Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

FP16 tensorcore = 312 tflops

x 2 (counting sparse as dense) = 624 tflops

x 8 GPUs = 5 "pflops"

The missing 8x you are looking for is just because tensorcore math is much faster than their normal fma path.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: