Anyone know of a more recent comparison of the performance advantage of GPUs over CPUs for neural nets? Everybody seems to cite this 2004 paper that claims a 20x advantage... surely there's a more recent post-transformers study?
Conversation
Replying to
Not an AI expert, but I haven't heard *anyone* in the startup field talk about using CPUs, and there are a lot of AI/ML startups on my radar. AI got rescued from AI winter by GPUs, so.
1
Replying to
I'm not talking about actually using CPUs. I just want to know the theoretical performance advantage.
This Tweet was deleted by the Tweet author. Learn more
thanks that looks good
Replying to
If you are looking for pure numbers the MLPerf benchmark is probably the best bet as it has each side with all optimizations, and it is updated regularly. Downside is companies only submit when their results are competitive mlcommons.org/en/
1


