B200 FP8-e4m3 mantissa precision

[lisuying214]

thank you for the support for B200 FP8gemm.
As we all known that in H100, e4m3 tensor core(QGMMA)only supports highest 14bits(include 1 sign bit ) mantissa.
I wonder if the accumulator bits is tested in B200 tensor core of e4m3 tested?

[StuartSul]

Hi @lisuying214 , could you clarify your question? It seems like you are referring to the WGMMA instruction which is not available on B200s