| 1 |
|
| 2 |
Platform: NVIDIA CUDA |
| 3 |
Device: Tesla K80 |
| 4 |
Driver version : 455.32.00 (Linux x64) |
| 5 |
Compute units : 13 |
| 6 |
Clock frequency : 823 MHz |
| 7 |
|
| 8 |
Global memory bandwidth (GBPS) |
| 9 |
float : 147.19 |
| 10 |
float2 : 148.99 |
| 11 |
float4 : 152.33 |
| 12 |
float8 : 141.67 |
| 13 |
float16 : 68.77 |
| 14 |
|
| 15 |
Single-precision compute (GFLOPS) |
| 16 |
float : 2835.78 |
| 17 |
float2 : 2834.16 |
| 18 |
float4 : 3700.81 |
| 19 |
float8 : 3518.41 |
| 20 |
float16 : 3288.67 |
| 21 |
|
| 22 |
No half precision support! Skipped |
| 23 |
|
| 24 |
Double-precision compute (GFLOPS) |
| 25 |
double : 1400.02 |
| 26 |
double2 : 1399.04 |
| 27 |
double4 : 1394.24 |
| 28 |
double8 : 1396.52 |
| 29 |
double16 : 1386.00 |
| 30 |
|
| 31 |
Integer compute (GIOPS) |
| 32 |
int : 711.60 |
| 33 |
int2 : 711.39 |
| 34 |
int4 : 711.65 |
| 35 |
int8 : 711.87 |
| 36 |
int16 : 711.75 |
| 37 |
|
| 38 |
Integer compute Fast 24bit (GIOPS) |
| 39 |
int : 711.52 |
| 40 |
int2 : 711.37 |
| 41 |
int4 : 711.58 |
| 42 |
int8 : 711.36 |
| 43 |
int16 : 709.81 |
| 44 |
|
| 45 |
Transfer bandwidth (GBPS) |
| 46 |
enqueueWriteBuffer : 9.35 |
| 47 |
enqueueReadBuffer : 11.89 |
| 48 |
enqueueWriteBuffer non-blocking : 8.42 |
| 49 |
enqueueReadBuffer non-blocking : 11.13 |
| 50 |
enqueueMapBuffer(for read) : 9.99 |
| 51 |
memcpy from mapped ptr : 9.80 |
| 52 |
enqueueUnmap(after write) : 12.05 |
| 53 |
memcpy to mapped ptr : 9.57 |
| 54 |
|
| 55 |
Kernel launch latency : 8.20 us |
| 56 |
|