GLM-4.5 — which GPUs fit

Z.ai · 358.3B (active 32B) · MoE · ctx 128K · MIT · HuggingFace

GLM-4.5 (Z.ai) is a 358.3B parameters MoE (32B active per token) open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 204.47 GB VRAM for Q4 (weights 197.07 GB + KV 6.4 GB + 1 GB overhead) and 724.0 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 ≈ 400.8 tok/s (estimated); MI300X 192GB ≈ 265.5 tok/s (estimated); H200 141GB SXM5 ≈ 129.7 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.

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

Hardware	VRAM	Quant	tok/s	Fits?	tok/s/$
B200 192GB SXM	192	Q4	400.8 est.	No	0.01
MI300X 192GB	192	Q4	265.5 est.	No	0.018
H200 141GB SXM5	141	Q4	129.7 est.	No	0.004
Gaudi 3 128GB	128	Q4	72.9 est.	No	0.006
Mac Studio M2 Ultra 192GB	192	Q4	40.1 est.	No	0.006
H100 80GB SXM5	80	Q4	29.1 est.	No	0.001
A100 80GB SXM4	80	Q4	17.7 est.	No	0.001
MacBook Pro M4 Max 128GB	128	Q4	12.2 est.	No	0.003
Mac Studio M3 Ultra 96GB	96	Q4	10.3 est.	No	0.003
MacBook Pro M3 Max 128GB	128	Q4	8.9 est.	No	0.002
B200 192GB SXM	192	FP16	8.8 est.	No	0.0
MI300X 192GB	192	FP16	5.8 est.	No	0.0
Mac mini M4 Pro 64GB	64	Q4	1.5 est.	No	0.001
Mac Studio M2 Ultra 192GB	192	FP16	0.9 est.	No	0.0