| 1 |
Platform: NVIDIA CUDA |
| 2 |
Device: Tesla K40c |
| 3 |
Driver version : 355.11 (Linux x64) |
| 4 |
Compute units : 15 |
| 5 |
Clock frequency : 745 MHz |
| 6 |
|
| 7 |
Global memory bandwidth (GBPS) |
| 8 |
float : 207.57 |
| 9 |
float2 : 213.87 |
| 10 |
float4 : 218.45 |
| 11 |
float8 : 178.41 |
| 12 |
float16 : 127.86 |
| 13 |
|
| 14 |
Single-precision compute (GFLOPS) |
| 15 |
float : 2798.13 |
| 16 |
float2 : 3495.13 |
| 17 |
float4 : 3340.98 |
| 18 |
float8 : 3481.63 |
| 19 |
float16 : 3168.49 |
| 20 |
|
| 21 |
Double-precision compute (GFLOPS) |
| 22 |
double : 1427.69 |
| 23 |
double2 : 1426.97 |
| 24 |
double4 : 1423.42 |
| 25 |
double8 : 1418.63 |
| 26 |
double16 : 1408.26 |
| 27 |
|
| 28 |
Integer compute (GIOPS) |
| 29 |
int : 715.75 |
| 30 |
int2 : 715.52 |
| 31 |
int4 : 715.50 |
| 32 |
int8 : 715.39 |
| 33 |
int16 : 715.43 |
| 34 |
|
| 35 |
Transfer bandwidth (GBPS) |
| 36 |
enqueueWriteBuffer : 5.93 |
| 37 |
enqueueReadBuffer : 5.86 |
| 38 |
enqueueMapBuffer(for read) : 5.85 |
| 39 |
memcpy from mapped ptr : 4.80 |
| 40 |
enqueueUnmap(after write) : 6.16 |
| 41 |
memcpy to mapped ptr : 4.84 |
| 42 |
|
| 43 |
Kernel launch latency : 9.27 us |