Best GPU for Mistral Large 2 (123B): VRAM Requirements
Mistral Large 2 (123B) needs about 273.8 GB of GPU memory in BF16 and 70.9 GB at 4 bits. See which GPUs fit it for inference and fine-tuning.
Mistral Large 2 (123B) needs about 273.8 GB of VRAM in FP16, 138.5 GB in FP8 and 70.9 GB at 4 bits for inference with an 8,192-token context. In FP16 it fits on 2 × NVIDIA B200 180GB; a 4-bit build fits on one NVIDIA A100 PCIe 80GB.
Mistral Large 2 (123B) is a 123B-parameter language model from Mistral AI, released in July 2024, with a 131,072-token context window and published under the Mistral Research License. Its weights alone take 246 GB in BF16 and 61.5 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 Large 2 (123B)
Best for inference
1 GPU × NVIDIA B200 180GB
FP8 or INT8 weights, 138.5 GB at 8,192 tokens.
Best for fine-tuning
2 GPUs × NVIDIA B200 180GB
LoRA on BF16 weights, 319.8 GB.
Budget pick
1 GPU × NVIDIA RTX PRO 6000 Blackwell 96GB
Smallest card class that holds a 4-bit build, 70.9 GB.
Mistral Large 2 (123B) 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 | 246 GB | 3 GB | 273.8 GB |
| FP8 | 123 GB | 3 GB | 138.5 GB |
| INT4 | 61.5 GB | 3 GB | 70.9 GB |
Mistral Large 2 (123B) 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) | 187.3 GB |
| QLoRA (4-bit base) | 80 GB |
| LoRA (BF16 base) | 319.8 GB |
| Full fine-tuning (BF16, AdamW) | 2,164.8 GB |
GPUs that fit Mistral Large 2 (123B)
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 (273.8 GB) | FP8 inference (138.5 GB) | INT4 inference (70.9 GB) | QLoRA (80 GB) | LoRA (319.8 GB) |
|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5080 16GB | 16 GB | — | — | 8 GPUs | 8 GPUs | — |
| NVIDIA GeForce RTX 4090 24GB | 24 GB | — | 8 GPUs | 4 GPUs | 4 GPUs | — |
| NVIDIA L4 24GB | 24 GB | — | 8 GPUs | 4 GPUs | 4 GPUs | — |
| NVIDIA GeForce RTX 5090 32GB | 32 GB | — | 8 GPUs | 4 GPUs | 4 GPUs | — |
| NVIDIA Tesla V100 32GB | 32 GB | — | 8 GPUs | 4 GPUs | 4 GPUs | — |
| NVIDIA A100 PCIe 40GB | 40 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA L40S 48GB | 48 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA Quadro RTX 8000 48GB | 48 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA RTX 6000 Ada 48GB | 48 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA RTX A6000 48GB | 48 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA A100 PCIe 80GB | 80 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA A100 SXM 80GB | 80 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA H100 PCIe 80GB | 80 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA H100 SXM 80GB | 80 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA RTX PRO 6000 Blackwell 96GB | 96 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA H200 141GB | 141 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA B200 180GB | 180 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA B300 288GB | 288 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA GB300 NVL | 288 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
What is Mistral Large 2 (123B)?
Mistral Large 2 is a dense 123-billion-parameter model from Mistral AI with a 128K-token context, aimed at reasoning, coding and function calling. Version 24.11 (Mistral Large Instruct 2411) improves long context, function calling and system prompt handling over the July 2024 release.
- Developer
- Mistral AI
- Released
- July 2024
- Parameters
- 123B
- Context window
- 131,072 tokens
- License
- Mistral Research License
The Mistral Research License allows research and non-commercial use; commercial self-hosting needs a separate license from Mistral AI.
Source: Mistral Large Instruct 2411 model card (as of November 17, 2026).
Frequently asked questions
- How much VRAM does Mistral Large 2 (123B) need?
- For inference with an 8,192-token context and one request, Mistral Large 2 (123B) needs about 273.8 GB in BF16, 138.5 GB in FP8, 70.9 GB in INT4, including a 10 % runtime margin. Real usage depends on the inference engine and its settings.
- Can Mistral Large 2 (123B) run on a single GPU?
- Yes. In FP8 it needs about 138.5 GB, which fits one NVIDIA B200 180GB (180 GB). Higher precisions need a larger card or several GPUs.
- How much GPU memory does fine-tuning Mistral Large 2 (123B) take?
- QLoRA (4-bit base): about 80 GB; LoRA (BF16 base): about 319.8 GB; Full fine-tuning (BF16, AdamW): about 2,164.8 GB. Longer training sequences and larger batches add activation memory on top.
- What is the best GPU for Mistral Large 2 (123B)?
- Best for inference: 1 GPU × NVIDIA B200 180GB (FP8 or INT8 weights, 138.5 GB at 8,192 tokens). Best for fine-tuning: 2 GPUs × NVIDIA B200 180GB (LoRA on BF16 weights, 319.8 GB). Budget pick: 1 GPU × NVIDIA RTX PRO 6000 Blackwell 96GB (Smallest card class that holds a 4-bit build, 70.9 GB). Check live offers before booking, as availability changes.
- Can Mistral Large 2 (123B) run on an RTX 4090?
- Not on one card: even a 4-bit build needs about 70.9 GB, so it takes 4 RTX 4090 cards (96 GB combined) with the model split across them.
- Can Mistral Large 2 (123B) run on an RTX 5090?
- Not on one card: even a 4-bit build needs about 70.9 GB, so it takes 4 RTX 5090 cards (128 GB combined) with the model split across them.
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.