| 1 |
|
| 2 |
Platform: NVIDIA CUDA |
| 3 |
Device: GeForce GTX 660 |
| 4 |
Driver version : 331.20 (Linux x86) |
| 5 |
Compute units : 5 |
| 6 |
|
| 7 |
Global memory bandwidth (GBPS) |
| 8 |
float : 107.96 |
| 9 |
float2 : 111.36 |
| 10 |
float4 : 113.08 |
| 11 |
float8 : 57.77 |
| 12 |
float16 : 37.33 |
| 13 |
|
| 14 |
Single-precision compute (GFLOPS) |
| 15 |
float : 1412.18 |
| 16 |
float2 : 1862.79 |
| 17 |
float4 : 1785.61 |
| 18 |
float8 : 1832.08 |
| 19 |
float16 : 1784.82 |
| 20 |
|
| 21 |
Double-precision compute (GFLOPS) |
| 22 |
double : 89.72 |
| 23 |
double2 : 89.60 |
| 24 |
double4 : 89.42 |
| 25 |
double8 : 89.10 |
| 26 |
double16 : 88.38 |
| 27 |
|
| 28 |
Integer compute (GIOPS) |
| 29 |
int : 358.32 |
| 30 |
int2 : 358.40 |
| 31 |
int4 : 358.11 |
| 32 |
int8 : 358.62 |
| 33 |
int16 : 358.41 |
| 34 |
|
| 35 |
Transfer bandwidth (GBPS) |
| 36 |
enqueueWriteBuffer : 6.53 |
| 37 |
enqueueReadBuffer : 6.58 |
| 38 |
enqueueMapBuffer(for read) : 2.07 |
| 39 |
memcpy from mapped ptr : 10.12 |
| 40 |
enqueueUnmap(after write) : 3.78 |
| 41 |
memcpy to mapped ptr : 10.29 |
| 42 |
|
| 43 |
Kernel launch latency : 6.89 us |
| 44 |
|