "JOULES": 57.25203536033631, "POWER_AFTER": [20.68, 20.44, 20.68, 20.68, 20.56, 20.88, 20.92, 20.88, 21.0, 20.96]} diff --git a/pytorch/output_cpu/altra_10_10_de2010_10000.output b/pytorch/output_cpu/altra_10_10_de2010_10000.output new file mode 100644 index 0000000..97dc072 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_de2010_10000.output @@ -0,0 +1,25 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3470980 queued and waiting for resources +srun: job 3470980 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 13, 21, ..., 116047, 116051, + 116056]), + col_indices=tensor([ 250, 251, 757, ..., 23334, 23553, 24050]), + values=tensor([ 14900., 33341., 20255., ..., 164227., 52413., + 16949.]), size=(24115, 24115), nnz=116056, + layout=torch.sparse_csr) +tensor([0.3562, 0.7994, 0.9047, ..., 0.2891, 0.3611, 0.5704]) +Matrix: de2010 +Shape: torch.Size([24115, 24115]) +Size: 581533225 +NNZ: 116056 +Density: 0.0001995689928120616 +Time: 1.1378686428070068 seconds + diff --git a/pytorch/output_cpu/altra_10_10_email-Enron_10000.json b/pytorch/output_cpu/altra_10_10_email-Enron_10000.json new file mode 100644 index 0000000..d7162df --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_email-Enron_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "email-Enron", "MATRIX_SHAPE": [36692, 36692], "MATRIX_SIZE": 1346302864, "MATRIX_NNZ": 367662, "MATRIX_DENSITY": 0.0002730901120626302, "TIME_S": 1.3314027786254883, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.8, 20.64, 20.6, 20.6, 20.48, 20.8, 20.72, 20.72, 20.92, 20.92], "POWER": [28.4, 43.96, 54.4, 55.28, 55.08], "JOULES": 73.5336650466919, "POWER_AFTER": [20.88, 20.8, 20.8, 20.8, 20.64, 20.64, 20.64, 20.48, 20.52, 20.72]} diff --git a/pytorch/output_cpu/altra_10_10_email-Enron_10000.output b/pytorch/output_cpu/altra_10_10_email-Enron_10000.output new file mode 100644 index 0000000..1499d7f --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_email-Enron_10000.output @@ -0,0 +1,24 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3470985 queued and waiting for resources +srun: job 3470985 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 1, 71, ..., 367660, 367661, + 367662]), + col_indices=tensor([ 1, 0, 2, ..., 36690, 36689, 8203]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(36692, 36692), + nnz=367662, layout=torch.sparse_csr) +tensor([0.7107, 0.7540, 0.8321, ..., 0.9503, 0.7781, 0.9277]) +Matrix: email-Enron +Shape: torch.Size([36692, 36692]) +Size: 1346302864 +NNZ: 367662 +Density: 0.0002730901120626302 +Time: 1.3314027786254883 seconds + diff --git a/pytorch/output_cpu/altra_10_10_fl2010_10000.json b/pytorch/output_cpu/altra_10_10_fl2010_10000.json new file mode 100644 index 0000000..939ef0b --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_fl2010_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "fl2010", "MATRIX_SHAPE": [484481, 484481], "MATRIX_SIZE": 234721839361, "MATRIX_NNZ": 2346294, "MATRIX_DENSITY": 9.99606174861054e-06, "TIME_S": 2.924255609512329, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.8, 20.88, 20.72, 20.64, 20.56, 20.92, 20.92, 21.0, 20.96, 20.84], "POWER": [73.32, 93.24, 93.64, 82.2, 61.36, 61.36, 58.0], "JOULES": 176.3268253517151, "POWER_AFTER": [20.76, 20.56, 20.76, 20.72, 20.76, 20.76, 20.76, 20.88, 20.68, 20.68]} diff --git a/pytorch/output_cpu/altra_10_10_fl2010_10000.output b/pytorch/output_cpu/altra_10_10_fl2010_10000.output new file mode 100644 index 0000000..7b4ecc2 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_fl2010_10000.output @@ -0,0 +1,25 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3471002 queued and waiting for resources +srun: job 3471002 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 2, 5, ..., 2346288, + 2346292, 2346294]), + col_indices=tensor([ 1513, 5311, 947, ..., 484460, 482463, + 484022]), + values=tensor([28364., 12497., 11567., ..., 8532., 22622., 35914.]), + size=(484481, 484481), nnz=2346294, layout=torch.sparse_csr) +tensor([0.5561, 0.7849, 0.5628, ..., 0.5545, 0.2543, 0.1741]) +Matrix: fl2010 +Shape: torch.Size([484481, 484481]) +Size: 234721839361 +NNZ: 2346294 +Density: 9.99606174861054e-06 +Time: 2.924255609512329 seconds + diff --git a/pytorch/output_cpu/altra_10_10_ga2010_10000.json b/pytorch/output_cpu/altra_10_10_ga2010_10000.json new file mode 100644 index 0000000..19767fe --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_ga2010_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "ga2010", "MATRIX_SHAPE": [291086, 291086], "MATRIX_SIZE": 84731059396, "MATRIX_NNZ": 1418056, "MATRIX_DENSITY": 1.6735964475229304e-05, "TIME_S": 2.341104745864868, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.32, 20.28, 20.24, 20.44, 20.52, 20.8, 20.64, 20.68, 20.6, 20.36], "POWER": [33.84, 53.08, 66.2, 66.52, 67.36, 59.0], "JOULES": 154.00518000602722, "POWER_AFTER": [20.28, 20.32, 20.52, 20.6, 20.6, 20.84, 21.12, 20.96, 20.76, 20.8]} diff --git a/pytorch/output_cpu/altra_10_10_ga2010_10000.output b/pytorch/output_cpu/altra_10_10_ga2010_10000.output new file mode 100644 index 0000000..dabd461 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_ga2010_10000.output @@ -0,0 +1,25 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3470989 queued and waiting for resources +srun: job 3470989 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 3, 10, ..., 1418047, + 1418054, 1418056]), + col_indices=tensor([ 1566, 1871, 1997, ..., 291064, 289820, + 290176]), + values=tensor([18760., 17851., 18847., ..., 65219., 56729., 77629.]), + size=(291086, 291086), nnz=1418056, layout=torch.sparse_csr) +tensor([0.0746, 0.8150, 0.2560, ..., 0.7929, 0.2552, 0.7733]) +Matrix: ga2010 +Shape: torch.Size([291086, 291086]) +Size: 84731059396 +NNZ: 1418056 +Density: 1.6735964475229304e-05 +Time: 2.341104745864868 seconds + diff --git a/pytorch/output_cpu/altra_10_10_mac_econ_fwd500_10000.json b/pytorch/output_cpu/altra_10_10_mac_econ_fwd500_10000.json new file mode 100644 index 0000000..7e78fe9 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_mac_econ_fwd500_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "mac_econ_fwd500", "MATRIX_SHAPE": [206500, 206500], "MATRIX_SIZE": 42642250000, "MATRIX_NNZ": 1273389, "MATRIX_DENSITY": 2.9862143765866013e-05, "TIME_S": 1.6093401908874512, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.76, 20.72, 20.28, 20.2, 20.24, 20.56, 20.72, 21.12, 21.24, 21.0], "POWER": [48.6, 65.2, 65.2, 61.84, 62.88, 59.36], "JOULES": 99.0504337310791, "POWER_AFTER": [20.76, 20.4, 20.64, 20.68, 20.68, 20.56, 20.48, 20.68, 20.64, 20.88]} diff --git a/pytorch/output_cpu/altra_10_10_mac_econ_fwd500_10000.output b/pytorch/output_cpu/altra_10_10_mac_econ_fwd500_10000.output new file mode 100644 index 0000000..aad4fb3 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_mac_econ_fwd500_10000.output @@ -0,0 +1,26 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3471003 queued and waiting for resources +srun: job 3471003 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 3, 8, ..., 1273376, + 1273379, 1273389]), + col_indices=tensor([ 3, 30, 44, ..., 206363, 206408, + 206459]), + values=tensor([-3.7877e-03, -1.5420e-01, 9.5305e-04, ..., + 1.2290e-01, 2.2235e-01, -1.0000e+00]), + size=(206500, 206500), nnz=1273389, layout=torch.sparse_csr) +tensor([0.8982, 0.5128, 0.1053, ..., 0.5733, 0.7437, 0.9673]) +Matrix: mac_econ_fwd500 +Shape: torch.Size([206500, 206500]) +Size: 42642250000 +NNZ: 1273389 +Density: 2.9862143765866013e-05 +Time: 1.6093401908874512 seconds + diff --git a/pytorch/output_cpu/altra_10_10_mc2depi_10000.json b/pytorch/output_cpu/altra_10_10_mc2depi_10000.json new file mode 100644 index 0000000..f54e717 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_mc2depi_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "mc2depi", "MATRIX_SHAPE": [525825, 525825], "MATRIX_SIZE": 276491930625, "MATRIX_NNZ": 2100225, "MATRIX_DENSITY": 7.595972132902821e-06, "TIME_S": 2.123237371444702, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.68, 20.68, 20.68, 20.64, 20.72, 20.6, 20.84, 20.76, 20.92, 20.96], "POWER": [52.52, 76.2, 82.92, 85.4, 72.28, 58.76], "JOULES": 164.92142794609072, "POWER_AFTER": [20.68, 20.72, 20.84, 20.88, 20.84, 21.16, 21.04, 21.16, 20.88, 20.88]} diff --git a/pytorch/output_cpu/altra_10_10_mc2depi_10000.output b/pytorch/output_cpu/altra_10_10_mc2depi_10000.output new file mode 100644 index 0000000..c4f19b8 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_mc2depi_10000.output @@ -0,0 +1,25 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3470981 queued and waiting for resources +srun: job 3470981 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 2, 5, ..., 2100220, + 2100223, 2100225]), + col_indices=tensor([ 0, 1, 1, ..., 525824, 525821, + 525824]), + values=tensor([-2025., 2025., -2026., ..., 2025., 1024., -1024.]), + size=(525825, 525825), nnz=2100225, layout=torch.sparse_csr) +tensor([0.8254, 0.0543, 0.1764, ..., 0.7650, 0.8254, 0.6404]) +Matrix: mc2depi +Shape: torch.Size([525825, 525825]) +Size: 276491930625 +NNZ: 2100225 +Density: 7.595972132902821e-06 +Time: 2.123237371444702 seconds + diff --git a/pytorch/output_cpu/altra_10_10_p2p-Gnutella04_10000.json b/pytorch/output_cpu/altra_10_10_p2p-Gnutella04_10000.json new file mode 100644 index 0000000..072f297 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_p2p-Gnutella04_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "p2p-Gnutella04", "MATRIX_SHAPE": [10879, 10879], "MATRIX_SIZE": 118352641, "MATRIX_NNZ": 39994, "MATRIX_DENSITY": 0.0003379223282393842, "TIME_S": 0.9692902565002441, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.6, 20.48, 20.64, 20.64, 20.64, 20.56, 20.52, 20.44, 20.24, 20.12], "POWER": [25.92, 43.16, 50.56, 48.4, 49.28], "JOULES": 47.76662384033203, "POWER_AFTER": [20.4, 20.52, 20.44, 20.64, 20.72, 20.64, 20.8, 20.6, 20.6, 20.64]} diff --git a/pytorch/output_cpu/altra_10_10_p2p-Gnutella04_10000.output