Llama 3.1 405B GPU requirements: 203 GB at 4-bit, 812 GB at BF16
Llama 3.1 405B needs 812 GB of GPU memory for its weights at BF16, 406 GB at FP8 and 203 GB at 4-bit, plus 16.9 GB of KV cache for a 32K context. The cheapest node that runs it here is 4× A100 PCIe 80GB at 4-bit, for $0.36 an hour on spot.
Weights plus a 32K 16-bit KV cache, with 10% of the node's memory kept free. Billed per minute.
- Parameters
- 405BDense
- Weights, 4-bit
- 203 GBBF16: 812 GB
- KV cache, 32K
- 16.9 GBOne sequence, 16-bit cache
- Context window
- 128KNative
How much VRAM Llama 3.1 405B needs, format by format.
Weights are parameters × bytes per parameter. Each context column adds the KV cache of one sequence of that length.
| Format | Weights | + 8Kcontext | + 32Kcontext | + 128Kcontext | Cheapest node, 32K |
|---|---|---|---|---|---|
| BF16Published weights, fine-tuning | 812 GB | 816 GB | 829 GB | 879 GB | 4× MI325X · $4.36/h |
| FP8Hopper, Ada, Blackwell, MI300 | 406 GB | 410 GB | 423 GB | 473 GB | 8× RTX PRO 6000 Blackwell · $1.52/h |
| 4-bitAWQ, GPTQ, GGUF Q4 | 203 GB | 207 GB | 220 GB | 271 GB | 4× A100 PCIe 80GB · $0.36/h |
The serving engine adds its own buffers, and an 8-bit KV cache halves the context columns. We keep 10% of a node's memory free when we say a model fits. Speed is a separate question: it follows memory bandwidth and, for a mixture of experts, the active parameters.
GPUs that run Llama 3.1 405B, cheapest first.
For each format, the smallest node of each GPU model that holds the weights and a 32K context, sorted by spot price. One row per GPU model; larger nodes of the same GPU work too.
| Format | GPU | GPU memory | Spot | On-demand | 730 h spot | Action |
|---|---|---|---|---|---|---|
| 4-bit | 4× A100 PCIe 80GBAmpere · 220 GB needed | 320 GB | $0.36 | $0.76 | $262.80 | Deploy |
| 4-bit | 8× A40Ampere · 220 GB needed | 384 GB | $0.48 | $0.72 | $350.40 | Deploy |
| 4-bit | 8× A100 40GBAmpere · 220 GB needed | 320 GB | $0.72 | $1.36 | $525.60 | Deploy |
| 4-bit | 8× L40Ada Lovelace · 220 GB needed | 384 GB | $0.72 | $1.52 | $525.60 | Deploy |
| FP8 | 8× RTX PRO 6000 BlackwellBlackwell · 423 GB needed | 768 GB | $1.52 | $3.12 | $1,109.60 | Deploy |
| FP8 | 2× MI325XCDNA 3 · 423 GB needed | 512 GB | $2.18 | $3.18 | $1,591.40 | Deploy |
| FP8 | 4× H200 NVLHopper · 423 GB needed | 564 GB | $2.76 | $3.96 | $2,014.80 | Deploy |
| FP8 | 4× H200 SXMHopper · 423 GB needed | 564 GB | $3.00 | $4.36 | $2,190.00 | Deploy |
| BF16 | 4× MI325XCDNA 3 · 829 GB needed | 1024 GB | $4.36 | $6.36 | $3,182.80 | Deploy |
| BF16 | 8× H200 NVLHopper · 829 GB needed | 1128 GB | $5.52 | $7.92 | $4,029.60 | Deploy |
| BF16 | 8× H200 SXMHopper · 829 GB needed | 1128 GB | $6.00 | $8.72 | $4,380.00 | Deploy |
| BF16 | 8× MI300XCDNA 3 · 829 GB needed | 1536 GB | $7.92 | $11.60 | $5,781.60 | Deploy |
Spot capacity can be reclaimed with a 2-minute notice; for a live endpoint, use on-demand. How spot works · serving a language model.
