X86 3x performance difference NNVM vs Relay

I have read previous posts and i understand that relay does not use avx2. Is this resolved as i have compared performance with various different parameters and the results are always consistent.

i don;t think there will be any limitations in terms of avx2. As of now we always recommend relay over nnvm

Thanks for replying.

Build Configuration: TVM github build LLVM 6.0.1 Openblas Centos 8.2

target = tvm.target.create(“llvm -mcpu=core-avx2”)

This means i am missing something in the relay tuning script. Please suggest me the changes i can make in following script to improve performance.

It would be helpful if you can dig deeper, e.g. what is the model you observed perf and look into the generated code. It is hard to tell as it is what could be wrong