+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: Edge AI & Gadgets (https://www.datasculpting.ai/inference/forumdisplay.php?fid=21)
+--- Forum: SBCs / Mini PCs (https://www.datasculpting.ai/inference/forumdisplay.php?fid=74)
+--- Thread: SBC inference experiments: early impressions (/showthread.php?tid=410)
Pages:
1
2
SBC inference experiments: early impressions - homelabhenry - 09-22-2026
Thought I'd post a small recipe because a lot of threads ask similar things.
Just sharing a setup that ended up working reasonably well: Llama, vLLM, 7900 XTX, with a lot of attention on small local stacks.
Feel free to tear it apart if you see something dumb.
Re: SBC inference experiments: early impressions - batchfield55 - 09-23-2026
Ran a quick check on my box. On mini PC with llama.cpp, I saw basically the same behavior. The surprise was that memory pressure showed up before throughput collapsed.
Re: SBC inference experiments: early impressions - byteyard - 09-26-2026
What exact setup are you running? What model, what runtime, and roughly what context size?
Re: SBC inference experiments: early impressions - vramghost - 09-29-2026
Quote:What exact setup are you running? What model, what runtime, and roughly what context size?
Ran a quick check on my box. My result was mixed: first run looked bad, second run looked normal. The surprise was that memory pressure showed up before throughput collapsed.
Re: SBC inference experiments: early impressions - homelabhenry - 09-29-2026
Good point. That matches what I saw on M3 Max. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.
Re: SBC inference experiments: early impressions - quantsupply - 09-29-2026
Appreciate the detail. The pattern around continuous batching has been pretty similar for me.
Re: SBC inference experiments: early impressions - phasespark - 09-29-2026
Quote:Appreciate the detail. The pattern around continuous batching has been pretty similar for me.
A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at batch size. Not glamorous advice, but repeatable baselines save a lot of time.
Re: SBC inference experiments: early impressions - sparrowworks88 - 10-01-2026
That matches what I was seeing. I don't think your conclusion is far off.
Re: SBC inference experiments: early impressions - promptfield - 10-04-2026
Appreciate the detail. I don't think your conclusion is far off. The biggest change for me was simplifying the stack instead of adding one more optimization layer.
Re: SBC inference experiments: early impressions - lablab - 10-04-2026
What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with Qwen and compare from there. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.