4 ms·yesby sabalaba 8y agoyeswmf 8y agoInteresting. I would have expected FP16 with tensor cores to give at least a 100% speedup, not just 60%.