I built two open-source tools to fix AMD GPU local LLM setup (ROCmFix) and benchmark Vulkan vs HIP performance (InferBench) (lemmy.world)

submitted 4 days ago by Xanpavle@lemmy.world to c/localllama@sh.itjust.works

1 comments fedilink hide all child comments

Running local LLMs on consumer AMD GPUs often comes with two pain points: configuring backend overrides (HSA_OVERRIDE_GFX_VERSION) and figuring out whether Vulkan or HIP/ROCm gives better token throughput on your setup.

I created two open-source tools to solve this:

1. ROCmFix Repo: github.com/xanpavle/rocmfix

Zero-dependency Python script runnable via single-line terminal command.

Detects GPU PCI IDs directly (Windows Registry / Linux lspci) to map the correct HSA_OVERRIDE_GFX_VERSION.

Automatically updates shell profiles (PowerShell, CMD, Bash, Zsh, Fish) and includes an undo command and driver diagnostic runner (rocmfix doctor).

2. InferBench Repo: github.com/xanpavle/inferbench

Automated benchmarking tool for local LLM servers (Ollama, LM Studio).

Rigorously compares Vulkan vs HIP/ROCm performance with warm-up passes, median of N runs, and cold-start VRAM unloading between test cycles.

Both projects are MIT-licensed. Check them out and let me know your benchmarking results!

top 1 comments

sorted by: hot top new old

[-] Eyekaytee@aussie.zone 2 points 4 days ago

ty!

this post was submitted on 19 Sep 2026

9 points (80.0% liked)

LocalLLaMA

5169 readers

25 users here now

Welcome to LocalLLaMA! Here we discuss running and developing machine learning models at home. Lets explore cutting edge open source neural network technology together.

Get support from the community! Ask questions, share prompts, discuss benchmarks, get hyped at the latest and greatest model releases! Enjoy talking about our awesome hobby.

As ambassadors of the self-hosting machine learning community, we strive to support each other and share our enthusiasm in a positive constructive way.

Rules:

Rule 1 - No harassment or personal character attacks of community members. I.E no namecalling, no generalizing entire groups of people that make up our community, no baseless personal insults.

Rule 2 - No comparing artificial intelligence/machine learning models to cryptocurrency. I.E no comparing the usefulness of models to that of NFTs, no comparing the resource usage required to train a model is anything close to maintaining a blockchain/ mining for crypto, no implying its just a fad/bubble that will leave people with nothing of value when it burst.

Rule 3 - No comparing artificial intelligence/machine learning to simple text prediction algorithms. I.E statements such as "llms are basically just simple text predictions like what your phone keyboard autocorrect uses, and they're still using the same algorithms since <over 10 years ago>.

Rule 4 - No implying that models are devoid of purpose or potential for enriching peoples lives.

founded 3 years ago

MODERATORS

pax@sh.itjust.works

noneabove1182@sh.itjust.works

Smokeydope@lemmy.world

MonsterBug@sh.itjust.works