AVX2 is a minimum requirement？

[Ecthellin203]

I got a 7900XTX with E5v2 CPUs. When Ollama runs models, it will cause an error“Error: llama runner process has terminated: exit status 0xc000001d”.
I want to confirm if this is due to the lack of AVX2?

[metal3d]

I'm working on an "old" computer (13 years old) with a RTX 3070 and 24Go RAM. I don't have AVX2 support on this CPU, only AVX. And it works.

Linux Fedora 40.

[Ecthellin203]

Obviously, Nvidia does much better than AMD in this regard. Now i believe this is a problem with ROCm, not llama.cpp.