b/pytorch/output_cpu/altra_10_10_p2p-Gnutella04_10000.output new file mode 100644 index 0000000..e250b7b --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_p2p-Gnutella04_10000.output @@ -0,0 +1,23 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3471000 queued and waiting for resources +srun: job 3471000 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 10, 20, ..., 39994, 39994, 39994]), + col_indices=tensor([ 1, 2, 3, ..., 9711, 10875, 10876]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(10879, 10879), + nnz=39994, layout=torch.sparse_csr) +tensor([0.2688, 0.1431, 0.7891, ..., 0.0735, 0.7672, 0.4174]) +Matrix: p2p-Gnutella04 +Shape: torch.Size([10879, 10879]) +Size: 118352641 +NNZ: 39994 +Density: 0.0003379223282393842 +Time: 0.9692902565002441 seconds + diff --git a/pytorch/output_cpu/altra_10_10_p2p-Gnutella24_10000.json b/pytorch/output_cpu/altra_10_10_p2p-Gnutella24_10000.json new file mode 100644 index 0000000..e8ce518 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_p2p-Gnutella24_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "p2p-Gnutella24", "MATRIX_SHAPE": [26518, 26518], "MATRIX_SIZE": 703204324, "MATRIX_NNZ": 65369, "MATRIX_DENSITY": 9.295875717624285e-05, "TIME_S": 0.9848971366882324, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [16.32, 16.36, 16.36, 16.32, 16.56, 16.64, 16.72, 16.92, 16.76, 16.96], "POWER": [22.56, 40.8, 42.16, 42.16, 39.84], "JOULES": 39.23830192565919, "POWER_AFTER": [16.56, 16.44, 16.44, 16.68, 16.72, 16.72, 16.76, 16.68, 16.68, 16.92]} diff --git a/pytorch/output_cpu/altra_10_10_p2p-Gnutella24_10000.output b/pytorch/output_cpu/altra_10_10_p2p-Gnutella24_10000.output new file mode 100644 index 0000000..892cbce --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_p2p-Gnutella24_10000.output @@ -0,0 +1,23 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3471012 queued and waiting for resources +srun: job 3471012 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 9, 9, ..., 65369, 65369, 65369]), + col_indices=tensor([ 1, 2, 3, ..., 15065, 9401, 26517]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(26518, 26518), + nnz=65369, layout=torch.sparse_csr) +tensor([0.6126, 0.7089, 0.2938, ..., 0.5143, 0.3903, 0.8766]) +Matrix: p2p-Gnutella24 +Shape: torch.Size([26518, 26518]) +Size: 703204324 +NNZ: 65369 +Density: 9.295875717624285e-05 +Time: 0.9848971366882324 seconds + diff --git a/pytorch/output_cpu/altra_10_10_p2p-Gnutella25_10000.json b/pytorch/output_cpu/altra_10_10_p2p-Gnutella25_10000.json new file mode 100644 index 0000000..bef61e3 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_p2p-Gnutella25_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "p2p-Gnutella25", "MATRIX_SHAPE": [22687, 22687], "MATRIX_SIZE": 514699969, "MATRIX_NNZ": 54705, "MATRIX_DENSITY": 0.00010628522108964806, "TIME_S": 1.064000129699707, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.4, 20.68, 20.76, 20.6, 20.64, 20.48, 20.36, 20.48, 20.52, 20.52], "POWER": [33.4, 49.92, 52.44, 52.44, 51.68], "JOULES": 55.747526702880855, "POWER_AFTER": [20.96, 20.76, 20.96, 21.08, 20.64, 20.84, 20.84, 20.56, 20.28, 20.48]} diff --git a/pytorch/output_cpu/altra_10_10_p2p-Gnutella25_10000.output b/pytorch/output_cpu/altra_10_10_p2p-Gnutella25_10000.output new file mode 100644 index 0000000..2320569 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_p2p-Gnutella25_10000.output @@ -0,0 +1,23 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3470999 queued and waiting for resources +srun: job 3470999 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 9, 9, ..., 54704, 54704, 54705]), + col_indices=tensor([ 1, 2, 3, ..., 17949, 22685, 144]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(22687, 22687), + nnz=54705, layout=torch.sparse_csr) +tensor([0.1096, 0.4722, 0.2402, ..., 0.8482, 0.4609, 0.1028]) +Matrix: p2p-Gnutella25 +Shape: torch.Size([22687, 22687]) +Size: 514699969 +NNZ: 54705 +Density: 0.00010628522108964806 +Time: 1.064000129699707 seconds + diff --git a/pytorch/output_cpu/altra_10_10_p2p-Gnutella30_10000.json b/pytorch/output_cpu/altra_10_10_p2p-Gnutella30_10000.json new file mode 100644 index 0000000..b444ddf --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_p2p-Gnutella30_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "p2p-Gnutella30", "MATRIX_SHAPE": [36682, 36682], "MATRIX_SIZE": 1345569124, "MATRIX_NNZ": 88328, "MATRIX_DENSITY": 6.564359899804003e-05, "TIME_S": 1.022092580795288, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.44, 20.56, 20.76, 20.6, 20.64, 21.08, 20.76, 20.32, 20.32, 20.44], "POWER": [25.64, 36.88, 51.72, 49.6, 50.84], "JOULES": 50.723186807632445, "POWER_AFTER": [20.56, 20.68, 20.6, 20.88, 21.08, 20.76, 20.76, 20.92, 20.32, 20.24]} diff --git a/pytorch/output_cpu/altra_10_10_p2p-Gnutella30_10000.output b/pytorch/output_cpu/altra_10_10_p2p-Gnutella30_10000.output new file mode 100644 index 0000000..d00ff11 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_p2p-Gnutella30_10000.output @@ -0,0 +1,23 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3471007 queued and waiting for resources +srun: job 3471007 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 10, 10, ..., 88328, 88328, 88328]), + col_indices=tensor([ 1, 2, 3, ..., 36675, 36676, 36677]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(36682, 36682), + nnz=88328, layout=torch.sparse_csr) +tensor([0.4265, 0.5292, 0.2746, ..., 0.3064, 0.8544, 0.6969]) +Matrix: p2p-Gnutella30 +Shape: torch.Size([36682, 36682]) +Size: 1345569124 +NNZ: 88328 +Density: 6.564359899804003e-05 +Time: 1.022092580795288 seconds + diff --git a/pytorch/output_cpu/altra_10_10_ri2010_10000.json b/pytorch/output_cpu/altra_10_10_ri2010_10000.json new file mode 100644 index 0000000..7eb8f84 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_ri2010_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "ri2010", "MATRIX_SHAPE": [25181, 25181], "MATRIX_SIZE": 634082761, "MATRIX_NNZ": 125750, "MATRIX_DENSITY": 0.00019831796057928155, "TIME_S": 0.7675364017486572, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.64, 20.64, 20.64, 20.64, 20.8, 20.8, 20.8, 20.96, 20.92, 20.84], "POWER": [26.52, 43.16, 47.12, 46.0, 47.48], "JOULES": 36.442628355026244, "POWER_AFTER": [20.48, 20.44, 20.6, 20.64, 20.6, 20.68, 20.6, 20.8, 20.6, 20.6]} diff --git a/pytorch/output_cpu/altra_10_10_ri2010_10000.output b/pytorch/output_cpu/altra_10_10_ri2010_10000.output new file mode 100644 index 0000000..7d61a84 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_ri2010_10000.output @@ -0,0 +1,24 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3470987 queued and waiting for resources +srun: job 3470987 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 3, 8, ..., 125742, 125747, + 125750]), + col_indices=tensor([ 25, 56, 662, ..., 21738, 22279, 23882]), + values=tensor([17171., 37318., 5284., ..., 25993., 24918., 803.]), + size=(25181, 25181), nnz=125750, layout=torch.sparse_csr) +tensor([0.8235, 0.3045, 0.3176, ..., 0.8277, 0.2909, 0.5754]) +Matrix: ri2010 +Shape: torch.Size([25181, 25181]) +Size: 634082761 +NNZ: 125750 +Density: 0.00019831796057928155 +Time: 0.7675364017486572 seconds + diff --git a/pytorch/output_cpu/altra_10_10_rma10_10000.json b/pytorch/output_cpu/altra_10_10_rma10_10000.json new file mode 100644 index 0000000..411d0b7 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_rma10_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "rma10", "MATRIX_SHAPE": [46835, 46835], "MATRIX_SIZE": 2193517225, "MATRIX_NNZ": 2374001, "MATRIX_DENSITY": 0.0010822805369125833, "TIME_S": 2.688584089279175, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.24, 20.24, 20.4, 20.44, 20.76, 20.76, 20.68, 20.72, 20.56, 20.44], "POWER": [53.84, 65.36, 65.36, 65.6, 62.2, 50.6], "JOULES": 162.64235491752623, "POWER_AFTER": [20.28, 20.4, 20.48, 20.44, 20.4, 20.48, 20.52, 20.44, 20.44, 20.44]} diff --git a/pytorch/output_cpu/altra_10_10_rma10_10000.output b/pytorch/output_cpu/altra_10_10_rma10_10000.output new file mode 100644 index 0000000..83cecf3 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_rma10_10000.output @@ -0,0 +1,25 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3471006 queued and waiting for resources +srun: job 3471006 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 17, 34, ..., 2373939, + 2373970, 2374001]), + col_indices=tensor([ 0, 1, 2, ..., 46831, 46833, 46834]), + values=tensor([ 1.2636e+05, -1.6615e+07, -8.2015e+04, ..., + 8.3378e+01, 2.5138e+00, 1.2184e+03]), + size=(46835, 46835), nnz=2374001, layout=torch.sparse_csr) +tensor([0.3759, 0.1778, 0.4707, ..., 0.4812, 0.6721, 0.5216]) +Matrix: rma10 +Shape: torch.Size([46835, 46835]) +Size: 2193517225 +NNZ: 2374001 +Density: 0.0010822805369125833 +Time: 2.688584089279175 seconds + diff --git a/pytorch/output_cpu/altra_10_10_soc-sign-Slashdot090216_10000.json b/pytorch/output_cpu/altra_10_10_soc-sign-Slashdot090216_10000.json new file mode 100644 index 0000000..9ea8d7c --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_soc-sign-Slashdot090216_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "soc-sign-Slashdot090216", "MATRIX_SHAPE": [81871, 81871], "MATRIX_SIZE": 6702860641, "MATRIX_NNZ": 545671, "MATRIX_DENSITY": 8.140867447881048e-05, "TIME_S": 1.4809374809265137, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [21.16, 20.96, 20.92, 20.92, 20.76, 20.72, 21.04, 21.04, 21.08, 20.84], "POWER": [38.4, 56.52, 60.12, 59.64, 58.44], "JOULES": 87.74598638534546, "POWER_AFTER": [20.56, 20.56, 20.68, 20.52, 21.16, 21.16, 21.28, 21.0, 21.12, 20.84]} diff --git a/pytorch/output_cpu/altra_10_10_soc-sign-Slashdot090216_10000.output b/pytorch/output_cpu/altra_10_10_soc-sign-Slashdot090216_10000.output new file mode 100644 index 0000000..c4b74ca --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_soc-sign-Slashdot090216_10000.output @@ -0,0 +1,24 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3471015 