| 1 |
Platform: NVIDIA CUDA |
| 2 |
Device: Tesla T4 |
| 3 |
Driver version : 560.35.03 (Linux x64) |
| 4 |
Compute units : 40 |
| 5 |
Clock frequency : 1590 MHz |
| 6 |
|
| 7 |
Global memory bandwidth (GBPS) |
| 8 |
float : 235.00 |
| 9 |
float2 : 247.01 |
| 10 |
float4 : 253.11 |
| 11 |
float8 : 263.44 |
| 12 |
float16 : 252.38 |
| 13 |
|
| 14 |
Single-precision compute (GFLOPS) |
| 15 |
float : 8030.45 |
| 16 |
float2 : 8034.32 |
| 17 |
float4 : 7985.38 |
| 18 |
float8 : 7848.48 |
| 19 |
float16 : 7651.69 |
| 20 |
|
| 21 |
No half precision support! Skipped |
| 22 |
|
| 23 |
Double-precision compute (GFLOPS) |
| 24 |
double : 256.45 |
| 25 |
double2 : 256.03 |
| 26 |
double4 : 253.74 |
| 27 |
double8 : 252.76 |
| 28 |
double16 : 251.68 |
| 29 |
|
| 30 |
Integer compute (GIOPS) |
| 31 |
int : 5802.79 |
| 32 |
int2 : 5715.24 |
| 33 |
int4 : 5742.30 |
| 34 |
int8 : 5863.19 |
| 35 |
int16 : 5711.99 |
| 36 |
|
| 37 |
Transfer bandwidth (GBPS) |
| 38 |
enqueueWriteBuffer : 4.73 |
| 39 |
enqueueReadBuffer : 4.78 |
| 40 |
enqueueMapBuffer(for read) : 8.73 |
| 41 |
memcpy from mapped ptr : 5.39 |
| 42 |
enqueueUnmap(after write) : 12.17 |
| 43 |
memcpy to mapped ptr : 5.39 |
| 44 |
|
| 45 |
Kernel launch latency : 5.82 us |