News for hackers. Stuff that matters.
Re:Which GPUs to get for deep learning by sytelus on Monday September 07, 2020 @10:21PM (#24404692)
> fp16 is only 30-40% faster than fp32

It really depends on workload. For ImageNet and resnet type architecture its not unusual to get 3X speed up. It also depends on if you do full fp16, leave BNs alone etc. This is a lot because instead of training for 6 days, you now training for 2 days.

Re:Which GPUs to get for deep learning by solidasparagus on Tuesday September 08, 2020 @12:15AM (#24405143)
> ImageNet and resnet type architecture its not unusual to get 3X speed up

Source on this? I've done a good bit of CV benchmarking work and I don't recall anything like a 3x boost. 30-40% improvement is much more in line with what I remember.