These last few weeks have been godsend for 24GB (and below) gpu poor peeps. Killer models released (Gemma 4 / Qwen 3.6) Free intelligence via QAT Bonus speed via MTP We're at the tipping point where GPU poor (24gb and below) people are actually NOT poor any more. I was already happy with Gemma 4 31b running at 40tok/s but now its 70-80tok/s Its not a wonder 3090 prices are increasing. For ref: - limit=1, OSL=192, concurrency 1, temp=1.0/top_k=64/top_p=0.95, ctx=40960, q8_0 KV ...