adding performance data

Author: guacamoleo

doc/performance/clBLAS_2.6.0/S9150/README.txt

@@ -0,0 +1,35 @@
+################################
+#                              #
+#   Benchmarking Methodology   #
+#                              #
+################################
+
+############
+# Hardware #
+############
+S9150
+
+############
+# Software #
+############
+CentOS 6.6
+clBLAS 2.6.0
+driver 14.502
+
+############
+# Settings #
+############
+gpu clocks: set to max level using proprietary tool though public alternatives exist
+clBLAS:
+  m=n=k=lda=ldb=ldc (for simplicity)
+  alpha=beta=1
+  gemms were column-major, op(A,B)=N,T
+
+############
+# Sampling #
+############
+For each data point, we took 10 samples. Each sample consists of 10 gemm calls with a wait afterward. Outlying samples beyond 1 standard deviation were removed (rarely if ever did this actually need to happen). Before running the 10 samples, one warm-up sample was executed (but not included in the stastics).
+
+GFlop/s was calculated as
+(2*m*n*k flops) / (host time for 10 kernels / 10) // real data
+(8*m*n*k flops) / (host time for 10 kernels / 10) // complex data

doc/performance/clBLAS_2.6.0/S9150/dgemm_32.csv

@@ -0,0 +1,61 @@
+m,n,k,lda,ldb,ldc,offa,offb,offc,alpha,beta,order,transa,transb,side,uplo,diag,function,device,library,label,GFLOPS
+96,96,96,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,62.8587
+192,192,192,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,290.018
+288,288,288,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,592.678
+384,384,384,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,611.778
+480,480,480,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,751.024
+576,576,576,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,937.285
+672,672,672,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,909.679
+768,768,768,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1177.76
+864,864,864,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1494.86
+960,960,960,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1261.75
+1056,1056,1056,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1521.25
+1152,1152,1152,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1434.59
+1248,1248,1248,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1677.97
+1344,1344,1344,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1585.51
+1440,1440,1440,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1547.27
+1536,1536,1536,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1702.19
+1632,1632,1632,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1712.63
+1728,1728,1728,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1691.28
+1824,1824,1824,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1688.98
+1920,1920,1920,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1679.2
+2016,2016,2016,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1706.24
+2112,2112,2112,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1846.23
+2208,2208,2208,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1738.61
+2304,2304,2304,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1735.26
+2400,2400,2400,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1780.01
+2496,2496,2496,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1803.69
+2592,2592,2592,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1829.92
+2688,2688,2688,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1852.93
+2784,2784,2784,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1807.14
+2880,2880,2880,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1837.23
+2976,2976,2976,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1870.96
+3072,3072,3072,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1153.64
+3168,3168,3168,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1869.52
+3264,3264,3264,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1844.1
+3360,3360,3360,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1880.59
+3456,3456,3456,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1855.69
+3552,3552,3552,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1847.63
+3648,3648,3648,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1883.89
+3744,3744,3744,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1874.28
+3840,3840,3840,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1824.94
+3936,3936,3936,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1863.14
+4032,4032,4032,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1861.9
+4128,4128,4128,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1860.46
+4224,4224,4224,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1878.57
+4320,4320,4320,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1867.23
+4416,4416,4416,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1869.6
+4512,4512,4512,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1874.17
+4608,4608,4608,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,622.126
+4704,4704,4704,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1885.94
+4800,4800,4800,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1893.31
+4896,4896,4896,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1877.67
+4992,4992,4992,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1879.83
+5088,5088,5088,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1894.9
+5184,5184,5184,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1883.3
+5280,5280,5280,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1892.44
+5376,5376,5376,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1057.62
+5472,5472,5472,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1890.31
+5568,5568,5568,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1887.85
+5664,5664,5664,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1882.64
+5760,5760,5760,0,0,0,0,0,0,1.0,1.0,column,none,transpose,left,upper,unit,dgemm,gpu,clblas,S9150,1887.03

doc/performance/clBLAS_2.6.0/S9150/dgemm_96.csv

@@ -0,0 +1,31 @@
+m,n,k,lda,ldb,ldc,offa,offb,offc,alpha,beta,order,transa,transb,side,uplo,diag,function,device,library,label,GFLOPS
+192,192,192,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,8.9202
+384,384,384,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,46.185
+576,576,576,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,126.686
+768,768,768,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,235.366
+960,960,960,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,375.406
+1152,1152,1152,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,475.497
+1344,1344,1344,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,599.527
+1536,1536,1536,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,437.835
+1728,1728,1728,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,778.815
+1920,1920,1920,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,845.844
+2112,2112,2112,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,969.624
+2304,2304,2304,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,943.48
+2496,2496,2496,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1026.58
+2688,2688,2688,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1074.56
+2880,2880,2880,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1102.6
+3072,3072,3072,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,848.076
+3264,3264,3264,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1010.06
+3456,3456,3456,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1034.51
+3648,3648,3648,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1059.02
+3840,3840,3840,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1037.95
+4032,4032,4032,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1103.8
+4224,4224,4224,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1109.83
+4416,4416,4416,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1096.15
+4608,4608,4608,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1055.28
+4800,4800,4800,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1140.07
+4992,4992,4992,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1152.31
+5184,5184,5184,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1165.47
+5376,5376,5376,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1152.36
+5568,5568,5568,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1193.66
+5760,5760,5760,0,0,0,0,0,0,1.0,1.0,column,none,none,right,upper,unit,dtrsm,gpu,clblas,s9150_dtrsm_14502,1199.05

doc/performance/clBLAS_2.6.0/S9150/dtrsm_192.csv

@@ -0,0 +1,25 @@
+
+# sgemm AMD vs NVIDIA
+python ../../../../src/scripts/perf/plotPerformance.py \
+  -d peak_sp.csv \
+  -d ../../cuBLAS_7.0/Tesla_K40/peak_sp.csv \
+  -d sgemm_96.csv \
+  -d ../../cuBLAS_7.0/Tesla_K40/sgemm.csv \
+  -x sizem --x_axis_label "m,n,k" \
+  -y gflops --y_axis_label "GFlop/s" \
+  --x_axis_scale linear \
+  --plot label \
+  --title "sgemm S9150 vs K40" --outputfile sgemm_S9150_K40.png
+
+# dgemm AMD vs NVIDIA
+python ../../../../src/scripts/perf/plotPerformance.py \
+  -d peak_dp.csv \
+  -d ../../cuBLAS_7.0/Tesla_K40/peak_dp.csv \
+  -d dgemm_96.csv \
+  -d ../../cuBLAS_7.0/Tesla_K40/dgemm.csv \
+  -x sizem --x_axis_label "m,n,k" \
+  -y gflops --y_axis_label "GFlop/s" \
+  --x_axis_scale linear \
+  --plot label \
+  --title "dgemm S9150 vs K40" --outputfile dgemm_S9150_K40.png
+