| 1 |
Platform: Portable Computing Language |
| 2 |
Device: NVIDIA Tegra X1 |
| 3 |
Driver version : 1.3 (Linux ARM64) |
| 4 |
Compute units : 1 |
| 5 |
Clock frequency : 921 MHz |
| 6 |
|
| 7 |
Global memory bandwidth (GBPS) |
| 8 |
float : 17.95 |
| 9 |
float2 : 20.21 |
| 10 |
float4 : 20.92 |
| 11 |
float8 : 19.82 |
| 12 |
float16 : 15.14 |
| 13 |
|
| 14 |
Single-precision compute (GFLOPS) |
| 15 |
float : 214.09 |
| 16 |
float2 : 229.80 |
| 17 |
float4 : 230.95 |
| 18 |
float8 : 229.31 |
| 19 |
float16 : 228.80 |
| 20 |
|
| 21 |
Half-precision compute (GFLOPS) |
| 22 |
half : 212.93 |
| 23 |
half2 : 228.95 |
| 24 |
half4 : 228.69 |
| 25 |
half8 : 245.39 |
| 26 |
half16 : 238.39 |
| 27 |
|
| 28 |
Double-precision compute (GFLOPS) |
| 29 |
double : 7.32 |
| 30 |
double2 : 7.31 |
| 31 |
double4 : 7.30 |
| 32 |
double8 : 7.27 |
| 33 |
double16 : 7.21 |
| 34 |
|
| 35 |
Integer compute (GIOPS) |
| 36 |
int : 70.95 |
| 37 |
int2 : 74.95 |
| 38 |
int4 : 76.43 |
| 39 |
int8 : 76.62 |
| 40 |
int16 : 76.78 |
| 41 |
|
| 42 |
Transfer bandwidth (GBPS) |
| 43 |
enqueueWriteBuffer : 2.94 |
| 44 |
enqueueReadBuffer : 0.69 |
| 45 |
enqueueMapBuffer(for read) : 2487.73 |
| 46 |
memcpy from mapped ptr : 0.70 |
| 47 |
enqueueUnmap(after write) : 0.68 |
| 48 |
memcpy to mapped ptr : 3.68 |
| 49 |
|
| 50 |
Kernel launch latency : 32.77 us |
| 51 |
|
| 52 |
Note via POCL 1.3 |