| 1 |
Platform: NVIDIA CUDA |
| 2 |
Device: NVIDIA A100-SXM4-40GB |
| 3 |
Driver version : 515.48.07 (Linux x64) |
| 4 |
Compute units : 108 |
| 5 |
Clock frequency : 1410 MHz |
| 6 |
|
| 7 |
Global memory bandwidth (GBPS) |
| 8 |
float : 1292.96 |
| 9 |
float2 : 1377.07 |
| 10 |
float4 : 1419.22 |
| 11 |
float8 : 1443.80 |
| 12 |
float16 : 1464.44 |
| 13 |
|
| 14 |
Single-precision compute (GFLOPS) |
| 15 |
float : 19352.89 |
| 16 |
float2 : 19386.64 |
| 17 |
float4 : 19367.97 |
| 18 |
float8 : 19285.78 |
| 19 |
float16 : 19110.61 |
| 20 |
|
| 21 |
No half precision support! Skipped |
| 22 |
|
| 23 |
Double-precision compute (GFLOPS) |
| 24 |
double : 9723.09 |
| 25 |
double2 : 9707.90 |
| 26 |
double4 : 9686.22 |
| 27 |
double8 : 9636.75 |
| 28 |
double16 : 9547.99 |
| 29 |
|
| 30 |
Integer compute (GIOPS) |
| 31 |
int : 19278.83 |
| 32 |
int2 : 19335.19 |
| 33 |
int4 : 19273.49 |
| 34 |
int8 : 19357.90 |
| 35 |
int16 : 19346.59 |
| 36 |
|
| 37 |
Integer compute Fast 24bit (GIOPS) |
| 38 |
int : 19289.10 |
| 39 |
int2 : 19293.80 |
| 40 |
int4 : 19278.51 |
| 41 |
int8 : 19233.96 |
| 42 |
int16 : 19053.93 |
| 43 |
|
| 44 |
Transfer bandwidth (GBPS) |
| 45 |
enqueueWriteBuffer : 14.54 |
| 46 |
enqueueReadBuffer : 13.04 |
| 47 |
enqueueWriteBuffer non-blocking : 7.79 |
| 48 |
enqueueReadBuffer non-blocking : 7.21 |
| 49 |
enqueueMapBuffer(for read) : 20.03 |
| 50 |
memcpy from mapped ptr : 20.65 |
| 51 |
enqueueUnmap(after write) : 26.77 |
| 52 |
memcpy to mapped ptr : 20.86 |
| 53 |
|
| 54 |
Kernel launch latency : 5.76 us |
| 55 |
|
| 56 |
Device: NVIDIA A100-SXM4-40GB |
| 57 |
Driver version : 515.48.07 (Linux x64) |
| 58 |
Compute units : 108 |
| 59 |
Clock frequency : 1410 MHz |
| 60 |
|
| 61 |
Global memory bandwidth (GBPS) |
| 62 |
float : 1293.90 |
| 63 |
float2 : 1377.21 |
| 64 |
float4 : 1419.86 |
| 65 |
float8 : 1443.43 |
| 66 |
float16 : 1464.50 |
| 67 |
|
| 68 |
Single-precision compute (GFLOPS) |
| 69 |
float : 19349.87 |
| 70 |
float2 : 19381.94 |
| 71 |
float4 : 19362.20 |
| 72 |
float8 : 19280.28 |
| 73 |
float16 : 19105.41 |
| 74 |
|
| 75 |
No half precision support! Skipped |
| 76 |
|
| 77 |
Double-precision compute (GFLOPS) |
| 78 |
double : 9716.58 |
| 79 |
double2 : 9701.24 |
| 80 |
double4 : 9680.46 |
| 81 |
double8 : 9631.46 |
| 82 |
double16 : 9543.48 |
| 83 |
|
| 84 |
Integer compute (GIOPS) |
| 85 |
int : 19275.95 |
| 86 |
int2 : 19324.45 |
| 87 |
int4 : 19268.48 |
| 88 |
int8 : 19351.87 |
| 89 |
int16 : 19343.15 |
| 90 |
|
| 91 |
Integer compute Fast 24bit (GIOPS) |
| 92 |
int : 19283.22 |
| 93 |
int2 : 19287.28 |
| 94 |
int4 : 19272.75 |
| 95 |
int8 : 19230.13 |
| 96 |
int16 : 19047.88 |
| 97 |
|
| 98 |
Transfer bandwidth (GBPS) |
| 99 |
enqueueWriteBuffer : 14.50 |
| 100 |
enqueueReadBuffer : 13.05 |
| 101 |
enqueueWriteBuffer non-blocking : 7.71 |
| 102 |
enqueueReadBuffer non-blocking : 7.27 |
| 103 |
enqueueMapBuffer(for read) : 19.83 |
| 104 |
memcpy from mapped ptr : 19.54 |
| 105 |
enqueueUnmap(after write) : 26.77 |
| 106 |
memcpy to mapped ptr : 20.55 |
| 107 |
|
| 108 |
Kernel launch latency : 5.65 us |
| 109 |
|
| 110 |
Device: NVIDIA A100-SXM4-40GB |
