DeepSeek-V4-Pro — which GPUs fit

DeepSeek · 861.6B (active 49B) · MoE · ctx 1000K · MIT · HuggingFace

DeepSeek-V4-Pro (DeepSeek) is a 861.6B parameters MoE (49B active per token) open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 484.68 GB VRAM for Q4 (weights 473.88 GB + KV 9.8 GB + 1 GB overhead) and 1734.0 GB for FP16. About 0 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 46.6 tok/s (estimated); MI300X 192GB ≈ 30.9 tok/s (estimated); H200 141GB SXM5 ≈ 15.1 tok/s (estimated). Use the table below for VRAM fit and tok/s per dollar (list MSRP). Estimates follow memory-bandwidth math; measured rows override when present. See methodology for details. Methodology.

7 hardware configs — sorted by tok/s. Default fit context 8K.

Hardware	VRAM	Quant	tok/s	Fits?	tok/s/$
B200 192GB SXM	192	Q4	46.6 est.	No	0.001
MI300X 192GB	192	Q4	30.9 est.	No	0.002
H200 141GB SXM5	141	Q4	15.1 est.	No	0.0
Gaudi 3 128GB	128	Q4	8.5 est.	No	0.001
Mac Studio M2 Ultra 192GB	192	Q4	4.7 est.	No	0.001
MacBook Pro M4 Max 128GB	128	Q4	1.4 est.	No	0.0
MacBook Pro M3 Max 128GB	128	Q4	1.0 est.	No	0.0