zlacker
[parent]
[thread]
1 comments
1. stonog+(OP)
[view]
[source]
2025-12-04 21:33:58
Am I reading this wrong, or does this only support FP16 inputs, and compares its performance against an FP32 solver?
replies(1):
>>Bulat_+kT4
◧
2. Bulat_+kT4
[view]
[source]
2025-12-06 10:31:48
>>stonog+(OP)
They compare HGEMM implementations. At least CUBLAS has HGEMM functions.
HGEMM means half-precision (i.e. FP16) general matrix multiplication
[go to top]