[Feature]: Docker build for native support gfx1100 with gfx1201

[djdeniro]

🚀 The feature, motivation and pitch

Hello, we have a stack 8 AMD gpu, 6x7900xtx + 2 new R9700, can you create a docker image with ROCm 7.0 for this two archs?

Alternatives

Already using image rocm/vllm-dev:nightly_main_20250911 with Issue on idle mode:

detailed reddit post

Additional context

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[gshtras]

ROCm 7.0 will be used in the next official release, which has gfx1100 and 1201 support built in alongside others.
The tentative date for this release is Sep 30, but the release candidate preview dockers will be available sooner, possibly later this week.

[ajaxdude]

All I want is a docker of vllm that will run for my 7900XT. Right now it is near impossible to understand how to get it to work

[djdeniro]

All I want is a docker of vllm that will run for my 7900XT. Right now it is near impossible to understand how to get it to work

base version easy to run from my example, you can replace with other model and it will works!

last commits still not work for me, i launch model directly with transformers and it work, but super slow

[djdeniro]

problem still active for vllm-dev:nightly_main_20251021