cd /news/ai-infrastructure/cuda-rocm-dual-back-ends-llm-inferen… · home › topics › ai-infrastructure › article
[ARTICLE · art-166675] src=github.com ↗ pub= topic=ai-infrastructure verified=true sentiment=↑ positive

CUDA+ROCm dual-back ends LLM inference for mixed-vendor GPUs (up to 48% faster)

A contributor to the project-maya repository opened a pull request adding a separately built HIP library that lets LLM inference run on NVIDIA and AMD GPUs in the same machine, with the owner reporting up to 48% faster performance. The owner of mw00/project-maya said the change stays off unless the plugin path is set and leaves the normal build and output unchanged, and that the team wants to test it on its own NVIDIA + AMD Linux setup before merging, which "may take a few days." The owner will document the settings under their MAYA_ names, keep the test folder to the parts CI runs, and fit the change on top of the GLM helper work from pull requests #849–#851.

read2 min views1 publishedNov 10, 2026
CUDA+ROCm dual-back ends LLM inference for mixed-vendor GPUs (up to 48% faster)
Image: Michielbdejong (auto-discovered)

Repository navigation #

Conversation #

RenWuJun force-pushedthe hengha/glm-hip-helper branch from

to 8ab866d

bac40a6 Compare

November 6, 2026 02:08 Owner

| Thanks a lot for this - running an NVIDIA card and an AMD GPU together in one PC is something we really want, and the split into a separately built HIP library with a small versioned interface is a clean way to do it. We read it through: it stays off unless the plugin path is set, and the normal build and output don't change. We'd like to run it on our own NVIDIA + AMD Linux setup before we bring it in, so it may take a few days. On our side we'll list the settings under their MAYA_ names in the docs, keep the test folder to the parts CI runs, and fit it on top of the GLM helper changes from #849 -#851 - we'll handle that ourselves. |

  This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.

Learn more about bidirectional Unicode characters Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.

source & further reading
github.com — original article Product analytics events, nearly too cheap to meter Exact – a new cross platform UI framework by the Expo team Show HN: Vork – headless 3D and VFX engine for agents
── more in #ai-infrastructure 4 stories · sorted by recency
── more on @project-maya 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/cuda-rocm-dual-back-…] indexed:0 read:2min 2026-11-10 · —