| 1 |
Platform: NVIDIA CUDA |
| 2 |
Device: Tesla P40 |
| 3 |
Driver version : 550.54.14 (Linux x64) |
| 4 |
Compute units : 30 |
| 5 |
Clock frequency : 1531 MHz |
| 6 |
|
| 7 |
Global memory bandwidth (GBPS) |
| 8 |
float : 282.85 |
| 9 |
float2 : 294.10 |
| 10 |
float4 : 301.39 |
| 11 |
float8 : 279.29 |
| 12 |
float16 : 193.72 |
| 13 |
|
| 14 |
Single-precision compute (GFLOPS) |
| 15 |
float : 11153.70 |
| 16 |
float2 : 11505.40 |
| 17 |
float4 : 11475.82 |
| 18 |
float8 : 11410.92 |
| 19 |
float16 : 11367.69 |
| 20 |
|
| 21 |
No half precision support! Skipped |
| 22 |
|
| 23 |
Double-precision compute (GFLOPS) |
| 24 |
double : 367.62 |
| 25 |
double2 : 367.05 |
| 26 |
double4 : 366.32 |
| 27 |
double8 : 365.52 |
| 28 |
double16 : 362.97 |
| 29 |
|
| 30 |
Integer compute (GIOPS) |
| 31 |
int : 3897.08 |
| 32 |
int2 : 3889.65 |
| 33 |
int4 : 3904.29 |
| 34 |
int8 : 3610.75 |
| 35 |
int16 : 3540.68 |
| 36 |
|
| 37 |
Integer compute Fast 24bit (GIOPS) |
| 38 |
int : 3895.72 |
| 39 |
int2 : 3901.65 |
| 40 |
int4 : 3895.32 |
| 41 |
int8 : 3882.49 |
| 42 |
int16 : 3866.57 |
| 43 |
|
| 44 |
Integer char (8bit) compute (GIOPS) |
| 45 |
char : 10813.47 |
| 46 |
char2 : 11447.82 |
| 47 |
char4 : 11485.37 |
| 48 |
char8 : 11522.07 |
| 49 |
char16 : 11404.32 |
| 50 |
|
| 51 |
Integer short (16bit) compute (GIOPS) |
| 52 |
short : 10708.50 |
| 53 |
short2 : 11449.04 |
| 54 |
short4 : 11481.69 |
| 55 |
short8 : 11518.50 |
| 56 |
short16 : 11333.30 |
| 57 |
|
| 58 |
Transfer bandwidth (GBPS) |
| 59 |
enqueueWriteBuffer : 6.17 |
| 60 |
enqueueReadBuffer : 6.45 |
| 61 |
enqueueWriteBuffer non-blocking : 5.68 |
| 62 |
enqueueReadBuffer non-blocking : 6.37 |
| 63 |
enqueueMapBuffer(for read) : 5.75 |
| 64 |
memcpy from mapped ptr : 9.36 |
| 65 |
enqueueUnmap(after write) : 6.27 |
| 66 |
memcpy to mapped ptr : 9.36 |
| 67 |
|
| 68 |
Kernel launch latency : 3.78 us |