Does tensorflow support AMD Radeon graphics cards?

General Discussion
, ,
#1

Hi,
First post here. I’m looking to buy a new laptop and I’ve found one that has an AMD Radeon graphics card. I’ll be using Linux, if that plays any role.
I’d like to have GPU support with Tensorflow, so I’d like to ask if there is support for AMD Radeon cards, as I can’t find a lot of information online.

Thanks

#3

Hi @AlexandrosDrymonitis, For using tensorflow on GPU you need to install CUDA toolkit, but CUDA was developed by NVIDIA, and it is specifically designed to work with NVIDIA GPUs. so it is not possible to use tensorflow on AMD GPUs. Thank You.

#5

Thanks for your reply. Is there an official guide on which GPUs are supported by Tensorflow? I might be able to find a laptop with an NVIDIA GPU, but I want to make sure I can use it with Tensorflow.

#6

Hi @AlexandrosDrymonitis, Please refer to document to know the cuda enabled GPU devices that are supported tensorflow. Thank You!

#8

Thanks for this. I have read that this guide is not up-to-date. Is that true? I also read in the link you provided the following:

For GPUs with unsupported CUDA® architectures , or to avoid JIT compilation from PTX, or to use different versions of the NVIDIA® libraries, see the Linux build from source guide.

Can someone verify that it is possible to get GPU support with an NVIDIA graphics card* not included here, if TensorFlow is built from source?

*I can also find laptops with NVIDIA cards, but even those are not included in the GPUs list of CUDA.

#10

Actually, there is a fork maintained by AMD themselves to use ROCM.

More information here: TensorFlow Installation for ROCm — ROCm 5.6.0 Documentation Home

1 Like
#12

Nice to hear ! What about performances ? Does it work with AMD APU like 680m or 780m?