Mixtral 8x7B GPU requirements: 23.4 GB at 4-bit, 93.4 GB at BF16
Mixtral 8x7B needs 93.4 GB of GPU memory for its weights at BF16, 46.7 GB at FP8 and 23.4 GB at 4-bit, plus 4.29 GB of KV cache for a 32K context. The cheapest node that runs it here is 2× RTX A4000 at 4-bit, for $0.06 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
- 46.7B12.9B active per token
- Weights, 4-bit
- 23.4 GBBF16: 93.4 GB
- KV cache, 32K
- 4.29 GBOne sequence, 16-bit cache
- Context window
- 32KNative
How much VRAM Mixtral 8x7B 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 | Cheapest node, 32K |
|---|---|---|---|---|
| BF16Published weights, fine-tuning | 93.4 GB | 94.5 GB | 97.7 GB | 2× A100 PCIe 80GB · $0.18/h |
| FP8Hopper, Ada, Blackwell, MI300 | 46.7 GB | 47.8 GB | 51.0 GB | 2× RTX PRO 4500 Blackwell · $0.18/h |
| 4-bitAWQ, GPTQ, GGUF Q4 | 23.4 GB | 24.4 GB | 27.6 GB | 2× RTX A4000 · $0.06/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 Mixtral 8x7B, 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 | 2× RTX A4000Ampere · 27.6 GB needed | 32 GB | $0.06 | $0.10 | $43.80 | Deploy |
| 4-bit | 1× A40Ampere · 27.6 GB needed | 48 GB | $0.06 | $0.09 | $43.80 | Deploy |
| 4-bit | 2× T4Turing · 27.6 GB needed | 32 GB | $0.08 | $0.14 | $58.40 | Deploy |
| 4-bit | 1× RTX PRO 4500 BlackwellBlackwell · 27.6 GB needed | 32 GB | $0.09 | $0.19 | $65.70 | Deploy |
| FP8 | 2× RTX PRO 4500 BlackwellBlackwell · 51.0 GB needed | 64 GB | $0.18 | $0.38 | $131.40 | Deploy |
| FP8 | 2× RTX 5000 AdaAda Lovelace · 51.0 GB needed | 64 GB | $0.18 | $0.34 | $131.40 | Deploy |
| FP8 | 2× RTX 5090Blackwell · 51.0 GB needed | 64 GB | $0.18 | $0.38 | $131.40 | Deploy |
| FP8 | 2× L40Ada Lovelace · 51.0 GB needed | 96 GB | $0.18 | $0.38 | $131.40 | Deploy |
| BF16 | 2× A100 PCIe 80GBAmpere · 97.7 GB needed | 160 GB | $0.18 | $0.38 | $131.40 | Deploy |
| BF16 | 4× V100 32GBVolta · 97.7 GB needed | 128 GB | $0.24 | $0.36 | $175.20 | Deploy |
| BF16 | 4× A40Ampere · 97.7 GB needed | 192 GB | $0.24 | $0.36 | $175.20 | Deploy |
| BF16 | 4× RTX PRO 4500 BlackwellBlackwell · 97.7 GB needed | 128 GB | $0.36 | $0.76 | $262.80 | 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 32 layers.
| Context | KV cache, 1 sequence | 8 sequences |
|---|---|---|
| 8K tokens | 1.07 GB | 8.59 GB |
| 32K tokens | 4.29 GB | 34.4 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 — 34.4 GB here.
- Recommended node2× RTX A4000
- Weights at 4-bit23.4 GB
- Context that fits, 1 sequence32K
Paged attention (vLLM, SGLang) allocates the cache in blocks as requests grow, so a node holds more short requests than the worst case suggests.
Mixtral 8x7B, as published.
- Developer
- Mistral AIClassic sparse mixture of experts
- Parameters
- 46.7B total, 12.9B active per token
- Architecture
- Mixture of experts, 32 layersGrouped-query attention, 8 KV heads of 128 across 32 layers
- Context window
- 32,768 tokensAs set in the published configuration
- Published format
- BF16, 93.4 GB
- Licence
- Apache 2.0
- Released
- 11 December 2023
- Source
- mistralai/Mixtral-8x7B-Instruct-v0.1config.json read 26 October 2026
Serve it with vLLM
template: vLLMOn 2× RTX A4000, from the instance terminal:
$ vllm serve mistralai/Mixtral-8x7B-Instruct-v0.1 \ --tensor-parallel-size 2 \ --max-model-len 32768
Very recent architectures need a recent vLLM; the model card gives the minimum version and any extra flags.
Mixtral 8x7B on a GPU — the questions that come up
How much VRAM does Mixtral 8x7B need?
For the weights alone: 93.4 GB at BF16, 46.7 GB at FP8 and 23.4 GB at 4-bit. Add 4.29 GB of KV cache for one 32K-token context, and keep about 10% of the memory free for activations.
Can Mixtral 8x7B run on a single GPU?
One RTX 4090 (24 GB): no. One RTX 5090 (32 GB): yes, at 4-bit. One H100 SXM (80 GB): yes, at FP8. One H200 SXM (141 GB): yes, at BF16. One B200 SXM (180 GB): yes, at BF16. Those answers assume a 32K context and the widest format that still fits.
What is the cheapest way to run Mixtral 8x7B?
Here, 2× RTX A4000 at 4-bit: $0.06 per hour on spot and $0.10 on-demand, billed per minute — $43.80 for a 730-hour month on spot. That node has 32 GB of GPU memory for the 27.6 GB this setup needs.
Does Mixtral 8x7B need less memory because only 12.9B 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 46.7B total. What the 12.9B 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 (mistralai/Mixtral-8x7B-Instruct-v0.1) 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.
Mixtral 8x7B on 2× RTX A4000 for $0.06 an hour.
Launch the node with the vLLM template, point it at mistralai/Mixtral-8x7B-Instruct-v0.1, 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.