Accelerate with AMD GPUs

[eliranwong]

This pull request is created to extend piper support of AMD GPUs or ROCm-enabled GPUs via 'ROCMExecutionProvider' or 'MIGraphXExecutionProvider'.

Original issue is recorded at: #483

To support ROCm-enabled GPUs via 'ROCMExecutionProvider' or 'MIGraphXExecutionProvider':

1. Install piper-tts

pip install piper-tts

2. Uninstall onnxruntime

pip uninstall onnxruntime

3. Install onnxruntime-rocm

pip3 install https://repo.radeon.com/rocm/manylinux/rocm-rel-6.0.2/onnxruntime_rocm-inference-1.17.0-cp310-cp310-linux_x86_64.whl

Remarks: Wheel files that support different ROCm versions are available at: https://repo.radeon.com/rocm/manylinux

To verify:

python3

$ import onnxruntime
$ onnxruntime.get_available_providers()

Output:

['MIGraphXExecutionProvider', 'ROCMExecutionProvider', 'CPUExecutionProvider']

To accelerate with AMD GPUs:

piper --migraphx

To accelerate with ROCm-enabled GPUs:

piper --rocm

Remarks: Tested on Ubuntu 22.04.4 + Kernel 6.6.32 + ROCm 6.0.2

Setup notes are available at: https://github.com/eliranwong/MultiAMDGPU_AIDev_Ubuntu/tree/main

[GrumpyMeow]

This seems like a legit Pull Request. Having this Pull request merged, would be appreciated.

[kvrban]

Can you show benchmarks using this solution?

[remon-nashid]

crickets

[eliranwong]

It appears that the developer does not care much about AMD users.

[remon-nashid]

It appears that the developer does not care much about AMD users.

I'd understand the lack of interest in AMD if he's a sole maintainer who uses nvidia, but I don't get why he'd ignore such a well prepared PR to improve the capabilities of piper! I think I'll use your fork @eliranwong for now.

[Digitalpeer1]

Has anyone got training to work with AMD GPU yet?

[alukortti]

Has anyone got training to work with AMD GPU yet?

I got it working in AMD rocm docker environment but my old RTX2080S outperforms my new 7900xtx GPU. Trying to figure out how to get it working better with AMD GPU.