I haven't not notice much of improvement for AVX512 wheel compare to AVX2, so it's kind of pointless.
In theory, VNNI should have improve performance quite a bit for inference.
https://community.intel.com/t5/Blogs/Tech-Innovation/Artificial-Intelligence-AI/Deep-Learning-Performance-Boost-by-Intel-VNNI/post/1335670
Kind of no brainer to use AVX512_VNNI.
Thanks.
I haven't not notice much of improvement for AVX512 wheel compare to AVX2, so it's kind of pointless.
In theory, VNNI should have improve performance quite a bit for inference.
https://community.intel.com/t5/Blogs/Tech-Innovation/Artificial-Intelligence-AI/Deep-Learning-Performance-Boost-by-Intel-VNNI/post/1335670
Kind of no brainer to use AVX512_VNNI.
Thanks.