What a long context costs in memory.
Grouped-query attention, 8 KV heads of 128 across 126 layers.
| Context | KV cache, 1 sequence | 8 sequences |
|---|---|---|
| 8K tokens | 4.23 GB | 33.8 GB |
| 32K tokens | 16.9 GB | 135 GB |
| 128K tokens | 67.6 GB | 541 GB |
Serving several users
KV cacheEvery concurrent request keeps its own cache, so memory for context grows with users × context length. A server batching 8 requests of 32K needs eight times the single-sequence figure — 135 GB here.
- Recommended node4× A100 PCIe 80GB
- Weights at 4-bit203 GB
- Context that fits, 1 sequence128K
Paged attention (vLLM, SGLang) allocates the cache in blocks as requests grow, so a node holds more short requests than the worst case suggests.
Llama 3.1 405B, as published.
- Developer
- MetaThe largest dense open model
- Parameters
- 405B
- Architecture
- Dense model, 126 layersGrouped-query attention, 8 KV heads of 128 across 126 layers
- Context window
- 131,072 tokensAs set in the published configuration
- Published format
- BF16, 812 GB
- Licence
- Llama 3.1 Community License
- Released
- 23 July 2024
- Source
- meta-llama/Llama-3.1-405B-Instructconfig.json read 26 October 2026
Serve it with vLLM
template: vLLMOn 4× A100 PCIe 80GB, from the instance terminal:
$ vllm serve meta-llama/Llama-3.1-405B-Instruct \ --tensor-parallel-size 4 \ --max-model-len 32768
Very recent architectures need a recent vLLM; the model card gives the minimum version and any extra flags.
Llama 3.1 405B on a GPU — the questions that come up
How much VRAM does Llama 3.1 405B need?
For the weights alone: 812 GB at BF16, 406 GB at FP8 and 203 GB at 4-bit. Add 16.9 GB of KV cache for one 32K-token context, or 67.6 GB at the full 128K window, and keep about 10% of the memory free for activations.
Can Llama 3.1 405B run on a single GPU?
One RTX 4090 (24 GB): no. One RTX 5090 (32 GB): no. One H100 SXM (80 GB): no. One H200 SXM (141 GB): no. One B200 SXM (180 GB): no. Those answers assume a 32K context and the widest format that still fits.
What is the cheapest way to run Llama 3.1 405B?
Here, 4× A100 PCIe 80GB at 4-bit: $0.36 per hour on spot and $0.76 on-demand, billed per minute — $262.80 for a 730-hour month on spot. That node has 320 GB of GPU memory for the 220 GB this setup needs.
How long a context fits on the recommended node?
With the weights at 4-bit on 4× A100 PCIe 80GB, the remaining memory holds a 16-bit KV cache for about 131,072 tokens of context for one sequence (128K). An 8-bit KV cache doubles that; serving several users at once divides it between them.
Which format should I run it in?
BF16 keeps the published weights untouched and is the format for fine-tuning. FP8 halves the memory and runs in hardware on Hopper, Ada, Blackwell and MI300-class GPUs. 4-bit quantisation (AWQ, GPTQ, GGUF) quarters it, at a quality cost that is larger on small models than on big ones; check it against your own evaluation before relying on it.
How do I deploy it?
Pick the node in the table above and launch it with the vLLM template; the model downloads from Hugging Face (meta-llama/Llama-3.1-405B-Instruct) on first start. Spread the weights over the node with --tensor-parallel-size equal to the GPU count, and set --max-model-len to the context you need. New architectures need a recent vLLM: check the model card for the minimum version.
Sizing another model? GPU requirements for 64 open models · serving a language model · fine-tuning.
Llama 3.1 405B on 4× A100 PCIe 80GB for $0.36 an hour.
Launch the node with the vLLM template, point it at meta-llama/Llama-3.1-405B-Instruct, stop it when you are done. Prepaid, billed by the minute.
Billed per minute from the moment the instance is reachable. Minimum credit $30, no subscription.