M4 Max

fp32-scalar = 15407 GFLOPS
fp32-vec4 = 14517 GFLOPS
fp16-scalar = 15678 GFLOPS
fp16-vec4 = 16035 GFLOPS
fp16-matrix = 0 GFLOPS
fp64-scalar = 0 GFLOPS
fp64-vec4 = 0 GFLOPS
int32-scalar = 3962 GIOPS
int32-vec4 = 3920 GIOPS
int16-scalar = 3978 GIOPS
int16-vec4 = 3955 GIOPS

希望我7 80歲的時候也能像川普這樣發瘋

Iris Xe Graphics

fp32-scalar = 1585 GFLOPS
fp32-vec4 = 797 GFLOPS
fp16-scalar = 3143 GFLOPS
fp16-vec4 = 1592 GFLOPS
fp16-matrix = 0 GFLOPS
fp64-scalar = 0 GFLOPS
fp64-vec4 = 0 GFLOPS
int32-scalar = 527 GIOPS
int32-vec4 = 266 GIOPS
int16-scalar = 1580 GIOPS
int16-vec4 = 1101 GIOPS

UHD 770

fp32-scalar = 774 GFLOPS
fp32-vec4 = 393 GFLOPS
fp16-scalar = 1507 GFLOPS
fp16-vec4 = 786 GFLOPS
fp16-matrix = 0 GFLOPS
fp64-scalar = 0 GFLOPS
fp64-vec4 = 0 GFLOPS
int32-scalar = 261 GIOPS
int32-vec4 = 131 GIOPS
int16-scalar = 760 GIOPS
int16-vec4 = 540 GIOPS

以first principle來思考，HLS之於FPGA會不會就是之前的CUDA之於GPU呢?

Tesla P40

fp32-scalar = 11359 GFLOPS
fp32-vec4 = 11653 GFLOPS
fp16-scalar = 0 GFLOPS
fp16-vec4 = 0 GFLOPS
fp16-matrix = 0 GFLOPS
fp64-scalar = 368 GFLOPS
fp64-vec4 = 368 GFLOPS
int32-scalar = 3919 GIOPS
int32-vec4 = 3905 GIOPS
int16-scalar = 0 GIOPS
int16-vec4 = 0 GIOPS

第一次煮水餃

4060M

fp32-scalar = 7771 GFLOPS
fp32-vec4 = 10174 GFLOPS
fp16-scalar = 7634 GFLOPS
fp16-vec4 = 8839 GFLOPS
fp16-matrix = 54578 GFLOPS
fp64-scalar = 240 GFLOPS
fp64-vec4 = 240 GFLOPS
int32-scalar = 7288 GIOPS
int32-vec4 = 7186 GIOPS
int16-scalar = 4597 GIOPS
int16-vec4 = 5085 GIOPS

受傷了被寶貝關心 我真的好幸福喔