Llama 4 Maverick GPU requirements: 201 GB at 4-bit, 803 GB at BF16
Llama 4 Maverick needs 803 GB of GPU memory for its weights at BF16, 402 GB at FP8 and 201 GB at 4-bit, plus 2.82 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
- 400B17B active per token
- Weights, 4-bit
- 201 GBBF16: 803 GB
- KV cache, 32K
- 2.82 GBOne sequence, 16-bit cache
- Context window
- 1MNative
How much VRAM Llama 4 Maverick 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 | + 1Mcontext | Cheapest node, 32K |
|---|---|---|---|---|---|---|
| BF16Published weights, fine-tuning | 803 GB | 805 GB | 806 GB | 811 GB | 856 GB | 4× MI325X · $4.36/h |
| FP8Hopper, Ada, Blackwell, MI300 | 402 GB | 403 GB | 404 GB | 409 GB | 454 GB | 8× RTX PRO 6000 Blackwell · $1.52/h |
| 4-bitAWQ, GPTQ, GGUF Q4 | 201 GB | 202 GB | 204 GB | 208 GB | 254 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 4 Maverick, 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 · 204 GB needed | 320 GB | $0.36 | $0.76 | $262.80 | Deploy |
| 4-bit | 8× A40Ampere · 204 GB needed | 384 GB | $0.48 | $0.72 | $350.40 | Deploy |
| 4-bit | 8× A100 40GBAmpere · 204 GB needed | 320 GB | $0.72 | $1.36 | $525.60 | Deploy |
| 4-bit | 8× L40Ada Lovelace · 204 GB needed | 384 GB | $0.72 | $1.52 | $525.60 | Deploy |
| FP8 | 8× RTX PRO 6000 BlackwellBlackwell · 404 GB needed | 768 GB | $1.52 | $3.12 | $1,109.60 | Deploy |
| FP8 | 2× MI325XCDNA 3 · 404 GB needed | 512 GB | $2.18 | $3.18 | $1,591.40 | Deploy |
| FP8 | 4× H200 NVLHopper · 404 GB needed | 564 GB | $2.76 | $3.96 | $2,014.80 | Deploy |
| FP8 | 4× H200 SXMHopper · 404 GB needed | 564 GB | $3.00 | $4.36 | $2,190.00 | Deploy |
| BF16 | 4× MI325XCDNA 3 · 806 GB needed | 1024 GB | $4.36 | $6.36 | $3,182.80 | Deploy |
| BF16 | 8× H200 NVLHopper · 806 GB needed | 1128 GB | $5.52 | $7.92 | $4,029.60 | Deploy |
| BF16 | 8× H200 SXMHopper · 806 GB needed | 1128 GB | $6.00 | $8.72 | $4,380.00 | Deploy |
| BF16 | 8× MI300XCDNA 3 · 806 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.
12 global layers and 36 layers with chunked local attention (8192 tokens), 8 KV heads of 128.
| Context | KV cache, 1 sequence | 8 sequences |
|---|---|---|
| 8K tokens | 1.61 GB | 12.9 GB |
| 32K tokens | 2.82 GB | 22.5 GB |
| 128K tokens | 7.65 GB | 61.2 GB |
| 1M tokens | 52.7 GB | 422 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 — 22.5 GB here.
- Recommended node4× A100 PCIe 80GB
- Weights at 4-bit201 GB
- Context that fits, 1 sequence1M
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 4 Maverick, as published.
- Developer
- MetaMultimodal mixture of experts, 128 experts, 1M-token context
- Parameters
- 400B total, 17B active per token
- Architecture
- Mixture of experts, 48 layers12 global layers and 36 layers with chunked local attention (8192 tokens), 8 KV heads of 128
- Context window
- 1,048,576 tokensAs set in the published configuration
- Published format
- BF16, 803 GB
- Licence
- Llama 4 Community License
- Released
- 5 April 2025
- Source
- meta-llama/Llama-4-Maverick-17B-128E-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-4-Maverick-17B-128E-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 4 Maverick on a GPU — the questions that come up
How much VRAM does Llama 4 Maverick need?
For the weights alone: 803 GB at BF16, 402 GB at FP8 and 201 GB at 4-bit. Add 2.82 GB of KV cache for one 32K-token context, or 52.7 GB at the full 1M window, and keep about 10% of the memory free for activations.
Can Llama 4 Maverick 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 4 Maverick?
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 204 GB this setup needs.
Does Llama 4 Maverick need less memory because only 17B parameters are active?
No. A mixture-of-experts model routes each token through a few experts, but every expert has to be resident in GPU memory, so the memory bill follows the 400B total. What the 17B active parameters change is the work per token: generation reads far less of the weights each step than a dense model of the same size would, which is why it is fast for its size.
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-4-Maverick-17B-128E-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 4 Maverick on 4× A100 PCIe 80GB for $0.36 an hour.
Launch the node with the vLLM template, point it at meta-llama/Llama-4-Maverick-17B-128E-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.