+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: Optimization & Performance (https://www.datasculpting.ai/inference/forumdisplay.php?fid=10)
+--- Forum: Quantization (https://www.datasculpting.ai/inference/forumdisplay.php?fid=40)
+--- Thread: AWQ vs GPTQ vs GGUF right now (/showthread.php?tid=152)
Pages:
1
2
AWQ vs GPTQ vs GGUF right now - phasestream53 - 09-06-2026
Did a bit of testing and thought I'd post the result here.
I tried this with FLUX using Ollama on 5090. The headline result was better than I expected, but prefix caching still ended up being the limiting factor.
Happy to post more details if anyone wants them.
Re: AWQ vs GPTQ vs GGUF right now - marbletrack - 09-07-2026
Quote:Did a bit of testing and thought I'd post the result here.
My first guess would be this: On 5090, the easiest win is often to back off the aggressive settings and see whether the problem survives. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.
Re: AWQ vs GPTQ vs GGUF right now - northquill - 09-07-2026
What worked for me was pretty simple. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at prefix caching. Once you know the plain path works, then add the faster path back in piece by piece.
Re: AWQ vs GPTQ vs GGUF right now - sanevault - 09-07-2026
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: AWQ vs GPTQ vs GGUF right now - benchplease - 09-07-2026
Good point. The pattern around structured output has been pretty similar for me.
Re: AWQ vs GPTQ vs GGUF right now - turboscope - 09-08-2026
Interesting, I hadn't tried that. The pattern around router policy has been pretty similar for me. The biggest change for me was simplifying the stack instead of adding one more optimization layer.
Re: AWQ vs GPTQ vs GGUF right now - benchplease - 09-08-2026
Quote:Interesting, I hadn't tried that. The pattern around router policy has been pretty similar for me. The biggest change for me was simplifying the stack
A couple things I'd check first. On M3 Max, the easiest win is often to back off the aggressive settings and see whether the problem survives.
Re: AWQ vs GPTQ vs GGUF right now - mangopeak72 - 09-08-2026
Interesting, I hadn't tried that. The pattern around token cost has been pretty similar for me.
Re: AWQ vs GPTQ vs GGUF right now - mangohouse - 09-10-2026
What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with Mistral and compare from there. Not glamorous advice, but repeatable baselines save a lot of time.
Re: AWQ vs GPTQ vs GGUF right now - quantshed - 09-10-2026
A couple questions before I guess too hard. Are you testing this on GB200 or something else?