Llama-4-Scout-17B-16E-Instruct

Meta · 109.0B (active 17.0B) · MoE · ctx 10240K · llama4-community · HuggingFace

Llama-4-Scout-17B-16E-Instruct (Meta) is a 109.0B parameters MoE (17.0B active per token) open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 64.35 GB VRAM for Q4 (weights 59.95 GB + KV 3.4 GB + 1 GB overhead) and 222.4 GB for FP16. About 10 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 299.5 tok/s (estimated); MI300X 192GB ≈ 198.4 tok/s (estimated); H200 141GB SXM5 ≈ 179.7 tok/s (estimated); H100 80GB SXM5 ≈ 125.4 tok/s (estimated); Gaudi 3 128GB ≈ 122.6 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.

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

Hardware	VRAM	Quant	tok/s	Fits?	tok/s/$
B200 192GB SXM	192	Q4	299.5 est.	Yes	0.007
MI300X 192GB	192	Q4	198.4 est.	Yes	0.013
H200 141GB SXM5	141	Q4	179.7 est.	Yes	0.005
B200 192GB SXM	192	FP16	175.4 est.	No	0.004
H100 80GB SXM5	80	Q4	125.4 est.	Yes	0.004
Gaudi 3 128GB	128	Q4	122.6 est.	Yes	0.01
MI300X 192GB	192	FP16	116.2 est.	No	0.008
A100 80GB SXM4	80	Q4	76.3 est.	Yes	0.005
RTX 6000 Ada 48GB	48	Q4	57.1 est.	No	0.008
H200 141GB SXM5	141	FP16	56.7 est.	No	0.002
L40S 48GB	48	Q4	51.4 est.	No	0.006
RTX 5090 32GB	32	Q4	47.4 est.	No	0.024
RTX A6000 48GB	48	Q4	45.7 est.	No	0.01
Gaudi 3 128GB	128	FP16	31.9 est.	No	0.003
Mac Studio M3 Ultra 96GB	96	Q4	30.7 est.	Yes	0.008
Mac Studio M2 Ultra 192GB	192	Q4	29.9 est.	Yes	0.004
Mac mini M4 Pro 64GB	64	Q4	28.9 est.	No	0.013
MacBook Pro M4 Max 128GB	128	Q4	20.4 est.	Yes	0.005
Mac Studio M2 Ultra 192GB	192	FP16	17.5 est.	No	0.003
RTX 4090 24GB	24	Q4	15.0 est.	No	0.009
RTX 4090D 24GB (China)	24	Q4	15.0 est.	No	0.009
MacBook Pro M3 Max 128GB	128	Q4	15.0 est.	Yes	0.004
RTX 3090 Ti 24GB	24	Q4	15.0 est.	No	0.008
RX 7900 XTX 24GB	24	Q4	14.3 est.	No	0.014
RTX 3090 24GB	24	Q4	13.9 est.	No	0.009
H100 80GB SXM5	80	FP16	12.7 est.	No	0.0
RX 7900 XT 20GB	20	Q4	8.3 est.	No	0.009
A100 80GB SXM4	80	FP16	7.8 est.	No	0.001
MacBook Pro M4 Max 128GB	128	FP16	5.3 est.	No	0.001
Mac Studio M3 Ultra 96GB	96	FP16	4.5 est.	No	0.001
MacBook Pro M3 Max 128GB	128	FP16	3.9 est.	No	0.001
Mac mini M4 32GB	32	Q4	3.2 est.	No	0.003
Mac mini M4 Pro 64GB	64	FP16	0.7 est.	No	0.0

Fit checks

• RTX 5090 32GB — does not fit · ≈47.4 tok/s Q4 est.
• RTX 3090 Ti 24GB — does not fit · ≈15.0 tok/s Q4 est.
• RTX 4090 24GB — does not fit · ≈15.0 tok/s Q4 est.
• RTX 6000 Ada 48GB — does not fit · ≈57.1 tok/s Q4 est.
• RX 7900 XTX 24GB — does not fit · ≈14.3 tok/s Q4 est.
• RTX 3090 24GB — does not fit · ≈13.9 tok/s Q4 est.
• L40S 48GB — does not fit · ≈51.4 tok/s Q4 est.
• Mac Studio M3 Ultra 96GB — fits · ≈30.7 tok/s Q4 est.
• Mac Studio M2 Ultra 192GB — fits · ≈29.9 tok/s Q4 est.
• RX 7900 XT 20GB — does not fit · ≈8.3 tok/s Q4 est.

12 popular retail GPUs; every row in the table above also links its Fits? verdict to the full fit check.

Same family: Llama

Llama 3.1 Llama-3.1-8B-Instruct · Llama-3.1-70B-Instruct
Llama 3.3 Llama-3.3-70B-Instruct
Llama 3.2 Llama-3.2-3B-Instruct

All Llama models →