| 111 |
Driver version : 515.48.07 (Linux x64) |
| 112 |
Compute units : 108 |
| 113 |
Clock frequency : 1410 MHz |
| 114 |
|
| 115 |
Global memory bandwidth (GBPS) |
| 116 |
float : 1304.11 |
| 117 |
float2 : 1376.87 |
| 118 |
float4 : 1419.82 |
| 119 |
float8 : 1444.07 |
| 120 |
float16 : 1465.06 |
| 121 |
|
| 122 |
Single-precision compute (GFLOPS) |
| 123 |
float : 19350.41 |
| 124 |
float2 : 19382.11 |
| 125 |
float4 : 19363.12 |
| 126 |
float8 : 19281.61 |
| 127 |
float16 : 19108.25 |
| 128 |
|
| 129 |
No half precision support! Skipped |
| 130 |
|
| 131 |
Double-precision compute (GFLOPS) |
| 132 |
double : 9719.24 |
| 133 |
double2 : 9704.38 |
| 134 |
double4 : 9682.93 |
| 135 |
double8 : 9633.92 |
| 136 |
double16 : 9544.74 |
| 137 |
|
| 138 |
Integer compute (GIOPS) |
| 139 |
int : 19277.98 |
| 140 |
int2 : 19332.19 |
| 141 |
int4 : 19269.01 |
| 142 |
int8 : 19352.73 |
| 143 |
int16 : 19343.15 |
| 144 |
|
| 145 |
Integer compute Fast 24bit (GIOPS) |
| 146 |
int : 19283.32 |
| 147 |
int2 : 19288.03 |
| 148 |
int4 : 19273.28 |
| 149 |
int8 : 19231.30 |
| 150 |
int16 : 19048.40 |
| 151 |
|
| 152 |
Transfer bandwidth (GBPS) |
| 153 |
enqueueWriteBuffer : 14.37 |
| 154 |
enqueueReadBuffer : 13.13 |
| 155 |
enqueueWriteBuffer non-blocking : 7.50 |
| 156 |
enqueueReadBuffer non-blocking : 6.90 |
| 157 |
enqueueMapBuffer(for read) : 19.81 |
| 158 |
memcpy from mapped ptr : 20.73 |
| 159 |
enqueueUnmap(after write) : 26.77 |
| 160 |
memcpy to mapped ptr : 20.62 |
| 161 |
|
| 162 |
Kernel launch latency : 5.75 us |
| 163 |
|
| 164 |
Device: NVIDIA A100-SXM4-40GB |
| 165 |
Driver version : 515.48.07 (Linux x64) |
| 166 |
Compute units : 108 |
| 167 |
Clock frequency : 1410 MHz |
| 168 |
|
| 169 |
Global memory bandwidth (GBPS) |
| 170 |
float : 1303.89 |
| 171 |
float2 : 1376.82 |
| 172 |
float4 : 1419.15 |
| 173 |
float8 : 1444.89 |
| 174 |
float16 : 1465.04 |
| 175 |
|
| 176 |
Single-precision compute (GFLOPS) |
| 177 |
float : 19339.44 |
| 178 |
float2 : 19388.10 |
| 179 |
float4 : 19371.42 |
| 180 |
float8 : 19289.58 |
| 181 |
float16 : 19115.54 |
| 182 |
|
| 183 |
No half precision support! Skipped |
| 184 |
|
| 185 |
Double-precision compute (GFLOPS) |
| 186 |
double : 9724.83 |
| 187 |
double2 : 9710.39 |
| 188 |
double4 : 9689.79 |
| 189 |
double8 : 9641.13 |
| 190 |
double16 : 9552.76 |
| 191 |
|
| 192 |
Integer compute (GIOPS) |
| 193 |
int : 19285.03 |
| 194 |
int2 : 19313.19 |
| 195 |
int4 : 19286.42 |
| 196 |
int8 : 19361.56 |
| 197 |
int16 : 19347.78 |
| 198 |
|
| 199 |
Integer compute Fast 24bit (GIOPS) |
| 200 |
int : 19292.73 |
| 201 |
int2 : 19297.12 |
| 202 |
int4 : 19282.58 |
| 203 |
int8 : 19238.22 |
| 204 |
int16 : 19056.33 |
| 205 |
|
| 206 |
Transfer bandwidth (GBPS) |
| 207 |
enqueueWriteBuffer : 14.48 |
| 208 |
enqueueReadBuffer : 13.16 |
| 209 |
enqueueWriteBuffer non-blocking : 7.18 |
| 210 |
enqueueReadBuffer non-blocking : 6.98 |
| 211 |
enqueueMapBuffer(for read) : 19.99 |
| 212 |
memcpy from mapped ptr : 19.35 |
| 213 |
enqueueUnmap(after write) : 26.77 |
| 214 |
memcpy to mapped ptr : 20.63 |
| 215 |
|
| 216 |
Kernel launch latency : 5.70 us |