AMD ROCm 5.5 Now Available on GitHub

[GFreeman]

As expected with AMD's activity on GitHub, ROCm 5.5 has now been officially released. It brings several big changes, including better RDNA 3 support. While officially focused on AMD's professional/workstation graphics cards, the ROCm 5.5 should also bring better support for Radeon RX 7000 series graphics cards on Linux.

Surprisingly, the release notes do not officially mention RDNA 3 improvements in its release notes, but those have been already tested and confirmed. The GPU support list is pretty short including AMD GFX9, RDNA, and CDNA GPUs, ranging from Radeon VII, Pro VII, W6800, V620, and Instinct lineup. The release notes do mention new HIP enhancements, enhanced stack size limit, raising it from 16k to 128k, new APIs, OpenMP enhancements, and more. You can check out the full release notes, downloads, and more details over at GitHub.



View at TechPowerUp Main Site | Source

 

[sam_86314]

where windows support

DirectML sucks in terms of performance I've found out. On my 6800 XT, a 512x512 generation on Stable Diffusion takes about 30 seconds on DirectML, and 5 seconds on ROCm.

 

[mx500torid]

Does it do anything for folding on a 6800xt?

 

[Leiesoldat]

AMD is targeting large compute systems with ROCm which is primarily UNIX based systems. Windows will come but is a lower priority than where the lion's share of profits are.

 

[vietanh2901]

where windows support

DirectML sucks in terms of performance I've found out. On my 6800 XT, a 512x512 generation on Stable Diffusion takes about 30 seconds on DirectML, and 5 seconds on ROCm.

What is your Stable Diffusion Webui experience with 6800XT? I'm still have to stick with a RTX 2080 because of ControlNet support

 

[sam_86314]

What is your Stable Diffusion Webui experience with 6800XT? I'm still have to stick with a RTX 2080 because of ControlNet support

Aside from some stability (ha) issues in Linux, it's great. Basically everything the card has enough VRAM to generate goes very fast.

I'm not sure how your 2080 would compare. Since it has AI-accelerating hardware; maybe it's faster?

I don't really go past 768x512 due to weird artifacts and deformities on generations. But AUTOMATIC1111 has a feature called "hires fix" that generates at a lower resolution and then adds more detail to a specified higher resolution. It works great at 512x320.

The stability issue happens when I generate an image too large for my GPU's framebuffer, where basically Linux freezes up and the only solution is to hard reset my PC.

 

[Patriot]

where windows support

DirectML sucks in terms of performance I've found out. On my 6800 XT, a 512x512 generation on Stable Diffusion takes about 30 seconds on DirectML, and 5 seconds on ROCm.

This year... should have many ROCm libraries, not everything, on windows.

Installation and Building for Windows — rocBLAS Documentation

rocblas.readthedocs.io

Some are available now.
I may have taken a gander behind the wall today, fun fact they auth After the page loads... so shadowplay helped me see things.
ROCm 5.6 Alpha with Windows support for HIP libraries.

 

[Mahboi]

Finally!

I'm gonna have to test this and try my hands at stable diffusion and some LLMs one of these weekends.

A 20Go card should bring some serious possibilities.

 

[Patriot]

Finally!

I'm gonna have to test this and try my hands at stable diffusion and some LLMs one of these weekends.

A 20Go card should bring some serious possibilities.

Someone got Stable diffusion working a couple of weeks ago on ROCm 5.5 beta, 5.5 launched today so it should be rebuild using final code soon.
5.5 silently enables 7xxx cards but they are not supported officially.

Step-by-step guide to run on RDNA3 (Linux) · AUTOMATIC1111 stable-diffusion-webui · Discussion #9591

AMD is keeping awfully quiet, but I somehow stumbled across a ROCm 5.5 release candidate Docker container that works properly on 7900XT/ 7900XTX cards - but you have to also compile PyTorch yoursel...

github.com

 

[Mahboi]

Someone got Stable diffusion working a couple of weeks ago on ROCm 5.5 beta, 5.5 launched today so it should be rebuild using final code soon.
5.5 silently enables 7xxx cards but they are not supported officially.

Step-by-step guide to run on RDNA3 (Linux) · AUTOMATIC1111 stable-diffusion-webui · Discussion #9591

AMD is keeping awfully quiet, but I somehow stumbled across a ROCm 5.5 release candidate Docker container that works properly on 7900XT/ 7900XTX cards - but you have to also compile PyTorch yoursel...

github.com

Yes I followed that since Reddit...but I am really expecting proper ROCm support anyway. This solution is a 3rd party, and just one person at that. I'd expect that if I pay 975€ for a GPU, AMD can actually properly support it