← AI Hashrate

Llama-3.1-405B-Instruct

Meta · 405.9B · ctx 128K · Llama 3.1 Community · HuggingFace

Llama-3.1-405B-Instruct (Meta) is a 405.9B parameters open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 305.25 GB VRAM for Q4 (weights 223.25 GB + KV 81.0 GB + 1 GB overhead) and 893.8 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 ≈ 14.2 tok/s (estimated); MI300X 192GB ≈ 9.4 tok/s (estimated); H200 141GB SXM5 ≈ 4.6 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.

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

HardwareVRAMQuanttok/sFits?tok/s/$
B200 192GB SXM192Q414.2 est.No0.0
MI300X 192GB192Q49.4 est.No0.001
H200 141GB SXM5141Q44.6 est.No0.0
Gaudi 3 128GB128Q42.6 est.No0.0
Mac Studio M2 Ultra 192GB192Q41.4 est.No0.0
H100 80GB SXM580Q41.0 est.No0.0
A100 80GB SXM480Q40.6 est.No0.0
MacBook Pro M4 Max 128GB128Q40.4 est.No0.0
Mac Studio M3 Ultra 96GB96Q40.4 est.No0.0
MacBook Pro M3 Max 128GB128Q40.3 est.No0.0