Best GPU for Mixtral 8x7B: VRAM Requirements
Mixtral 8x7B needs about 103.9 GB of GPU memory in BF16 and 26.9 GB at 4 bits. See which GPUs fit it for inference and fine-tuning.
Mixtral 8x7B needs about 103.9 GB of VRAM in FP16, 52.6 GB in FP8 and 26.9 GB at 4 bits for inference with an 8,192-token context. In FP16 it fits on one NVIDIA H200 141GB; a 4-bit build fits on one NVIDIA GeForce RTX 5090 32GB.
Mixtral 8x7B is a mixture-of-experts language model with 46.7B total and 12.9B active parameters per token from Mistral AI, released in December 2023, with a 32,768-token context window and published under the Apache 2.0. Its weights alone take 93.4 GB in BF16 and 23.4 GB at 4 bits per parameter; the tables below add the KV cache and runtime memory, then list the GPUs that hold each setup.
Recommended GPUs for Mixtral 8x7B
Best for inference
1 GPU × NVIDIA A100 PCIe 80GB
FP8 or INT8 weights, 52.6 GB at 8,192 tokens.
Best for fine-tuning
1 GPU × NVIDIA H200 141GB
LoRA on BF16 weights, 121.4 GB.
Budget pick
1 GPU × NVIDIA GeForce RTX 5090 32GB
Smallest card class that holds a 4-bit build, 26.9 GB.
Mixtral 8x7B VRAM requirements by precision
One request with an 8,192-token context, KV cache in BF16, plus a 10 % margin for activations and runtime buffers.
| Precision | Weights | KV cache | Total for inference |
|---|---|---|---|
| BF16 / FP16 | 93.4 GB | 1.1 GB | 103.9 GB |
| FP8 | 46.7 GB | 1.1 GB | 52.6 GB |
| INT4 | 23.4 GB | 1.1 GB | 26.9 GB |
Mixtral 8x7B VRAM by task
Fine-tuning rows assume a short sequence, batch 1 and gradient checkpointing; longer sequences add activation memory.
| Task | GPU memory |
|---|---|
| Inference at the full 32,768-token context (FP8) | 56.1 GB |
| QLoRA (4-bit base) | 30.4 GB |
| LoRA (BF16 base) | 121.4 GB |
| Full fine-tuning (BF16, AdamW) | 821.9 GB |
GPUs that fit Mixtral 8x7B
Number of cards whose combined memory holds each setup, assuming the model is split across them with tensor or pipeline parallelism. A dash means more than eight cards.
| GPU | Memory | BF16 inference (103.9 GB) | FP8 inference (52.6 GB) | INT4 inference (26.9 GB) | QLoRA (30.4 GB) | LoRA (121.4 GB) |
|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5080 16GB | 16 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA GeForce RTX 4090 24GB | 24 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA L4 24GB | 24 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA GeForce RTX 5090 32GB | 32 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA Tesla V100 32GB | 32 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA A100 PCIe 40GB | 40 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA L40S 48GB | 48 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA Quadro RTX 8000 48GB | 48 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA RTX 6000 Ada 48GB | 48 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA RTX A6000 48GB | 48 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA A100 PCIe 80GB | 80 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA A100 SXM 80GB | 80 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA H100 PCIe 80GB | 80 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA H100 SXM 80GB | 80 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA RTX PRO 6000 Blackwell 96GB | 96 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA H200 141GB | 141 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA B200 180GB | 180 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA B300 288GB | 288 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA GB300 NVL | 288 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
What is Mixtral 8x7B?
Mixtral 8x7B is a sparse mixture-of-experts model: each layer has eight expert networks and a router picks two per token. It holds 46.7 billion parameters in memory but computes about 12.9 billion per token, so it needs the memory of a 47B model with the speed of a 13B one.
- Developer
- Mistral AI
- Released
- December 2023
- Parameters
- 46.7B total, 12.9B active per token
- Context window
- 32,768 tokens
- License
- Apache 2.0
Source: Mixtral 8x7B Instruct v0.1 model card (as of November 17, 2026).
Frequently asked questions
- How much VRAM does Mixtral 8x7B need?
- For inference with an 8,192-token context and one request, Mixtral 8x7B needs about 103.9 GB in BF16, 52.6 GB in FP8, 26.9 GB in INT4, including a 10 % runtime margin. Real usage depends on the inference engine and its settings.
- Can Mixtral 8x7B run on a single GPU?
- Yes. In BF16 it needs about 103.9 GB, which fits one NVIDIA H200 141GB (141 GB).
- How much GPU memory does fine-tuning Mixtral 8x7B take?
- QLoRA (4-bit base): about 30.4 GB; LoRA (BF16 base): about 121.4 GB; Full fine-tuning (BF16, AdamW): about 821.9 GB. Longer training sequences and larger batches add activation memory on top.
- What is the best GPU for Mixtral 8x7B?
- Best for inference: 1 GPU × NVIDIA A100 PCIe 80GB (FP8 or INT8 weights, 52.6 GB at 8,192 tokens). Best for fine-tuning: 1 GPU × NVIDIA H200 141GB (LoRA on BF16 weights, 121.4 GB). Budget pick: 1 GPU × NVIDIA GeForce RTX 5090 32GB (Smallest card class that holds a 4-bit build, 26.9 GB). Check live offers before booking, as availability changes.
- Can Mixtral 8x7B run on an RTX 4090?
- Not on one card: even a 4-bit build needs about 26.9 GB, so it takes 2 RTX 4090 cards (48 GB combined) with the model split across them.
- Can Mixtral 8x7B run on an RTX 5090?
- Yes. In 4-bit it needs about 26.9 GB, which fits one RTX 5090 (32 GB). FP16 needs about 103.9 GB, 4 of these cards.
Estimates use the formulas of the LLM VRAM calculator (decimal GB). Real usage depends on the inference engine, quantization format and memory fragmentation: test on the target GPU before a long rental.
Explore the live market
Every offer on the live market comes from a reviewed supplier and is sold by EverXCloud Inc.