queued and waiting for resources +srun: job 3471015 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 29, 124, ..., 545669, 545669, + 545671]), + col_indices=tensor([ 1, 2, 3, ..., 81869, 81699, 81863]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(81871, 81871), + nnz=545671, layout=torch.sparse_csr) +tensor([0.7292, 0.5775, 0.7105, ..., 0.2374, 0.7415, 0.8438]) +Matrix: soc-sign-Slashdot090216 +Shape: torch.Size([81871, 81871]) +Size: 6702860641 +NNZ: 545671 +Density: 8.140867447881048e-05 +Time: 1.4809374809265137 seconds + diff --git a/pytorch/output_cpu/altra_10_10_soc-sign-Slashdot090221_10000.json b/pytorch/output_cpu/altra_10_10_soc-sign-Slashdot090221_10000.json new file mode 100644 index 0000000..ff323c2 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_soc-sign-Slashdot090221_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "soc-sign-Slashdot090221", "MATRIX_SHAPE": [82144, 82144], "MATRIX_SIZE": 6747636736, "MATRIX_NNZ": 549202, "MATRIX_DENSITY": 8.13917555860553e-05, "TIME_S": 1.608903408050537, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.68, 20.68, 20.64, 20.28, 20.32, 20.44, 20.44, 20.44, 20.44, 20.52], "POWER": [57.2, 57.2, 72.76, 72.52, 70.32, 58.68], "JOULES": 106.05045198440551, "POWER_AFTER": [20.96, 20.76, 20.84, 20.92, 20.92, 20.96, 21.12, 21.24, 21.16, 21.04]} diff --git a/pytorch/output_cpu/altra_10_10_soc-sign-Slashdot090221_10000.output b/pytorch/output_cpu/altra_10_10_soc-sign-Slashdot090221_10000.output new file mode 100644 index 0000000..5745a5e --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_soc-sign-Slashdot090221_10000.output @@ -0,0 +1,24 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3470983 queued and waiting for resources +srun: job 3470983 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 29, 124, ..., 549200, 549200, + 549202]), + col_indices=tensor([ 1, 2, 3, ..., 82142, 81974, 82136]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(82144, 82144), + nnz=549202, layout=torch.sparse_csr) +tensor([0.2718, 0.1909, 0.9904, ..., 0.8130, 0.5743, 0.4283]) +Matrix: soc-sign-Slashdot090221 +Shape: torch.Size([82144, 82144]) +Size: 6747636736 +NNZ: 549202 +Density: 8.13917555860553e-05 +Time: 1.608903408050537 seconds + diff --git a/pytorch/output_cpu/altra_10_10_soc-sign-epinions_10000.json b/pytorch/output_cpu/altra_10_10_soc-sign-epinions_10000.json new file mode 100644 index 0000000..8b8d867 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_soc-sign-epinions_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "soc-sign-epinions", "MATRIX_SHAPE": [131828, 131828], "MATRIX_SIZE": 17378621584, "MATRIX_NNZ": 841372, "MATRIX_DENSITY": 4.841419648464106e-05, "TIME_S": 4.555854320526123, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [16.4, 16.36, 16.48, 16.68, 16.32, 16.32, 16.56, 16.56, 16.64, 16.64], "POWER": [51.6, 68.68, 77.56, 77.4, 61.4, 55.08, 54.44, 65.6], "JOULES": 284.7840434265137, "POWER_AFTER": [16.92, 16.88, 17.04, 16.92, 16.84, 16.92, 16.88, 16.8, 17.12, 17.12]} diff --git a/pytorch/output_cpu/altra_10_10_soc-sign-epinions_10000.output b/pytorch/output_cpu/altra_10_10_soc-sign-epinions_10000.output new file mode 100644 index 0000000..35af03c --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_soc-sign-epinions_10000.output @@ -0,0 +1,25 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3470998 queued and waiting for resources +srun: job 3470998 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 1, 2, ..., 841371, 841371, + 841372]), + col_indices=tensor([ 1, 128552, 3, ..., 131824, 131826, + 7714]), + values=tensor([-1., -1., 1., ..., 1., 1., 1.]), + size=(131828, 131828), nnz=841372, layout=torch.sparse_csr) +tensor([0.6727, 0.2484, 0.1189, ..., 0.2578, 0.7441, 0.8799]) +Matrix: soc-sign-epinions +Shape: torch.Size([131828, 131828]) +Size: 17378621584 +NNZ: 841372 +Density: 4.841419648464106e-05 +Time: 4.555854320526123 seconds + diff --git a/pytorch/output_cpu/altra_10_10_sx-mathoverflow_10000.json b/pytorch/output_cpu/altra_10_10_sx-mathoverflow_10000.json new file mode 100644 index 0000000..0f61308 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_sx-mathoverflow_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "sx-mathoverflow", "MATRIX_SHAPE": [24818, 24818], "MATRIX_SIZE": 615933124, "MATRIX_NNZ": 239978, "MATRIX_DENSITY": 0.00038961697406616504, "TIME_S": 1.0039293766021729, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.88, 21.0, 21.0, 20.92, 20.92, 20.8, 20.6, 20.6, 20.76, 20.92], "POWER": [29.76, 49.24, 50.6, 47.84, 47.84], "JOULES": 48.02798137664795, "POWER_AFTER": [20.96, 20.8, 20.92, 21.68, 22.4, 23.04, 23.76, 23.12, 22.6, 21.8]} diff --git a/pytorch/output_cpu/altra_10_10_sx-mathoverflow_10000.output b/pytorch/output_cpu/altra_10_10_sx-mathoverflow_10000.output new file mode 100644 index 0000000..458979b --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_sx-mathoverflow_10000.output @@ -0,0 +1,24 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3470984 queued and waiting for resources +srun: job 3470984 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 317, 416, ..., 239976, 239977, + 239978]), + col_indices=tensor([ 0, 1, 2, ..., 1483, 2179, 24817]), + values=tensor([151., 17., 6., ..., 1., 1., 1.]), + size=(24818, 24818), nnz=239978, layout=torch.sparse_csr) +tensor([0.8169, 0.9455, 0.2378, ..., 0.7183, 0.8285, 0.9774]) +Matrix: sx-mathoverflow +Shape: torch.Size([24818, 24818]) +Size: 615933124 +NNZ: 239978 +Density: 0.00038961697406616504 +Time: 1.0039293766021729 seconds + diff --git a/pytorch/output_cpu/altra_10_10_tn2010_10000.json b/pytorch/output_cpu/altra_10_10_tn2010_10000.json new file mode 100644 index 0000000..4d8c5bf --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_tn2010_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "tn2010", "MATRIX_SHAPE": [240116, 240116], "MATRIX_SIZE": 57655693456, "MATRIX_NNZ": 1193966, "MATRIX_DENSITY": 2.070855328296721e-05, "TIME_S": 2.2318568229675293, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.52, 20.52, 20.68, 20.6, 20.76, 20.84, 20.52, 20.44, 20.48, 20.4], "POWER": [47.04, 68.12, 70.92, 71.88, 71.88, 61.28], "JOULES": 157.9681861114502, "POWER_AFTER": [21.04, 20.76, 20.8, 20.72, 20.76, 20.84, 20.92, 21.04, 20.8, 20.8]} diff --git a/pytorch/output_cpu/altra_10_10_tn2010_10000.output b/pytorch/output_cpu/altra_10_10_tn2010_10000.output new file mode 100644 index 0000000..d7a77ad --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_tn2010_10000.output @@ -0,0 +1,26 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3470986 queued and waiting for resources +srun: job 3470986 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 4, 20, ..., 1193961, + 1193963, 1193966]), + col_indices=tensor([ 1152, 1272, 1961, ..., 238254, 239142, + 240113]), + values=tensor([ 5728., 2871., 418449., ..., 10058., 33324., + 34928.]), size=(240116, 240116), nnz=1193966, + layout=torch.sparse_csr) +tensor([0.2593, 0.6684, 0.1857, ..., 0.6282, 0.3314, 0.7454]) +Matrix: tn2010 +Shape: torch.Size([240116, 240116]) +Size: 57655693456 +NNZ: 1193966 +Density: 2.070855328296721e-05 +Time: 2.2318568229675293 seconds + diff --git a/pytorch/output_cpu/altra_10_10_ut2010_10000.json b/pytorch/output_cpu/altra_10_10_ut2010_10000.json new file mode 100644 index 0000000..6ba8d96 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_ut2010_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "ut2010", "MATRIX_SHAPE": [115406, 115406], "MATRIX_SIZE": 13318544836, "MATRIX_NNZ": 572066, "MATRIX_DENSITY": 4.295259032005559e-05, "TIME_S": 1.5120632648468018, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [23.36, 22.84, 22.36, 21.92, 21.48, 21.48, 21.72, 22.08, 22.64, 23.28], "POWER": [43.48, 59.4, 65.28, 65.16, 62.16], "JOULES": 96.98985254287719, "POWER_AFTER": [22.56, 22.8, 22.24, 21.84, 21.4, 21.32, 20.96, 21.28, 21.36, 21.08]} diff --git a/pytorch/output_cpu/altra_10_10_ut2010_10000.output b/pytorch/output_cpu/altra_10_10_ut2010_10000.output new file mode 100644 index 0000000..5956478 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_ut2010_10000.output @@ -0,0 +1,26 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3471001 queued and waiting for resources +srun: job 3471001 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 3, 9, ..., 572056, 572061, + 572066]), + col_indices=tensor([ 453, 1291, 1979, ..., 113521, 114509, + 114602]), + values=tensor([160642., 31335., 282373., ..., 88393., 99485., + 18651.]), size=(115406, 115406), nnz=572066, + layout=torch.sparse_csr) +tensor([0.9240, 0.3751, 0.9849, ..., 0.9377, 0.9441, 0.6765]) +Matrix: ut2010 +Shape: torch.Size([115406, 115406]) +Size: 13318544836 +NNZ: 572066 +Density: 4.295259032005559e-05 +Time: 1.5120632648468018 seconds + diff --git a/pytorch/output_cpu/altra_10_10_va2010_10000.json b/pytorch/output_cpu/altra_10_10_va2010_10000.json new file mode 100644 index 0000000..e22845a --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_va2010_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "va2010", "MATRIX_SHAPE": [285762, 285762], "MATRIX_SIZE": 81659920644, "MATRIX_NNZ": 1402128, "MATRIX_DENSITY": 1.717033263003816e-05, "TIME_S": 2.1484014987945557, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.76, 20.72, 20.76, 20.88, 20.88, 20.96, 20.96, 20.96, 20.8, 20.6], "POWER": [65.16, 84.16, 87.88, 82.08, 64.16, 59.44], "JOULES": 155.0609850883484, "POWER_AFTER": [20.52, 20.52, 20.72, 20.56, 20.64, 20.64, 20.72, 20.92, 21.16, 21.32]} diff --git a/pytorch/output_cpu/altra_10_10_va2010_10000.output b/pytorch/output_cpu/altra_10_10_va2010_10000.output new file mode 100644 index 0000000..de218b7 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_va2010_10000.output @@ -0,0 +1,26 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3471004 queued and waiting for resources +srun: job 3471004 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 2, 8, ..., 