GPT3 is so big it would take 355 years to train on a nvidia V100, so your example is also not really useful for comparison. It would be interesting to see some

GPT3 is so big it would take 355 years to train on a nvidia V100, so your example is also not really useful for comparison. It would be interesting to see some mid-sized nn benchmarks though.