Llama-3.1-70B-Instruct

Meta · 70.6B · ctx 128K · Llama 3.1 Community · HuggingFace

Llama-3.1-70B-Instruct (Meta) is a 70.6B parameters open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 53.83 GB VRAM for Q4 (weights 38.83 GB + KV 14.0 GB + 1 GB overhead) and 156.2 GB for FP16. About 11 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 520.0 tok/s (measured); MI300X 192GB ≈ 48.2 tok/s (estimated); H200 141GB SXM5 ≈ 43.6 tok/s (estimated); H100 80GB SXM5 ≈ 30.5 tok/s (estimated); Gaudi 3 128GB ≈ 29.8 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.

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

Hardware	VRAM	Quant	tok/s	Fits?	tok/s/$
B200 192GB SXM	192	Q4	520.0 measured	Yes	0.013
H100 80GB SXM5	80	FP16	180.0 measured	No	0.006
MI300X 192GB	192	FP16	165.0 measured	Yes	0.011
MI300X 192GB	192	Q4	48.2 est.	Yes	0.003
H200 141GB SXM5	141	Q4	43.6 est.	Yes	0.001
H100 80GB SXM5	80	Q4	30.5 est.	Yes	0.001
Gaudi 3 128GB	128	Q4	29.8 est.	Yes	0.002
H200 141GB SXM5	141	FP16	27.9 est.	No	0.001
B200 192GB SXM	192	FP16	20.0 est.	Yes	0.001
RTX 6000 Ada 48GB	48	Q4	19.8 est.	No	0.003
RTX A6000 48GB	48	Q4	18.0 measured	No	0.004
L40S 48GB	48	Q4	17.8 est.	No	0.002
RTX 5090 32GB	32	Q4	16.4 est.	No	0.008
Gaudi 3 128GB	128	FP16	15.7 est.	No	0.001
Mac Studio M3 Ultra 96GB	96	Q4	14.1 measured	Yes	0.004
MacBook Pro M4 Max 128GB	128	Q4	12.0 measured	Yes	0.003
MacBook Pro M3 Max 128GB	128	Q4	9.0 measured	Yes	0.002
A100 80GB SXM4	80	Q4	8.7 measured	Yes	0.001
Mac Studio M2 Ultra 192GB	192	Q4	8.0 measured	Yes	0.001
RTX 4090D 24GB (China)	24	Q4	5.2 est.	No	0.003
RTX 3090 Ti 24GB	24	Q4	5.2 est.	No	0.003
RX 7900 XTX 24GB	24	Q4	5.0 est.	No	0.005
Mac mini M4 Pro 64GB	64	Q4	5.0 measured	Yes	0.002
RTX 3090 24GB	24	Q4	4.8 est.	No	0.003
A100 80GB SXM4	80	FP16	3.8 est.	No	0.0
RTX 4090 24GB	24	Q4	3.1 measured	No	0.002
RX 7900 XT 20GB	20	Q4	2.9 est.	No	0.003
MacBook Pro M4 Max 128GB	128	FP16	2.6 est.	No	0.001
RTX 5080 16GB	16	Q4	2.2 est.	No	0.002
Mac Studio M3 Ultra 96GB	96	FP16	2.2 est.	No	0.001
RTX 5070 Ti 16GB	16	Q4	2.1 est.	No	0.003
Mac Studio M2 Ultra 192GB	192	FP16	2.0 est.	Yes	0.0
MacBook Pro M3 Max 128GB	128	FP16	1.9 est.	No	0.0
RTX 4080 Super 16GB	16	Q4	1.7 est.	No	0.002
RTX 4070 Ti Super 16GB	16	Q4	1.5 est.	No	0.002
RX 9070 16GB	16	Q4	1.5 est.	No	0.003
RX 9070 XT 16GB	16	Q4	1.5 est.	No	0.003
RX 7800 XT 16GB	16	Q4	1.4 est.	No	0.003
RX 7900 GRE 16GB	16	Q4	1.3 est.	No	0.002
Arc A770 16GB	16	Q4	1.3 est.	No	0.004
Mac mini M4 32GB	32	Q4	1.1 est.	No	0.001
RTX 5060 Ti 16GB	16	Q4	1.0 est.	No	0.002
T4 16GB	16	Q4	0.7 est.	No	0.0
RTX 4060 Ti 16GB	16	Q4	0.7 est.	No	0.001
RX 7600 XT 16GB	16	Q4	0.7 est.	No	0.002
L40S 48GB	48	FP16	0.6 est.	No	0.0
RTX 6000 Ada 48GB	48	FP16	0.6 est.	No	0.0
RTX A6000 48GB	48	FP16	0.5 est.	No	0.0
Mac mini M4 Pro 64GB	64	FP16	0.3 est.	No	0.0

Fit checks

• RTX 5090 32GB — does not fit · ≈16.4 tok/s Q4 est.
• RTX 3090 Ti 24GB — does not fit · ≈5.2 tok/s Q4 est.
• RTX 4090 24GB — does not fit · ≈3.1 tok/s Q4 measured
• RTX 5080 16GB — does not fit · ≈2.2 tok/s Q4 est.
• RTX 6000 Ada 48GB — does not fit · ≈19.8 tok/s Q4 est.
• RX 7900 XTX 24GB — does not fit · ≈5.0 tok/s Q4 est.
• RTX 3090 24GB — does not fit · ≈4.8 tok/s Q4 est.
• RTX 5070 Ti 16GB — does not fit · ≈2.1 tok/s Q4 est.
• L40S 48GB — does not fit · ≈17.8 tok/s Q4 est.
• Mac Studio M3 Ultra 96GB — fits · ≈14.1 tok/s Q4 measured
• Mac Studio M2 Ultra 192GB — fits · ≈8.0 tok/s Q4 measured
• RX 7900 XT 20GB — does not fit · ≈2.9 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 3.3 Llama-3.3-70B-Instruct
Llama 3.2 Llama-3.2-3B-Instruct
Llama 4 Llama-4-Scout-17B-16E-Instruct

All Llama models →