How to get startet with my setup

Hallo there!

I have really tried to google, use claude/chatgpt, and searched this forum, but without any major success.

I have 2x Arc Pro B70 cards which I wish to try and use for running llama.cpp.

But I can’t seem to find any idiot proof guide to where to start.

I know this:

  1. I need to use llama.cpp SYCL (llama.cpp/docs/backend/SYCL.md at master · ggml-org/llama.cpp · GitHub)
  2. I need to install the correct drivers for the cards

I just don’t exactly trust the responses I get from chatgpt..

It might be a language thing, English isn’t my first language..

I have installed ubuntu server 24.04 and updated everything.

I could just willy nilly try and install llama.cpp, but I wan’t to make sure to install stuff in the right order..

Could some one please link an article which explains how and why I need to install stuff?

Thank you!

1 Like

Hi, I like the choice of Username, welcome!

I could only help with German if that helps better. Generally all guides unfortunately will be primarily in English (some Russian or Chinese, but need to use their own search engines). Me, not having an Arc can’t give a tested guide, but maybe help in getting you started.

Fortunately, it seems you are not running on a production machine or anything you need to use on a daily basis, so do not be discouraged in trying different things to get things to work, note down what worked, and post anything you did and how it failed (with input + output messages), so people can assist in providing potential next steps.

The link you provided has the best guide there can be, especially as it hints into the first actual todo after the basic install: Installing the driver (Setting up the environment)
Being on older ubuntu, but new enough, it might help moving on to using “LTS 2523.x Releases” instead of the linked “LTS 2350.x Releases”.
Or, try with latest version for a start: Ubuntu 26.04 (no upgrade path from 24.04 out of the box, sorry). This ships with latest kernel (7+) and has a driver built in for you, ready to use. And latest possible kernel is always a good thing, especially for fast moving targets like LLMs.

Afterwards, the llama.cpp wiki has a nice summary of steps:

git clone --depth=1 https://github.com/ggerganov/llama.cpp
cd llama.cpp
cmake -Bbuild
cmake --build build -D...
cd build
cpack -G DEB
dpkg -i *.deb

(I personally prefer building things as a dedicated user, and not root, but the tool running will be needing very low level access, and it being a dedicated machine… writing this to have it noted)

Hope this helps. If anything, don’t get discouraged, software is a tricky thing, stuff might go wrong where it never did before. Likely there are people who can assist. Posting your actual hardware around the Intel Arc GPUs might help as well for future debugging.

1 Like

Hi,

wanted to followup and see if it worked out for you, or if there had been any trouble at any step that you might want to share the solution you found to. Might help others or allow the actual docs to be updated :slight_smile: