Q: How fast is DeepSeek-V4.1-Flash on the Apple M2?

~3.2–3.9 tok/s (active-expert estimate) on the Apple M2. computed roofline: 100 GB/s × 0.80–1.00 efficiency window / 25.3GB (Q4_K_M) — bands, never points. MoE: file is 168.9GB (that decides fit) but only ~82.8B of experts activate per token (~25.3GB streamed) — band below is the active-expert estimate, optimistic; community MoE rows run several× the naive bandwidth math

Question 1

Can the Apple M2 run DeepSeek-V4.1-Flash?

Accepted Answer

DeepSeek-V4.1-Flash needs about 170.9GB working set (weights 168.9GB + KV cache 2GB at 4k context). The Apple M2 offers ~6GB usable — verdict: no. 170.9 GB vs 6 GB usable — does not fit

Question 2

How fast is DeepSeek-V4.1-Flash on the Apple M2?

Accepted Answer

~3.2–3.9 tok/s (active-expert estimate)

computed roofline: 100 GB/s × 0.80–1.00 efficiency window / 25.3GB (Q4_K_M) — bands, never points.

MoE: file is 168.9GB (that decides fit) but only ~82.8B of experts activate per token (~25.3GB streamed) — band below is the active-expert estimate, optimistic; community MoE rows run several× the naive bandwidth math

Question 3

What should I run DeepSeek-V4.1-Flash on instead?

Accepted Answer

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

Can the Apple M2 run DeepSeek-V4.1-Flash?

Run DeepSeek-V4.1-Flash on instead