2 ms·
There's also the 512-bit matrix units which would be more like 3840 FLOP/cycle (single precision) for workloads that can be expressed that way
by gok 6y ago
There's also the 512-bit matrix units which would be more like 3840 FLOP/cycle (single precision) for workloads that can be expressed that way
- dragontamer 6y agoI'm not sure if that plays to the advantage of POWER10. NVidia 2070 Super also has FP16 matrix-multiplication units, achieving 57 FP16-matrix TFLOPs. These are the NVidia "Tensor Cores". Ampere (rumored to be released within a few weeks...) even has sparse matrix-multiplication (!!) units being added in.