Question 1

Can the NVIDIA L40S 48GB run qwen3 235B-A22B MoE?

Accepted Answer

qwen3 235B-A22B MoE needs about 142.73GB working set (weights 142GB + KV cache 0.73GB at 4k context). The NVIDIA L40S 48GB offers ~48GB usable — verdict: no. 142.7 GB vs 48 GB usable — does not fit

Question 2

How fast is qwen3 235B-A22B MoE on the NVIDIA L40S 48GB?

Accepted Answer

~30–49 tok/s (active-expert estimate) on the NVIDIA L40S 48GB. computed roofline: 864 GB/s × 0.46–0.76 efficiency window / 13.3GB (Q4_K_M) — bands, never points. MoE: file is 142GB (that decides fit) but only ~22B of experts activate per token (~13.3GB streamed) — band below is the active-expert estimate, optimistic; community MoE rows run several× the naive bandwidth math

Question 3

What should I run qwen3 235B-A22B MoE on instead?

Accepted Answer

ASUS ExpertCenter Pro ET900N G3 (GB300) runs it fully ($60000).

Can the NVIDIA L40S 48GB run qwen3 235B-A22B MoE?

Run qwen3 235B-A22B MoE on instead