Tensorflow not working with AMD EPYC 7302 (3,00 GHz, 16-Core, 128 MB)

Dear Members,

I have a Gigabyte GeForce RTX 3090 Turbo (24GB, 10496 Cores), (2x HDMI, 2x DP) installed on a 4 GPU slot server having a AMD EPYC 7302 (3,00 GHz, 16-Core, 128 MB) cpu.

Currently I am trying to run a python3 keras script (using tensorflow) but even when running a minimum script like

import tensorflow as tf
print("Num GPUs Available: ", len(tf.config.list_physical_devices(‘GPU’)))

this leads to the following Error:

The TensorFlow library was compiled to use SSE4.1 instructions, but these aren’t available on your machine. Aborted

It seems the AMD CPU has no support for SSE4.1 and tensorflow needs to be run on a recent Intel CPU.

Noone told us that before and now the hardware is set up and we cant change it anymore.

Because of time pressure I need to find a most simple workaround to make our scripts run flawlessly.

Further details:

  • The host uses PCI passthrough to make the GPU available from within a Debian 11 VM (Works, CUDA 11.5 installed and working)
  • Python3 with venv used.

requirements.txt contains:

The examples are working on a Win10 machine using a Intel I7 CPU and Gforce GTX1080 mobile (so the code is ok).

I really need help on this!

Yours faithfully


It seems that the new Version of the Proxmox Hypervisor used to host the vm already has an option to change the emulation type of the cpu (default kvm) to EPYC.

A quick smoke up test shows that the error message is gone, but now cudnn Files are missing which is not a related problem.
(Hint given by https://www.thomas-krenn.com support team)

I will close this issue as soon as i can proof that the problem is solved.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.