AMD Radeon RX 6700 XT GPU not supported with TensorFlow ROCm

[abutair]

Issue type

Bug

Have you reproduced the bug with TensorFlow Nightly?

Yes

Source

source

TensorFlow version

2.11.0

Custom code

Yes

OS platform and distribution

Linux Ubuntu 22.04.4 LTS

Mobile device

No response

Python version

Python 3.9

Bazel version

No response

GCC/compiler version

No response

CUDA/cuDNN version

Not applicable (as ROCm is being used)

GPU model and memory

AMD Radeon RX 6700 XT, 12GB VRAM

Current behavior?

TensorFlow is not recognizing the AMD Radeon RX 6700 XT GPU due to unsupported amdgpu version gfx1031, resulting in the GPU being ignored during computations.

Standalone code to reproduce the issue

import tensorflow as tf
print("TensorFlow version:", tf.__version__)
print("Is built with ROCm:", tf.test.is_built_with_rocm())
print("GPUs available:", tf.config.list_physical_devices('GPU'))

Relevant log output

TensorFlow version: 2.11.0
Is built with ROCm: True
GPUs available: []

[abutair]

After checking other supported versions, I found that the support list includes gfx900, gfx906, gfx908, gfx90a, gfx940, gfx941, gfx942, gfx1030, and gfx1100. However, my GPU has gfx1031, and it seems there is no official version that supports it. Is this a bug, or am I mistaken about the support status for my GPU?

[Redhawk18]

It's sad to see how inactive this project is for a billion dollar company but this is what I found. I just had the same issue.

It's very odd it doesn't work for you I have a 5700xt which is unsupported but works fine, but I thought RDNA2+ had full official rocm support.

[harkgill-amd]

Hi @abutair, the RX 6700 XT is not officially supported for usage with ROCm(see Supported GPU List). As @Redhawk18 mentioned, you can utilize the HSA_OVERRIDE_GFX_VERSION=10.3.0 variable to mimic a supported architecture. This can be set as an environment variable for all runs with

export HSA_OVERRIDE_GFX_VERSION=10.3.0

Please let me know if you have any questions or run into further issues. Thanks!