GPU benchmarks for exllamav2

[volodymyr-barannik]

Hello.

There is a GPU benchmark ran on llama.cpp.

My question is whether we can safely translate these results to exllamav2? Or does exllamav2 have some features which make some GPUs perform better or worse than in that llama.cpp benchmark?

I am most interested in 2x3090/4090 vs A100.

Thank you!

P.S. if you benchmarked exllamav2 on several GPUs it would be really nice if you could share the results..