Llama 3.1 405B Instruct

Running Llama 3.1 405B Instruct locally

At Q4_K_M and 4K of context, Llama 3.1 405B Instruct needs about 258.7 GB — 233.2 GB of weights, 1.97 GB of KV cache and a 23.5 GB activation buffer. The smallest card in this index that clears that comfortably is the Mac M5 Ultra 512G at 512 GB, and 1 of the 63 cards here do. These are calculated figures, not measurements: the estimate stops counting a card as comfortable at 88% of its VRAM, which is roughly the room a desktop session needs.

What longer context costs

Going from 4K to 32K adds about 13.78 GB, taking the total to 273.9 GB. Weights do not move with context — only the KV cache does, and it grows linearly, so this is the number to watch when planning for long documents. The model's native window is 128K; holding all of it at Q4_K_M would need about 326 GB.

Which build to download

This index tracks 2 formats for it — GGUF, AWQ — across 3 levels. Q4_K_M carries the lowest published perplexity loss at 2.3%. The fastest level measured here is AWQ INT4 at 10 tok/s on an RTX 4090, batch 1. GGUF runs on llama.cpp and Ollama across NVIDIA, AMD and Apple silicon; AWQ and GPTQ target vLLM on CUDA and ROCm; EXL2 is ExLlamaV2 and CUDA only.

Common questions

How much VRAM does Llama 3.1 405B Instruct need?

About 258.7 GB at Q4_K_M with 4K of context and batch 1, rising to roughly 273.9 GB at 32K. That figure is weights plus KV cache plus a 10% activation buffer, calculated from the model's architecture rather than measured on a card.

Will Llama 3.1 405B Instruct run on a 512GB GPU?

Yes — at Q4_K_M and 4K context it needs about 258.7 GB, which leaves 253.3 GB spare on a Mac M5 Ultra 512G. That is the smallest card in this index that clears it comfortably; 1 of 63 do.

Which quantization of Llama 3.1 405B Instruct should I use?

Q4_K_M has the lowest published quality loss (2.3%), and Q4_K_M is the level most people run. All 3 levels in the index are Q4_K_M, Q3_K_M, AWQ INT4.