1402119, + 1402123, 1402128]), + col_indices=tensor([ 2006, 2464, 1166, ..., 285581, 285634, + 285760]), + values=tensor([125334., 3558., 1192., ..., 10148., 1763., + 9832.]), size=(285762, 285762), nnz=1402128, + layout=torch.sparse_csr) +tensor([0.5972, 0.8492, 0.1772, ..., 0.7912, 0.0415, 0.8296]) +Matrix: va2010 +Shape: torch.Size([285762, 285762]) +Size: 81659920644 +NNZ: 1402128 +Density: 1.717033263003816e-05 +Time: 2.1484014987945557 seconds + diff --git a/pytorch/output_cpu/altra_10_10_vt2010_10000.json b/pytorch/output_cpu/altra_10_10_vt2010_10000.json new file mode 100644 index 0000000..78aeae3 --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_vt2010_10000.json @@ -0,0 +1 @@ +{"CPU": "ALTRA", "ITERATIONS": 10000, "MATRIX_FILE": "vt2010", "MATRIX_SHAPE": [32580, 32580], "MATRIX_SIZE": 1061456400, "MATRIX_NNZ": 155598, "MATRIX_DENSITY": 0.00014658915806621921, "TIME_S": 0.8885588645935059, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [20.84, 20.84, 20.84, 20.92, 20.96, 20.88, 20.72, 20.52, 20.28, 20.28], "POWER": [24.52, 34.36, 45.4, 48.68, 47.56], "JOULES": 42.25985960006714, "POWER_AFTER": [20.36, 20.48, 20.56, 20.8, 21.08, 21.08, 21.28, 21.6, 21.68, 21.48]} diff --git a/pytorch/output_cpu/altra_10_10_vt2010_10000.output b/pytorch/output_cpu/altra_10_10_vt2010_10000.output new file mode 100644 index 0000000..996ee9d --- /dev/null +++ b/pytorch/output_cpu/altra_10_10_vt2010_10000.output @@ -0,0 +1,24 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: ################################################################################ +srun: # Please note that the oasis compute nodes have aarch64 architecture CPUs. # +srun: # All submission nodes and all other compute nodes have x86_64 architecture # +srun: # CPUs. Programs, environments, or other software that was built on x86_64 # +srun: # nodes may need to be rebuilt to properly execute on these nodes. # +srun: ################################################################################ +srun: job 3471005 queued and waiting for resources +srun: job 3471005 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at /space/jenkins/workspace/Releases/pytorch-dls/pytorch-dls/aten/src/ATen/SparseCsrTensorImpl.cpp:55.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 4, 7, ..., 155588, 155592, + 155598]), + col_indices=tensor([ 131, 561, 996, ..., 32237, 32238, 32570]), + values=tensor([79040., 7820., 15136., ..., 2828., 17986., 2482.]), + size=(32580, 32580), nnz=155598, layout=torch.sparse_csr) +tensor([0.7980, 0.7955, 0.8301, ..., 0.2464, 0.9642, 0.0961]) +Matrix: vt2010 +Shape: torch.Size([32580, 32580]) +Size: 1061456400 +NNZ: 155598 +Density: 0.00014658915806621921 +Time: 0.8885588645935059 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_ASIC_680k_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_ASIC_680k_10000.json new file mode 100644 index 0000000..af8b627 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_ASIC_680k_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "ASIC_680k", "MATRIX_SHAPE": [682862, 682862], "MATRIX_SIZE": 466300511044, "MATRIX_NNZ": 3871773, "MATRIX_DENSITY": 8.303171256088674e-06, "TIME_S": 7.5851967334747314, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [41.12, 39.22, 38.67, 39.0, 39.11, 39.2, 39.03, 39.06, 39.93, 38.51], "POWER": [122.77], "JOULES": 931.2346029686928, "POWER_AFTER": [40.16, 38.97, 38.8, 39.29, 39.44, 38.77, 39.27, 38.71, 38.69, 38.72]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_ASIC_680k_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_ASIC_680k_10000.output new file mode 100644 index 0000000..02b19c9 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_ASIC_680k_10000.output @@ -0,0 +1,20 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470899 queued and waiting for resources +srun: job 3470899 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 3, 4, ..., 3871767, + 3871770, 3871773]), + col_indices=tensor([ 0, 11698, 11699, ..., 169456, 645874, + 682861]), + values=tensor([ 3.8333e-04, -3.3333e-04, -5.0000e-05, ..., + 0.0000e+00, 0.0000e+00, 7.9289e-02]), + size=(682862, 682862), nnz=3871773, layout=torch.sparse_csr) +tensor([0.6720, 0.5431, 0.4163, ..., 0.4625, 0.4662, 0.2085]) +Matrix: ASIC_680k +Shape: torch.Size([682862, 682862]) +Size: 466300511044 +NNZ: 3871773 +Density: 8.303171256088674e-06 +Time: 7.5851967334747314 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_Oregon-2_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_Oregon-2_10000.json new file mode 100644 index 0000000..a5dcb6a --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_Oregon-2_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "Oregon-2", "MATRIX_SHAPE": [11806, 11806], "MATRIX_SIZE": 139381636, "MATRIX_NNZ": 65460, "MATRIX_DENSITY": 0.0004696458003979807, "TIME_S": 0.4882948398590088, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [41.11, 38.63, 39.35, 38.39, 39.53, 38.33, 39.4, 39.34, 42.37, 41.56], "POWER": [78.62], "JOULES": 38.38974030971527, "POWER_AFTER": [41.57, 38.36, 39.18, 38.33, 39.47, 38.52, 39.07, 38.29, 39.18, 38.38]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_Oregon-2_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_Oregon-2_10000.output new file mode 100644 index 0000000..018245e --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_Oregon-2_10000.output @@ -0,0 +1,17 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470898 queued and waiting for resources +srun: job 3470898 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 583, 584, ..., 65459, 65460, 65460]), + col_indices=tensor([ 2, 23, 27, ..., 3324, 958, 841]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(11806, 11806), + nnz=65460, layout=torch.sparse_csr) +tensor([0.6755, 0.7426, 0.3350, ..., 0.5898, 0.3954, 0.3897]) +Matrix: Oregon-2 +Shape: torch.Size([11806, 11806]) +Size: 139381636 +NNZ: 65460 +Density: 0.0004696458003979807 +Time: 0.4882948398590088 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_as-caida_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_as-caida_10000.json new file mode 100644 index 0000000..827b10e --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_as-caida_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "as-caida", "MATRIX_SHAPE": [31379, 31379], "MATRIX_SIZE": 984641641, "MATRIX_NNZ": 106762, "MATRIX_DENSITY": 0.00010842726485909405, "TIME_S": 0.6748511791229248, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [40.0, 38.56, 38.3, 38.46, 39.39, 38.44, 38.81, 38.3, 38.45, 38.62], "POWER": [80.47], "JOULES": 54.30527438402176, "POWER_AFTER": [40.22, 38.5, 39.18, 38.29, 39.13, 38.27, 38.85, 38.25, 38.39, 38.34]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_as-caida_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_as-caida_10000.output new file mode 100644 index 0000000..604b029 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_as-caida_10000.output @@ -0,0 +1,18 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470879 queued and waiting for resources +srun: job 3470879 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 0, 0, ..., 106761, 106761, + 106762]), + col_indices=tensor([ 106, 329, 1040, ..., 155, 160, 12170]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(31379, 31379), + nnz=106762, layout=torch.sparse_csr) +tensor([0.6412, 0.3070, 0.9642, ..., 0.0959, 0.1216, 0.7825]) +Matrix: as-caida +Shape: torch.Size([31379, 31379]) +Size: 984641641 +NNZ: 106762 +Density: 0.00010842726485909405 +Time: 0.6748511791229248 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_dc2_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_dc2_10000.json new file mode 100644 index 0000000..b9a980e --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_dc2_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "dc2", "MATRIX_SHAPE": [116835, 116835], "MATRIX_SIZE": 13650417225, "MATRIX_NNZ": 766396, "MATRIX_DENSITY": 5.614451099680581e-05, "TIME_S": 2.0699713230133057, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [39.82, 38.4, 39.36, 38.95, 39.48, 38.39, 39.32, 38.43, 39.06, 38.43], "POWER": [98.5], "JOULES": 203.8921753168106, "POWER_AFTER": [39.63, 39.36, 38.51, 38.63, 38.49, 39.69, 38.57, 39.3, 38.49, 39.44]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_dc2_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_dc2_10000.output new file mode 100644 index 0000000..f90c1ff --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_dc2_10000.output @@ -0,0 +1,20 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470885 queued and waiting for resources +srun: job 3470885 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 1, 2, ..., 766390, 766394, + 766396]), + col_indices=tensor([ 0, 1, 2, ..., 116833, 89, + 116834]), + values=tensor([-1.0000e+00, -1.0000e+00, -1.0000e+00, ..., + 1.0331e+01, -1.0000e-03, 1.0000e-03]), + size=(116835, 116835), nnz=766396, layout=torch.sparse_csr) +tensor([0.9489, 0.1111, 0.7586, ..., 0.1064, 0.9062, 0.5747]) +Matrix: dc2 +Shape: torch.Size([116835, 116835]) +Size: 13650417225 +NNZ: 766396 +Density: 5.614451099680581e-05 +Time: 2.0699713230133057 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_de2010_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_de2010_10000.json new file mode 100644 index 0000000..ab2718e --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_de2010_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "de2010", "MATRIX_SHAPE": [24115, 24115], "MATRIX_SIZE": 581533225, "MATRIX_NNZ": 116056, "MATRIX_DENSITY": 0.0001995689928120616, "TIME_S": 0.5970535278320312, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [39.59, 38.0, 38.9, 38.03, 39.01, 37.93, 38.85, 37.93, 38.11, 38.06], "POWER": [79.47], "JOULES": 47.447843856811524, "POWER_AFTER": [40.4, 38.96, 38.62, 38.82, 38.75, 38.73, 38.33, 38.84, 38.46, 39.16]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_de2010_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_de2010_10000.output