Can 4x RTX 3090 run Qwen3 235B-A22B?

Q: Can 4x RTX 3090 run Qwen3 235B-A22B?

Yes. Measured 18 tok/s generation on 4x RTX 3090, via llama.cpp, IQ4_XS. Source: https://www.reddit.com/r/LocalLLaMA/comments/1khmaah/5_commands_to_run_qwen3235ba22b_q3_inference_on/

Yes. Measured 18 tok/s generation (llama.cpp, IQ4_XS; source).

Every measured result for Qwen3 235B-A22B on 4x RTX 3090

Build	Quant / setup	Engine	tok/s
4x RTX 3090 + EPYC 7443, 256 GB DDR4 8-channel	IQ4_XS	llama.cpp	18	source ⚠ unchecked
4x RTX 3090 + 32-core Threadripper, 192 GB DDR4	UD-Q3_K_XL	llama.cpp	12-14	source ⚠ unchecked

Speeds are decode (generation) tokens/sec as reported by each source; quant, engine and context differ, so small gaps are noise.

All measured speeds for Qwen3 235B-A22B: best home rig for Qwen3 235B-A22B.

Not enough VRAM? Rent before you buy. Run Qwen3 235B-A22B on a bigger GPU on a rented 24–96 GB cloud GPU by the hour and see the real speed before spending €1,000+. Try it on RunPod
Referral link: RigBase may earn RunPod credit if you sign up. It never affects our data.

SPONSOR
Your brand here. One sponsor slot on this page, shown to people choosing AI hardware. Sponsor RigBase →