2 x Intel Xeon Platinum 8490H oneDNN Intel AMX Sapphire Rapids benchmarks by Michael Larabel for a future article.
Compare your own system(s) to this result file with the
Phoronix Test Suite by running the command:
phoronix-test-suite benchmark 2301158-NE-ONEDNN30I77
oneDNN 3.0 Intel Sapphire Rapids AMX,
"oneDNN 3.0 - Harness: Deconvolution Batch shapes_3d - Data Type: bf16bf16bf16 - Engine: CPU",
Lower Results Are Better
"AVX512_CORE_AMX",0.442804,0.442051,0.444081,0.454079,0.444313,0.444061,0.453546,0.444452,0.442623
"AVX512_CORE_FP16",1.0407,1.03533,1.05636,1.03691,1.03315,1.0611,1.04354,1.03934,1.05952
"AVX512_CORE_BF16",1.03924,1.06412,1.04035,1.03686,1.03753,1.0783,1.03853,1.03707,1.06325
"AVX512_CORE_VNNI",2.37068,2.36473,2.41982,2.37507,2.35953,2.3737,2.41956,2.36196,2.36525
"AVX512_CORE",2.36408,2.41903,2.36968,2.36377,2.36896,2.41584,2.36915,2.3639,2.41812
"oneDNN 3.0 - Harness: Deconvolution Batch shapes_1d - Data Type: bf16bf16bf16 - Engine: CPU",
Lower Results Are Better
"AVX512_CORE_AMX",0.598145,0.568969,0.578906,0.576562
"AVX512_CORE_FP16",2.16436,2.17986,2.16797
"AVX512_CORE_BF16",2.17364,2.15568,2.18395
"AVX512_CORE_VNNI",3.04038,3.06944,3.09182
"AVX512_CORE",3.01676,3.10484,3.04677
"oneDNN 3.0 - Harness: Recurrent Neural Network Inference - Data Type: bf16bf16bf16 - Engine: CPU",
Lower Results Are Better
"AVX512_CORE_AMX",855.989,822.356,864.188,858.851
"AVX512_CORE_FP16",906.543,868.845,856.447,906.774,963.025,839.224,813.465,845.587,916.872,836.305,887.77,862.545
"AVX512_CORE_BF16",869.591,805.891,889.484,853.384,902.207,860.682,885.356,842.606,943.673,845.705,856.681,781.77,855.483,883.267,891.109
"AVX512_CORE_VNNI",893.748,853.166,803.985,803.615,851.034,831.766,870.874,959.132,900.387,786.044,889.629,877.707,851.889,945.646,817.687
"AVX512_CORE",881.903,783.223,846.883,855.181,811.966,852.868,765.159,858.671,878.966,901.05,886.481,881.708,862.903,880.562,905.912
"CPU Temperature Monitor - Phoronix Test Suite System Monitoring",
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"CPU Power Consumption Monitor - Phoronix Test Suite System Monitoring",
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"CPU Peak Freq (Highest CPU Core Frequency) Monitor - Phoronix Test Suite System Monitoring",
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Temperature Monitor",
Lower Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Power Consumption Monitor",
Lower Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Peak Freq (Highest CPU Core Frequency) Monitor",
Higher Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Temperature Monitor",
Lower Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Power Consumption Monitor",
Lower Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Peak Freq (Highest CPU Core Frequency) Monitor",
Higher Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - Harness: Matrix Multiply Batch Shapes Transformer - Data Type: bf16bf16bf16 - Engine: CPU",
Lower Results Are Better
"AVX512_CORE_AMX",0.327209,0.336893,0.325312,0.330428
"AVX512_CORE_FP16",0.383548,0.379863,0.388771,0.382956
"AVX512_CORE_BF16",0.393076,0.392035,0.383093,0.378341
"AVX512_CORE_VNNI",16.4146,16.5115,16.4002,16.9741
"AVX512_CORE",18.6439,15.2465,14.0044,16.7507,17.5147,16.9855,18.6119,14.8792,16.9141,16.9466,16.9602,13.7749,16.4758,16.5884,16.587
"oneDNN 3.0 - CPU Temperature Monitor",
Lower Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Power Consumption Monitor",
Lower Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Peak Freq (Highest CPU Core Frequency) Monitor",
Higher Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - Harness: IP Shapes 1D - Data Type: bf16bf16bf16 - Engine: CPU",
Lower Results Are Better
"AVX512_CORE_AMX",5.96669,4.78922,5.3854,5.84821,5.9136,5.51881,5.35714,5.75629,6.04431,6.17752,5.32039,5.51307,5.89329,5.95679,6.61687
"AVX512_CORE_FP16",6.47326,6.25167,8.48586,6.51583,7.46053,6.84247,5.51401,6.00881,5.4975,6.27394,5.14176,5.60164,5.63936,6.87232,8.67594
"AVX512_CORE_BF16",5.13188,5.70124,9.49014,5.23353,6.25089,7.32988,6.12513,5.63245,7.09837,5.03615,7.24469,6.68187,8.7799,5.80968,7.2803
"AVX512_CORE_VNNI",6.15586,5.52694,6.03825,7.32378,5.69321,5.84726,6.39206,8.36741,8.99012,6.71035,8.03225,6.48089,5.78387,5.96548,9.0572
"AVX512_CORE",5.1982,5.47376,5.98196,6.4422,6.35259,7.79575,6.5481,6.97536,7.85299,6.49428,6.19513,6.67782,6.12459,6.36077,9.24897
"oneDNN 3.0 - CPU Temperature Monitor",
Lower Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Power Consumption Monitor",
Lower Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Peak Freq (Highest CPU Core Frequency) Monitor",
Higher Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Temperature Monitor",
Lower Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Power Consumption Monitor",
Lower Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",
"oneDNN 3.0 - CPU Peak Freq (Highest CPU Core Frequency) Monitor",
Higher Results Are Better
"AVX512_CORE_AMX",
"AVX512_CORE_FP16",
"AVX512_CORE_BF16",
"AVX512_CORE_VNNI",
"AVX512_CORE",