new file mode 100644 index 0000000..19c0d6c --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_de2010_10000.output @@ -0,0 +1,19 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470877 queued and waiting for resources +srun: job 3470877 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 13, 21, ..., 116047, 116051, + 116056]), + col_indices=tensor([ 250, 251, 757, ..., 23334, 23553, 24050]), + values=tensor([ 14900., 33341., 20255., ..., 164227., 52413., + 16949.]), size=(24115, 24115), nnz=116056, + layout=torch.sparse_csr) +tensor([0.4656, 0.5143, 0.3514, ..., 0.7050, 0.9241, 0.6135]) +Matrix: de2010 +Shape: torch.Size([24115, 24115]) +Size: 581533225 +NNZ: 116056 +Density: 0.0001995689928120616 +Time: 0.5970535278320312 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_email-Enron_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_email-Enron_10000.json new file mode 100644 index 0000000..69cf030 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_email-Enron_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "email-Enron", "MATRIX_SHAPE": [36692, 36692], "MATRIX_SIZE": 1346302864, "MATRIX_NNZ": 367662, "MATRIX_DENSITY": 0.0002730901120626302, "TIME_S": 1.2558205127716064, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [39.74, 38.27, 38.91, 38.72, 38.52, 38.36, 38.35, 39.81, 38.58, 38.28], "POWER": [88.07], "JOULES": 110.60011255979538, "POWER_AFTER": [40.62, 39.93, 43.98, 39.02, 38.58, 38.9, 38.4, 38.52, 38.45, 38.36]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_email-Enron_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_email-Enron_10000.output new file mode 100644 index 0000000..fed87ab --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_email-Enron_10000.output @@ -0,0 +1,18 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470880 queued and waiting for resources +srun: job 3470880 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 1, 71, ..., 367660, 367661, + 367662]), + col_indices=tensor([ 1, 0, 2, ..., 36690, 36689, 8203]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(36692, 36692), + nnz=367662, layout=torch.sparse_csr) +tensor([0.0951, 0.9410, 0.9223, ..., 0.8038, 0.1166, 0.7786]) +Matrix: email-Enron +Shape: torch.Size([36692, 36692]) +Size: 1346302864 +NNZ: 367662 +Density: 0.0002730901120626302 +Time: 1.2558205127716064 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_fl2010_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_fl2010_10000.json new file mode 100644 index 0000000..9a15ca0 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_fl2010_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "fl2010", "MATRIX_SHAPE": [484481, 484481], "MATRIX_SIZE": 234721839361, "MATRIX_NNZ": 2346294, "MATRIX_DENSITY": 9.99606174861054e-06, "TIME_S": 3.8482837677001953, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [39.85, 44.42, 38.66, 39.02, 38.56, 38.59, 39.21, 38.44, 38.64, 38.76], "POWER": [123.85], "JOULES": 476.60994462966914, "POWER_AFTER": [41.95, 38.59, 39.04, 38.7, 38.69, 38.66, 39.8, 38.57, 39.54, 38.67]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_fl2010_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_fl2010_10000.output new file mode 100644 index 0000000..739be80 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_fl2010_10000.output @@ -0,0 +1,19 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470889 queued and waiting for resources +srun: job 3470889 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 2, 5, ..., 2346288, + 2346292, 2346294]), + col_indices=tensor([ 1513, 5311, 947, ..., 484460, 482463, + 484022]), + values=tensor([28364., 12497., 11567., ..., 8532., 22622., 35914.]), + size=(484481, 484481), nnz=2346294, layout=torch.sparse_csr) +tensor([0.2616, 0.5904, 0.5539, ..., 0.1315, 0.7299, 0.8588]) +Matrix: fl2010 +Shape: torch.Size([484481, 484481]) +Size: 234721839361 +NNZ: 2346294 +Density: 9.99606174861054e-06 +Time: 3.8482837677001953 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_ga2010_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_ga2010_10000.json new file mode 100644 index 0000000..e3f12ca --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_ga2010_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "ga2010", "MATRIX_SHAPE": [291086, 291086], "MATRIX_SIZE": 84731059396, "MATRIX_NNZ": 1418056, "MATRIX_DENSITY": 1.6735964475229304e-05, "TIME_S": 2.374833583831787, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [40.39, 38.45, 39.02, 38.45, 39.0, 38.2, 39.33, 38.48, 39.15, 40.07], "POWER": [110.89], "JOULES": 263.34529611110685, "POWER_AFTER": [39.66, 39.39, 38.41, 39.31, 38.38, 38.89, 38.48, 39.38, 38.53, 39.22]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_ga2010_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_ga2010_10000.output new file mode 100644 index 0000000..ceb943b --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_ga2010_10000.output @@ -0,0 +1,19 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470878 queued and waiting for resources +srun: job 3470878 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 3, 10, ..., 1418047, + 1418054, 1418056]), + col_indices=tensor([ 1566, 1871, 1997, ..., 291064, 289820, + 290176]), + values=tensor([18760., 17851., 18847., ..., 65219., 56729., 77629.]), + size=(291086, 291086), nnz=1418056, layout=torch.sparse_csr) +tensor([0.2127, 0.1840, 0.5883, ..., 0.9651, 0.0622, 0.2931]) +Matrix: ga2010 +Shape: torch.Size([291086, 291086]) +Size: 84731059396 +NNZ: 1418056 +Density: 1.6735964475229304e-05 +Time: 2.374833583831787 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_mac_econ_fwd500_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_mac_econ_fwd500_10000.json new file mode 100644 index 0000000..51d2df3 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_mac_econ_fwd500_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "mac_econ_fwd500", "MATRIX_SHAPE": [206500, 206500], "MATRIX_SIZE": 42642250000, "MATRIX_NNZ": 1273389, "MATRIX_DENSITY": 2.9862143765866013e-05, "TIME_S": 1.166548252105713, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [40.4, 38.82, 39.17, 38.67, 39.62, 39.37, 38.77, 38.83, 39.09, 38.67], "POWER": [96.59], "JOULES": 112.67689567089081, "POWER_AFTER": [39.82, 39.34, 39.56, 38.78, 38.54, 39.44, 38.58, 39.51, 38.79, 39.36]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_mac_econ_fwd500_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_mac_econ_fwd500_10000.output new file mode 100644 index 0000000..74c0f20 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_mac_econ_fwd500_10000.output @@ -0,0 +1,20 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470890 queued and waiting for resources +srun: job 3470890 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 3, 8, ..., 1273376, + 1273379, 1273389]), + col_indices=tensor([ 3, 30, 44, ..., 206363, 206408, + 206459]), + values=tensor([-3.7877e-03, -1.5420e-01, 9.5305e-04, ..., + 1.2290e-01, 2.2235e-01, -1.0000e+00]), + size=(206500, 206500), nnz=1273389, layout=torch.sparse_csr) +tensor([0.2225, 0.9184, 0.8891, ..., 0.8781, 0.9920, 0.8523]) +Matrix: mac_econ_fwd500 +Shape: torch.Size([206500, 206500]) +Size: 42642250000 +NNZ: 1273389 +Density: 2.9862143765866013e-05 +Time: 1.166548252105713 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_mc2depi_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_mc2depi_10000.json new file mode 100644 index 0000000..b0ff3f7 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_mc2depi_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "mc2depi", "MATRIX_SHAPE": [525825, 525825], "MATRIX_SIZE": 276491930625, "MATRIX_NNZ": 2100225, "MATRIX_DENSITY": 7.595972132902821e-06, "TIME_S": 1.4909443855285645, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [40.83, 38.41, 39.81, 39.67, 44.4, 38.42, 38.51, 39.33, 39.34, 39.73], "POWER": [105.92], "JOULES": 157.92082931518556, "POWER_AFTER": [41.76, 38.57, 38.98, 38.45, 39.54, 39.61, 44.62, 39.62, 39.65, 38.4]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_mc2depi_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_mc2depi_10000.output new file mode 100644 index 0000000..0da53ce --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_mc2depi_10000.output @@ -0,0 +1,19 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470883 queued and waiting for resources +srun: job 3470883 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 2, 5, ..., 2100220, + 2100223, 2100225]), + col_indices=tensor([ 0, 1, 1, ..., 525824, 525821, + 525824]), + values=tensor([-2025., 2025., -2026., ..., 2025., 1024., -1024.]), + size=(525825, 525825), nnz=2100225, layout=torch.sparse_csr) +tensor([0.1112, 0.0723, 0.0629, ..., 0.0188, 0.2120, 0.5563]) +Matrix: mc2depi +Shape: torch.Size([525825, 525825]) +Size: 276491930625 +NNZ: 2100225 +Density: 7.595972132902821e-06 +Time: 1.4909443855285645 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella04_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella04_10000.json new file mode 100644 index 0000000..08dcfed --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella04_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "p2p-Gnutella04", "MATRIX_SHAPE": [10879, 10879], "MATRIX_SIZE": 118352641, "MATRIX_NNZ": 39994, "MATRIX_DENSITY": 0.0003379223282393842, "TIME_S": 0.3917062282562256, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [40.7, 38.58, 38.9, 38.56, 44.06, 38.54, 38.5, 38.53, 39.05, 38.74], "POWER": [78.01], "JOULES": 30.55700286626816, "POWER_AFTER": [39.85, 39.15, 38.49, 38.63, 38.95, 38.93, 38.57, 41.29, 43.97, 38.61]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella04_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella04_10000.output new file mode 100644 index 0000000..23b38e4 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella04_10000.output @@ -0,0 +1,17 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470891 queued and waiting for resources +srun: job 3470891 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 10, 20, ..., 39994, 39994, 39994]), + col_indices=tensor([ 1, 2, 3, ..., 9711, 10875, 10876]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(10879, 10879), + nnz=39994, layout=torch.sparse_csr) +tensor([0.7836, 0.6102, 0.9911, ..., 0.3070, 0.4164, 0.1677]) +Matrix: p2p-Gnutella04 +Shape: torch.Size([10879, 10879]) +Size: 118352641 +NNZ: 39994 +Density: 0.0003379223282393842 +Time: 0.3917062282562256 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella24_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella24_10000.json new file mode 100644 index 0000000..2896e4d --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella24_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "p2p-Gnutella24", "MATRIX_SHAPE": [26518, 26518], "MATRIX_SIZE": 703204324, "MATRIX_NNZ": 65369, "MATRIX_DENSITY": 9.295875717624285e-05, "TIME_S": 0.5904901027679443, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [41.69, 38.5, 38.66, 38.85, 39.3, 38.41, 39.31, 38.3, 39.22, 38.45], "POWER": [78.57], "JOULES": 46.39480737447738, "POWER_AFTER": [39.95, 38.46, 44.94, 39.14, 39.06, 39.03, 38.93, 39.28, 38.52, 38.92]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella24_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella24_10000.output new file mode 100644 index 0000000..2557a07 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella24_10000.output @@ -0,0 +1,17 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470896 queued and waiting for resources +srun: job 3470896 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 9, 9, ..., 65369, 65369, 65369]), + col_indices=tensor([ 1, 2, 3, ..., 15065, 9401, 26517]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(26518, 26518), + nnz=65369, layout=torch.sparse_csr) +tensor([0.6457, 0.2082, 0.9929, ..., 0.5035, 0.5783, 0.4428]) +Matrix: p2p-Gnutella24 +Shape: torch.Size([26518, 26518]) +Size: 703204324 +NNZ: 65369 +Density: 9.295875717624285e-05 +Time: 0.5904901027679443 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella25_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella25_10000.json new file mode 100644 index 0000000..2f91672 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella25_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "p2p-Gnutella25", "MATRIX_SHAPE": [22687, 22687], "MATRIX_SIZE": 514699969, "MATRIX_NNZ": 54705, "MATRIX_DENSITY": 0.00010628522108964806, "TIME_S": 0.5463590621948242, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [40.5, 38.94, 39.13, 38.5, 38.9, 39.01, 38.41, 38.75, 38.72, 38.83], "POWER": [80.15], "JOULES": 43.79067883491516, "POWER_AFTER": [40.41, 39.07, 39.98, 38.86, 38.61, 39.01, 39.29, 38.36, 38.7, 39.03]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella25_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella25_10000.output new file mode 100644 index 0000000..ebb9072 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella25_10000.output @@ -0,0 +1,17 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470893 queued and waiting for resources +srun: job 3470893 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 9, 9, ..., 54704, 54704, 54705]), + col_indices=tensor([ 1, 2, 3, ..., 17949, 22685, 144]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(22687, 22687), + nnz=54705, layout=torch.sparse_csr) +tensor([0.1041, 0.8814, 0.8390, ..., 0.9768, 0.6919, 0.6912]) +Matrix: p2p-Gnutella25 +Shape: torch.Size([22687, 22687]) +Size: 514699969 +NNZ: 54705 +Density: 0.00010628522108964806 +Time: 0.5463590621948242 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella30_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella30_10000.json new file mode 100644 index 0000000..b283dff --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella30_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "p2p-Gnutella30", "MATRIX_SHAPE": [36682, 36682], "MATRIX_SIZE": 1345569124, "MATRIX_NNZ": 88328, "MATRIX_DENSITY": 6.564359899804003e-05, "TIME_S": 0.5678565502166748, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [40.93, 39.31, 38.47, 39.16, 38.63, 39.42, 38.44, 39.46, 38.84, 39.32], "POWER": [80.0], "JOULES": 45.428524017333984, "POWER_AFTER": [42.33, 38.87, 38.51, 39.86, 38.5, 39.57, 38.64, 39.3, 38.59, 39.52]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella30_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella30_10000.output new file mode 100644 index 0000000..c28b6a9 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_p2p-Gnutella30_10000.output @@ -0,0 +1,17 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470892 queued and waiting for resources +srun: job 3470892 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 10, 10, ..., 88328, 88328, 88328]), + col_indices=tensor([ 1, 2, 3, ..., 36675, 36676, 36677]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(36682, 36682), + nnz=88328, layout=torch.sparse_csr) +tensor([0.8398, 0.4628, 0.1152, ..., 0.1515, 0.0358, 0.8190]) +Matrix: p2p-Gnutella30 +Shape: torch.Size([36682, 36682]) +Size: 1345569124 +NNZ: 88328 +Density: 6.564359899804003e-05 +Time: 0.5678565502166748 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_ri2010_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_ri2010_10000.json new file mode 100644 index 0000000..ad9b1c9 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_ri2010_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "ri2010", "MATRIX_SHAPE": [25181, 25181], "MATRIX_SIZE": 634082761, "MATRIX_NNZ": 125750, "MATRIX_DENSITY": 0.00019831796057928155, "TIME_S": 0.610253095626831, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [39.97, 38.69, 38.83, 38.7, 38.39, 38.32, 38.45, 38.47, 38.36, 38.28], "POWER": [80.37], "JOULES": 49.04604129552841, "POWER_AFTER": [40.01, 38.51, 38.35, 38.59, 39.52, 38.74, 38.53, 38.37, 38.81, 39.15]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_ri2010_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_ri2010_10000.output new file mode 100644 index 0000000..533b0f1 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_ri2010_10000.output @@ -0,0 +1,18 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470884 queued and waiting for resources +srun: job 3470884 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 3, 8, ..., 125742, 125747, + 125750]), + col_indices=tensor([ 25, 56, 662, ..., 21738, 22279, 23882]), + values=tensor([17171., 37318., 5284., ..., 25993., 24918., 803.]), + size=(25181, 25181), nnz=125750, layout=torch.sparse_csr) +tensor([0.8353, 0.9273, 0.0726, ..., 0.3513, 0.9132, 0.6466]) +Matrix: ri2010 +Shape: torch.Size([25181, 25181]) +Size: 634082761 +NNZ: 125750 +Density: 0.00019831796057928155 +Time: 0.610253095626831 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_rma10_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_rma10_10000.json new file mode 100644 index 0000000..2264e68 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_rma10_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "rma10", "MATRIX_SHAPE": [46835, 46835], "MATRIX_SIZE": 2193517225, "MATRIX_NNZ": 2374001, "MATRIX_DENSITY": 0.0010822805369125833, "TIME_S": 3.9620909690856934, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [40.62, 38.96, 38.47, 39.5, 38.54, 39.4, 38.42, 39.25, 39.24, 38.47], "POWER": [112.26], "JOULES": 444.78433218955996, "POWER_AFTER": [41.23, 39.08, 39.0, 39.24, 38.82, 39.26, 38.75, 40.54, 45.02, 39.59]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_rma10_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_rma10_10000.output new file mode 100644 index 0000000..a52f402 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_rma10_10000.output @@ -0,0 +1,19 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470888 queued and waiting for resources +srun: job 3470888 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 17, 34, ..., 2373939, + 2373970, 2374001]), + col_indices=tensor([ 0, 1, 2, ..., 46831, 46833, 46834]), + values=tensor([ 1.2636e+05, -1.6615e+07, -8.2015e+04, ..., + 8.3378e+01, 2.5138e+00, 1.2184e+03]), + size=(46835, 46835), nnz=2374001, layout=torch.sparse_csr) +tensor([0.0694, 0.3886, 0.4209, ..., 0.6373, 0.6766, 0.6929]) +Matrix: rma10 +Shape: torch.Size([46835, 46835]) +Size: 2193517225 +NNZ: 2374001 +Density: 0.0010822805369125833 +Time: 3.9620909690856934 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-Slashdot090216_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-Slashdot090216_10000.json new file mode 100644 index 0000000..acd2056 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-Slashdot090216_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "soc-sign-Slashdot090216", "MATRIX_SHAPE": [81871, 81871], "MATRIX_SIZE": 6702860641, "MATRIX_NNZ": 545671, "MATRIX_DENSITY": 8.140867447881048e-05, "TIME_S": 1.620380163192749, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [41.09, 39.77, 38.98, 38.92, 39.2, 38.85, 38.8, 38.42, 38.67, 38.84], "POWER": [95.89], "JOULES": 155.3782538485527, "POWER_AFTER": [41.04, 38.44, 38.51, 39.43, 38.9, 38.66, 38.49, 38.69, 40.35, 38.43]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-Slashdot090216_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-Slashdot090216_10000.output new file mode 100644 index 0000000..1a4ac52 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-Slashdot090216_10000.output @@ -0,0 +1,18 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470897 queued and waiting for resources +srun: job 3470897 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 29, 124, ..., 545669, 545669, + 545671]), + col_indices=tensor([ 1, 2, 3, ..., 81869, 81699, 81863]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(81871, 81871), + nnz=545671, layout=torch.sparse_csr) +tensor([0.6601, 0.6699, 0.4597, ..., 0.4006, 0.0724, 0.2095]) +Matrix: soc-sign-Slashdot090216 +Shape: torch.Size([81871, 81871]) +Size: 6702860641 +NNZ: 545671 +Density: 8.140867447881048e-05 +Time: 1.620380163192749 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-Slashdot090221_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-Slashdot090221_10000.json new file mode 100644 index 0000000..3694b0b --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-Slashdot090221_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "soc-sign-Slashdot090221", "MATRIX_SHAPE": [82144, 82144], "MATRIX_SIZE": 6747636736, "MATRIX_NNZ": 549202, "MATRIX_DENSITY": 8.13917555860553e-05, "TIME_S": 1.6988587379455566, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [39.74, 38.23, 38.66, 38.34, 38.69, 38.27, 38.61, 38.22, 38.36, 38.7], "POWER": [95.19], "JOULES": 161.71436326503752, "POWER_AFTER": [39.47, 38.44, 38.74, 38.72, 38.97, 38.52, 38.32, 38.61, 38.31, 38.32]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-Slashdot090221_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-Slashdot090221_10000.output new file mode 100644 index 0000000..5482408 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-Slashdot090221_10000.output @@ -0,0 +1,18 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470876 queued and waiting for resources +srun: job 3470876 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 29, 124, ..., 549200, 549200, + 549202]), + col_indices=tensor([ 1, 2, 3, ..., 82142, 81974, 82136]), + values=tensor([1., 1., 1., ..., 1., 1., 1.]), size=(82144, 82144), + nnz=549202, layout=torch.sparse_csr) +tensor([0.5845, 0.4829, 0.3749, ..., 0.6026, 0.8058, 0.2362]) +Matrix: soc-sign-Slashdot090221 +Shape: torch.Size([82144, 82144]) +Size: 6747636736 +NNZ: 549202 +Density: 8.13917555860553e-05 +Time: 1.6988587379455566 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-epinions_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-epinions_10000.json new file mode 100644 index 0000000..997180b --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-epinions_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "soc-sign-epinions", "MATRIX_SHAPE": [131828, 131828], "MATRIX_SIZE": 17378621584, "MATRIX_NNZ": 841372, "MATRIX_DENSITY": 4.841419648464106e-05, "TIME_S": 2.818403720855713, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [40.15, 38.33, 39.01, 38.43, 38.82, 38.58, 39.15, 39.79, 38.39, 38.52], "POWER": [101.75], "JOULES": 286.7725785970688, "POWER_AFTER": [40.36, 38.72, 38.87, 38.41, 38.45, 38.49, 39.37, 38.38, 38.82, 38.92]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-epinions_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-epinions_10000.output new file mode 100644 index 0000000..ece723e --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_soc-sign-epinions_10000.output @@ -0,0 +1,19 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470886 queued and waiting for resources +srun: job 3470886 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 1, 2, ..., 841371, 841371, + 841372]), + col_indices=tensor([ 1, 128552, 3, ..., 131824, 131826, + 7714]), + values=tensor([-1., -1., 1., ..., 1., 1., 1.]), + size=(131828, 131828), nnz=841372, layout=torch.sparse_csr) +tensor([0.2911, 0.5946, 0.7956, ..., 0.3632, 0.5862, 0.9286]) +Matrix: soc-sign-epinions +Shape: torch.Size([131828, 131828]) +Size: 17378621584 +NNZ: 841372 +Density: 4.841419648464106e-05 +Time: 2.818403720855713 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_sx-mathoverflow_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_sx-mathoverflow_10000.json new file mode 100644 index 0000000..6ff876d --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_sx-mathoverflow_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "sx-mathoverflow", "MATRIX_SHAPE": [24818, 24818], "MATRIX_SIZE": 615933124, "MATRIX_NNZ": 239978, "MATRIX_DENSITY": 0.00038961697406616504, "TIME_S": 0.9492478370666504, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [39.95, 38.76, 38.89, 38.34, 38.41, 38.56, 38.77, 38.39, 43.82, 38.34], "POWER": [85.3], "JOULES": 80.97084050178528, "POWER_AFTER": [40.98, 38.86, 38.38, 38.62, 38.93, 38.52, 39.17, 38.59, 38.99, 38.46]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_sx-mathoverflow_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_sx-mathoverflow_10000.output new file mode 100644 index 0000000..f0fdd59 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_sx-mathoverflow_10000.output @@ -0,0 +1,18 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470882 queued and waiting for resources +srun: job 3470882 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 317, 416, ..., 239976, 239977, + 239978]), + col_indices=tensor([ 0, 1, 2, ..., 1483, 2179, 24817]), + values=tensor([151., 17., 6., ..., 1., 1., 1.]), + size=(24818, 24818), nnz=239978, layout=torch.sparse_csr) +tensor([0.5148, 0.6679, 0.4736, ..., 0.5604, 0.1954, 0.1745]) +Matrix: sx-mathoverflow +Shape: torch.Size([24818, 24818]) +Size: 615933124 +NNZ: 239978 +Density: 0.00038961697406616504 +Time: 0.9492478370666504 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_tn2010_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_tn2010_10000.json new file mode 100644 index 0000000..567a51c --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_tn2010_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "tn2010", "MATRIX_SHAPE": [240116, 240116], "MATRIX_SIZE": 57655693456, "MATRIX_NNZ": 1193966, "MATRIX_DENSITY": 2.070855328296721e-05, "TIME_S": 1.7836709022521973, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [40.32, 39.68, 39.86, 39.48, 38.52, 38.29, 40.37, 38.34, 39.45, 38.27], "POWER": [106.23], "JOULES": 189.47935994625092, "POWER_AFTER": [41.93, 39.25, 38.74, 39.37, 39.14, 39.52, 38.5, 38.58, 38.4, 39.3]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_tn2010_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_tn2010_10000.output new file mode 100644 index 0000000..c8e88f5 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_tn2010_10000.output @@ -0,0 +1,20 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470881 queued and waiting for resources +srun: job 3470881 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 4, 20, ..., 1193961, + 1193963, 1193966]), + col_indices=tensor([ 1152, 1272, 1961, ..., 238254, 239142, + 240113]), + values=tensor([ 5728., 2871., 418449., ..., 10058., 33324., + 34928.]), size=(240116, 240116), nnz=1193966, + layout=torch.sparse_csr) +tensor([0.0655, 0.4633, 0.1355, ..., 0.7193, 0.0926, 0.7299]) +Matrix: tn2010 +Shape: torch.Size([240116, 240116]) +Size: 57655693456 +NNZ: 1193966 +Density: 2.070855328296721e-05 +Time: 1.7836709022521973 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_ut2010_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_ut2010_10000.json new file mode 100644 index 0000000..5a910a7 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_ut2010_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "ut2010", "MATRIX_SHAPE": [115406, 115406], "MATRIX_SIZE": 13318544836, "MATRIX_NNZ": 572066, "MATRIX_DENSITY": 4.295259032005559e-05, "TIME_S": 0.7757325172424316, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [39.73, 39.13, 38.38, 39.43, 38.36, 39.7, 43.3, 38.81, 38.54, 39.24], "POWER": [90.41], "JOULES": 70.13397688388824, "POWER_AFTER": [40.58, 38.4, 39.23, 38.86, 38.48, 38.28, 39.25, 38.5, 40.62, 44.86]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_ut2010_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_ut2010_10000.output new file mode 100644 index 0000000..0f866b2 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_ut2010_10000.output @@ -0,0 +1,20 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470887 queued and waiting for resources +srun: job 3470887 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 3, 9, ..., 572056, 572061, + 572066]), + col_indices=tensor([ 453, 1291, 1979, ..., 113521, 114509, + 114602]), + values=tensor([160642., 31335., 282373., ..., 88393., 99485., + 18651.]), size=(115406, 115406), nnz=572066, + layout=torch.sparse_csr) +tensor([0.5921, 0.3895, 0.8812, ..., 0.2001, 0.1496, 0.7049]) +Matrix: ut2010 +Shape: torch.Size([115406, 115406]) +Size: 13318544836 +NNZ: 572066 +Density: 4.295259032005559e-05 +Time: 0.7757325172424316 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_va2010_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_va2010_10000.json new file mode 100644 index 0000000..7c1a790 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_va2010_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "va2010", "MATRIX_SHAPE": [285762, 285762], "MATRIX_SIZE": 81659920644, "MATRIX_NNZ": 1402128, "MATRIX_DENSITY": 1.717033263003816e-05, "TIME_S": 2.389526844024658, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [42.21, 38.46, 39.09, 39.12, 38.6, 38.8, 38.74, 38.74, 38.51, 38.8], "POWER": [112.07], "JOULES": 267.7942734098434, "POWER_AFTER": [41.44, 38.92, 38.62, 39.01, 38.95, 38.72, 39.78, 38.59, 38.49, 38.56]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_va2010_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_va2010_10000.output new file mode 100644 index 0000000..aa37cd6 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_va2010_10000.output @@ -0,0 +1,20 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470895 queued and waiting for resources +srun: job 3470895 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 2, 8, ..., 1402119, + 1402123, 1402128]), + col_indices=tensor([ 2006, 2464, 1166, ..., 285581, 285634, + 285760]), + values=tensor([125334., 3558., 1192., ..., 10148., 1763., + 9832.]), size=(285762, 285762), nnz=1402128, + layout=torch.sparse_csr) +tensor([0.7826, 0.8027, 0.4606, ..., 0.4410, 0.5591, 0.5693]) +Matrix: va2010 +Shape: torch.Size([285762, 285762]) +Size: 81659920644 +NNZ: 1402128 +Density: 1.717033263003816e-05 +Time: 2.389526844024658 seconds + diff --git a/pytorch/output_cpu/epyc_7313p_10_10_vt2010_10000.json b/pytorch/output_cpu/epyc_7313p_10_10_vt2010_10000.json new file mode 100644 index 