5

I am interested in running deep learning code on my Ryzen 3400g system. Online I saw:

I trying to use Vega 11 on my Ryzen 3400G, I'm aware that I cannot run HIP, but I possibly could run opencl on this APU, my first step is to import tensorflow, but I got an error,

This is from https://github.com/ROCmSoftwarePlatform/tensorflow-upstream/issues/669

Is it really impossible to use the ryzen 3400g with Radeon vega 11 gpu with tensorflow?

—————

Could it be made to work with opencl instead (as in https://missinglink.ai/guides/tensorflow/tensorflow-support-opencl/ )?

Simd
  • 19,447
  • 42
  • 136
  • 271

1 Answers1

5

According to the official HIP Github page,:

HIP is a C++ Runtime API and Kernel Language that allows developers to create portable applications for AMD and NVIDIA GPUs from single source code.

And yes you can use tensorflow library like a normal python module, but you cannot use tensorflow-gpu which leverages Cuda cores available only Nvidia GPUs. And the Radeon Open Compute project is basically aimed at 'HPC and Ultrascale computing' and you can see the official supported GPUs here. They have also mentioned that the ROCm project does not support built-in GPUs.

The integrated GPUs in AMD APUs are not officially supported targets for ROCm

So basically the answer is no, you won't be able to run tensorflow-gpu on your Ryzen 3600G system but you can use the normal tensorflow library without GPU acceleration. There are lots of threads discussing the issue so hopefully something will come up soon.

krxat
  • 513
  • 4
  • 16
  • Thank you although it’s a real shame if it’s really impossible. – Simd Aug 08 '20 at 06:36
  • Yeah the ROCm project is a very early one which is still in development as of my knowledge and AMD as a whole is very late to the "leveraging multiple cores for deep learning" game. – krxat Aug 08 '20 at 06:40
  • What is the current status? – Simd Aug 18 '21 at 13:53