Best GPU for Mistral Small 3.1 (24B): VRAM Requirements
Mistral Small 3.1 (24B) needs about 54.3 GB of GPU memory in BF16 and 14.7 GB at 4 bits. See which GPUs fit it for inference and fine-tuning.
Mistral Small 3.1 (24B) needs about 54.3 GB of VRAM in FP16, 27.9 GB in FP8 and 14.7 GB at 4 bits for inference with an 8,192-token context. In FP16 it fits on one NVIDIA A100 PCIe 80GB; a 4-bit build fits on one NVIDIA GeForce RTX 4090 24GB.
Mistral Small 3.1 (24B) is a 24B-parameter language model from Mistral AI, released in March 2025, with a 131,072-token context window and published under the Apache 2.0. Its weights alone take 48 GB in BF16 and 12 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 Mistral Small 3.1 (24B)
Best for inference
1 GPU × NVIDIA A100 PCIe 40GB
FP8 or INT8 weights, 27.9 GB at 8,192 tokens.
Best for fine-tuning
1 GPU × NVIDIA A100 PCIe 80GB
LoRA on BF16 weights, 62.4 GB.
Budget pick
1 GPU × NVIDIA GeForce RTX 4090 24GB
Smallest card class that holds a 4-bit build, 14.7 GB.
Mistral Small 3.1 (24B) 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 | 48 GB | 1.3 GB | 54.3 GB |
| FP8 | 24 GB | 1.3 GB | 27.9 GB |
| INT4 | 12 GB | 1.3 GB | 14.7 GB |
Mistral Small 3.1 (24B) 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 131,072-token context (FP8) | 50 GB |
| QLoRA (4-bit base) | 15.6 GB |
| LoRA (BF16 base) | 62.4 GB |
| Full fine-tuning (BF16, AdamW) | 422.4 GB |
GPUs that fit Mistral Small 3.1 (24B)
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 (54.3 GB) | FP8 inference (27.9 GB) | INT4 inference (14.7 GB) | QLoRA (15.6 GB) | LoRA (62.4 GB) |
|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5080 16GB | 16 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA GeForce RTX 4090 24GB | 24 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA L4 24GB | 24 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA GeForce RTX 5090 32GB | 32 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA Tesla V100 32GB | 32 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA A100 PCIe 40GB | 40 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA L40S 48GB | 48 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA Quadro RTX 8000 48GB | 48 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA RTX 6000 Ada 48GB | 48 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA RTX A6000 48GB | 48 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA A100 PCIe 80GB | 80 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA A100 SXM 80GB | 80 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA H100 PCIe 80GB | 80 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA H100 SXM 80GB | 80 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA RTX PRO 6000 Blackwell 96GB | 96 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| 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 Mistral Small 3.1 (24B)?
Mistral Small 3.1 builds on Mistral Small 3 with image understanding and a context window extended to 128K tokens, keeping its 24-billion-parameter dense design. Mistral states that it fits on a single RTX 4090 or a 32 GB Mac once quantized.
- Developer
- Mistral AI
- Released
- March 2025
- Parameters
- 24B
- Context window
- 131,072 tokens
- License
- Apache 2.0
The 24-billion-parameter total includes the vision encoder.
Source: Mistral Small 3.1 24B Instruct 2503 model card (as of November 17, 2026).
Frequently asked questions
- How much VRAM does Mistral Small 3.1 (24B) need?
- For inference with an 8,192-token context and one request, Mistral Small 3.1 (24B) needs about 54.3 GB in BF16, 27.9 GB in FP8, 14.7 GB in INT4, including a 10 % runtime margin. Real usage depends on the inference engine and its settings.
- Can Mistral Small 3.1 (24B) run on a single GPU?
- Yes. In BF16 it needs about 54.3 GB, which fits one NVIDIA A100 PCIe 80GB (80 GB).
- How much GPU memory does fine-tuning Mistral Small 3.1 (24B) take?
- QLoRA (4-bit base): about 15.6 GB; LoRA (BF16 base): about 62.4 GB; Full fine-tuning (BF16, AdamW): about 422.4 GB. Longer training sequences and larger batches add activation memory on top.
- What is the best GPU for Mistral Small 3.1 (24B)?
- Best for inference: 1 GPU × NVIDIA A100 PCIe 40GB (FP8 or INT8 weights, 27.9 GB at 8,192 tokens). Best for fine-tuning: 1 GPU × NVIDIA A100 PCIe 80GB (LoRA on BF16 weights, 62.4 GB). Budget pick: 1 GPU × NVIDIA GeForce RTX 4090 24GB (Smallest card class that holds a 4-bit build, 14.7 GB). Check live offers before booking, as availability changes.
- Can Mistral Small 3.1 (24B) run on an RTX 4090?
- Yes. In 4-bit it needs about 14.7 GB, which fits one RTX 4090 (24 GB). FP16 needs about 54.3 GB, 4 of these cards.
- Can Mistral Small 3.1 (24B) run on an RTX 5090?
- Yes. In FP8 it needs about 27.9 GB, which fits one RTX 5090 (32 GB). FP16 needs about 54.3 GB, 2 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.