I have the following rig
- B550M PRO-VDH WIFI (MS-7C95)
- 6-core model: AMD Ryzen 5 3600 bits: 64 type: MT MCP arch: Zen 2
- Intel Battlemage G21 [Arc B580]
- 2x DDR4 3600 KHX3600C18D4/16GX
I would say I’m getting decent results with Gemma3:12b in terms of response time and token generation.
My goal is to, in the end, have something equivalent to Fara-7b GitHub - microsoft/fara: Fara-7B: An Efficient Agentic Model for Computer Use on this rig, without it becoming an unbearable slog. I have not been able to make Fara-7b run as is on my Ubuntu-based setup, and I have understood that the culprit is my Intel B580 card.
Now, hardware wise, I would like to understand what or if there’s anything I could upgrade to improve performance on running models on this. I’m mainly using this as a hobby and I’m on a budget of somewhere around $500 as it stands right now.
I’m currently eyeing a DDR4 2x32-kit, but I don’t know if this would improve anything at all when it comes to AI performance.
So, in terms of performance for AI, is my current path buying more RAM improving anything, or would I do better in getting something else? I’ve very recently acquired the B580 and would like to avoid having to replace that if possible.