0000000..c68bb46 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_vt2010_10000.json @@ -0,0 +1 @@ +{"CPU": "EPYC_7313P", "ITERATIONS": 10000, "MATRIX_FILE": "vt2010", "MATRIX_SHAPE": [32580, 32580], "MATRIX_SIZE": 1061456400, "MATRIX_NNZ": 155598, "MATRIX_DENSITY": 0.00014658915806621921, "TIME_S": 0.8104038238525391, "BASELINE_TIME_S": 10, "BASELINE_DELAY_S": 10, "POWER_BEFORE": [41.86, 43.27, 40.2, 39.3, 38.73, 39.46, 39.11, 39.26, 38.8, 38.85], "POWER": [81.98], "JOULES": 66.43690547943116, "POWER_AFTER": [41.6, 38.57, 39.45, 38.33, 39.52, 44.29, 38.68, 38.83, 39.46, 38.86]} diff --git a/pytorch/output_cpu/epyc_7313p_10_10_vt2010_10000.output b/pytorch/output_cpu/epyc_7313p_10_10_vt2010_10000.output new file mode 100644 index 0000000..b5a2376 --- /dev/null +++ b/pytorch/output_cpu/epyc_7313p_10_10_vt2010_10000.output @@ -0,0 +1,18 @@ +srun: Job time limit was unset; set to partition default of 60 minutes +srun: job 3470894 queued and waiting for resources +srun: job 3470894 has been allocated resources +/nfshomes/vut/ampere_research/pytorch/spmv.py:22: UserWarning: Sparse CSR tensor support is in beta state. If you miss a functionality in the sparse tensor support, please submit a feature request to https://github.com/pytorch/pytorch/issues. (Triggered internally at ../aten/src/ATen/SparseCsrTensorImpl.cpp:53.) + ).to_sparse_csr().type(torch.float) +tensor(crow_indices=tensor([ 0, 4, 7, ..., 155588, 155592, + 155598]), + col_indices=tensor([ 131, 561, 996, ..., 32237, 32238, 32570]), + values=tensor([79040., 7820., 15136., ..., 2828., 17986., 2482.]), + size=(32580, 32580), nnz=155598, layout=torch.sparse_csr) +tensor([0.7387, 0.6923, 0.8988, ..., 0.9956, 0.0628, 0.8750]) +Matrix: vt2010 +Shape: torch.Size([32580, 32580]) +Size: 1061456400 +NNZ: 155598 +Density: 0.00014658915806621921 +Time: 0.8104038238525391 seconds + diff --git a/pytorch/power.py b/pytorch/power.py old mode 100755 new mode 100644 index 57764ac..6cf6674 --- a/pytorch/power.py +++ b/pytorch/power.py @@ -9,6 +9,7 @@ import argparse, subprocess import sys parser = argparse.ArgumentParser() +parser.add_argument('program', nargs='*') parser.add_argument('-s', '--seconds', type=int) args = parser.parse_args() @@ -19,11 +20,35 @@ if arch == "aarch64": i = 0 while i != upper_bound: proc = subprocess.Popen(['sensors'], stdout=subprocess.PIPE) - proc = subprocess.Popen(['awk', '/CPU power:/ {print $3}'], stdin=proc.stdout, stdout=subprocess.PIPE, text=True) + proc = subprocess.Popen( + ['awk', '/CPU power:/ {print $3; exit}'], + stdin=proc.stdout, + stdout=subprocess.PIPE, + text=True) power = proc.communicate()[0].strip().split('\n')[0] print(power) i += 1 time.sleep(1) elif arch == "x86_64": - pass + if args.seconds is None: + proc = subprocess.Popen( + ['turbostat', '-s', 'PkgWatt'] + args.program, + stdout=subprocess.DEVNULL, + stderr=subprocess.PIPE) + proc = subprocess.Popen( + ['sed', '-n', '/PkgWatt/{n;p}'], + stdin=proc.stderr, + stdout=subprocess.PIPE, + text=True) + else: + proc = subprocess.Popen( + ['turbostat', '-s', 'PkgWatt', '-n', str(args.seconds), '-i', '1'], + stdout=subprocess.PIPE) + proc = subprocess.Popen( + ['sed', '-n', '/PkgWatt/{n;p}'], + stdin=proc.stdout, + stdout=subprocess.PIPE, + text=True) + power = proc.communicate()[0].strip().split('\n') + print('\n'.join(power)) diff --git a/pytorch/power.sh.bak b/pytorch/power.sh.bak index 288c3dd..072c431 100644 --- a/pytorch/power.sh.bak +++ b/pytorch/power.sh.bak @@ -6,23 +6,26 @@ arch=$(uname -m) if [[ $arch = aarch64 ]]; then iter=0 function aarch64_power { - sensors | awk '/CPU power:/ {printf "Socket"++count[$1] " "; print $3}' + sensors | awk '/CPU power:/ {print $3; exit}' ((iter++)) sleep 1s } if [[ -z "$baseline_time_s" ]]; then - while true; do - aarch64_power - done - else - while [[ "$iter" -ne "$baseline_time_s" ]]; do - aarch64_power - done + baseline_time_s=-1 fi + while [[ "$iter" -ne "$baseline_time_s" ]]; do + aarch64_power + done elif [[ $arch = x86_64 ]]; then if [[ -z "$baseline_time_s" ]]; then - turbostat -s PkgWatt -i 1 2>/dev/null + #turbostat -s PkgWatt -i 1 2>/dev/null | awk -F: '/PkgWatt/ {getline; print $0}' + #turbostat -s PkgWatt -i 1 | awk '/PkgWatt/ {getline; print $0}' + turbostat -s PkgWatt -i 1 | sed -n "/PkgWatt/{n;p}" else - turbostat -s PkgWatt -n "$baseline_time_s" -i 1 2>/dev/null + #turbostat -s PkgWatt -n "$baseline_time_s" -i 1 2>/dev/null | awk -F: '/PkgWatt/ {getline; print $0}' + turbostat -s PkgWatt -n "$baseline_time_s" -i 1 | sed -n "/PkgWatt/{n;p}" fi +else + echo "Unrecognized arch!" + exit 1 fi diff --git a/pytorch/pytorch-altra.sif b/pytorch/pytorch-altra.sif index e1048ac..af7093a 100755 Binary files a/pytorch/pytorch-altra.sif and b/pytorch/pytorch-altra.sif differ diff --git a/pytorch/pytorch-epyc_7313p.Containerfile b/pytorch/pytorch-epyc_7313p.Containerfile new file mode 100644 index 0000000..9aa4f7b --- /dev/null +++ b/pytorch/pytorch-epyc_7313p.Containerfile @@ -0,0 +1,7 @@ +FROM --platform=linux/amd64 ubuntu:22.04 + +RUN apt-get update -y \ + && apt-get install -y python3 python3-pip git vim wget \ + && rm -rf /var/lib/apt/lists/* +RUN pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu +RUN pip install scipy diff --git a/pytorch/run.py b/pytorch/run.py old mode 100755 new mode 100644 index 5475b3f..9d2acc2 --- a/pytorch/run.py +++ b/pytorch/run.py @@ -1,5 +1,3 @@ -#! /bin/python3 - import data_stat from data_stat import Stat, Cpu @@ -21,11 +19,18 @@ parser.add_argument('-d', '--debug', action='store_true') args = parser.parse_args() args.cpu = Cpu[args.cpu.upper()] +python = { + Cpu.ALTRA: 'python3', + Cpu.EPYC_7313P: 'python3.11' +} program = { Cpu.ALTRA: [ 'apptainer', 'run', 'pytorch-altra.sif', '-c', 'numactl --cpunodebind=0 --membind=0 ' - + f'python spmv.py {args.matrix_file} {args.iterations}'] + + f'python spmv.py {args.matrix_file} {args.iterations}'], + Cpu.EPYC_7313P: [ + 'apptainer', 'run', 'pytorch-epyc_7313p.sif', + 'python3', 'spmv.py', f'{args.matrix_file}', f'{args.iterations}'] } perf = ['perf', 'stat'] perf_args = { @@ -37,12 +42,14 @@ perf_args = { ['-M', 'l2_cache_miss_ratio,l2_tlb_miss_ratio,ll_cache_read_miss_ratio']] } -def baseline_power(baseline_time_s: int) -> list: - power_process = subprocess.Popen(['./power.py', '-s', str(baseline_time_s)], +def baseline_power(cpu: Cpu, baseline_time_s: int) -> list: + power_process = subprocess.Popen([python[args.cpu], 'power.py', '-s', str(baseline_time_s)], stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True) - return [float(x) for x in power_process.communicate()[0].strip().split('\n')] + return [float(x) for x in power_process.communicate()[0].strip().split('\n') if len(x) != 0] def run_program(program: list[str]) -> tuple[dict, str]: + if args.debug: + print(program) process = subprocess.run(program, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True) if args.debug: @@ -53,36 +60,54 @@ def run_program(program: list[str]) -> tuple[dict, str]: result = dict() result[Stat.CPU.name] = args.cpu.name result[Stat.ITERATIONS.name] = args.iterations -result[Stat.BASELINE_TIME_S.name] = args.baseline_time_s -result[Stat.BASELINE_DELAY_S.name] = args.baseline_delay_s + +program_result = run_program(program[args.cpu]) +result |= program_result[0] +print(program_result[1], file=sys.stderr) if args.power: + result[Stat.BASELINE_TIME_S.name] = args.baseline_time_s + result[Stat.BASELINE_DELAY_S.name] = args.baseline_delay_s + time.sleep(args.baseline_delay_s) - result[Stat.POWER_BEFORE.name] = baseline_power(args.baseline_time_s) + result[Stat.POWER_BEFORE.name] = baseline_power(args.cpu, args.baseline_time_s) if args.debug: print(result) run_program(program[args.cpu]) # Warmup - power_process = subprocess.Popen(['./power.py'], - stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True) - result = result | run_program(program[args.cpu])[0] + if args.cpu == Cpu.ALTRA: + power_process = subprocess.Popen( + [python[args.cpu], 'power.py'], + stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True) - power_process.send_signal(signal.SIGINT) - if args.debug: - print(result) + run_program(program[args.cpu])[0] + + power_process.send_signal(signal.SIGINT) + if args.debug: + print(result) + + result[Stat.POWER.name] = [float(x) for x in power_process.communicate()[0].strip().split('\n')] + # Riemann Sum across the last (s) power recordings. + from math import ceil + result[Stat.JOULES.name] = ( + sum(result[Stat.POWER.name][-ceil(result[Stat.TIME_S.name]):-1]) + + (result[Stat.POWER.name][-1] * (result[Stat.TIME_S.name] % 1))) + + elif args.cpu == Cpu.EPYC_7313P: + power_process = subprocess.Popen( + [python[args.cpu], 'power.py'] + program[args.cpu], + stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True) + result[Stat.POWER.name] = [float(x) for x in power_process.communicate()[0].strip().split('\n')] + result[Stat.JOULES.name] = result[Stat.POWER.name][0] * result[Stat.TIME_S.name] - result[Stat.POWER.name] = [float(x) for x in power_process.communicate()[0].strip().split('\n')] - # Riemann Sum - from math import ceil - result[Stat.JOULES.name] = sum(result[Stat.POWER.name][-ceil(result[Stat.TIME_S.name]):-1]) + (result[Stat.POWER.name][-1] * (result[Stat.TIME_S.name] % 1)) if args.debug: print(result) #print(len(result['power'])) #print(sum(result['power']) / len(result['power'])) time.sleep(args.baseline_delay_s) - result[Stat.POWER_AFTER.name] = baseline_power(args.baseline_time_s) + result[Stat.POWER_AFTER.name] = baseline_power(args.cpu, args.baseline_time_s) if args.debug: print(result) diff --git a/pytorch/spmv.py b/pytorch/spmv.py index 3eeb948..a367ba7 100644 --- a/pytorch/spmv.py +++ b/pytorch/spmv.py @@ -28,8 +28,7 @@ print(vector, file=sys.stderr) start = time.time() for i in range(0, args.iterations): - torch.sparse.mm(matrix, vector.unsqueeze(-1)).squeeze(-1) - #print(i) + torch.mv(matrix, vector) end = time.time() result = dict()