[W NNPACK.cpp:80] Could not initialize NNPACK! Reason: Unsupported hardware

Experiencing this message when running on multiple Intel based CPUs.

On one system I compiled torch from sources (Python 3.9) following “getting started” instructions. On the other I installed via pip (Python 3.8). So this seems to be an issue with multiple CPU architectures and multiple versions.

  1. What is the performance implication?
  2. Do I need to fix this issue? Is it safe to ignore?
  3. Is it possible to silence the warning?

I just installed pytorch via anaconda to test the speech commands tutorial.

I have got as far as

new_sample_rate = 8000
transform = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=new_sample_rate)
transformed = transform(waveform)

ipd.Audio(transformed.numpy(), rate=new_sample_rate)

Then [W NNPACK.cpp:80] Could not initialize NNPACK! Reason: Unsupported hardware.

Ubuntu 20.04, I5-3570, GTX1050ti

OK, so after digging into NNPACK - requirements state avx2 and l3 cache are required.

one of my systems does not support avx2(Intel® Xeon® CPU E5-2690 v2 @ 3.00GHz) and the other does not have L3(Intel® Xeon Phi™ CPU 7250 @ 1.40GHz)

There’s nothing to be done, wish that this warning could be silenced.

1 Like

Hey @bloring so you mean there is no way to avoid this error because of no hardware support?
(building from source could be a solution but it is crashing my pc)