Skip to content

Best GPU for Codestral 22B: VRAM Requirements

Codestral 22B needs about 50.9 GB of GPU memory in BF16 and 14.3 GB at 4 bits. See which GPUs fit it for inference and fine-tuning.

Codestral 22B needs about 50.9 GB of VRAM in FP16, 26.5 GB in FP8 and 14.3 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 5080 16GB.

Codestral 22B is a 22.2B-parameter language model from Mistral AI, released in May 2024, with a 32,768-token context window and published under the Mistral AI Non-Production License. Its weights alone take 44.4 GB in BF16 and 11.1 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 Codestral 22B

Codestral 22B 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.

PrecisionWeightsKV cacheTotal for inference
BF16 / FP1644.4 GB1.9 GB50.9 GB
FP822.2 GB1.9 GB26.5 GB
INT411.1 GB1.9 GB14.3 GB

Codestral 22B VRAM by task

Fine-tuning rows assume a short sequence, batch 1 and gradient checkpointing; longer sequences add activation memory.

TaskGPU memory
Inference at the full 32,768-token context (FP8)32.7 GB
QLoRA (4-bit base)14.4 GB
LoRA (BF16 base)57.7 GB
Full fine-tuning (BF16, AdamW)390.7 GB

GPUs that fit Codestral 22B

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.

GPUMemoryBF16 inference (50.9 GB)FP8 inference (26.5 GB)INT4 inference (14.3 GB)QLoRA (14.4 GB)LoRA (57.7 GB)
NVIDIA GeForce RTX 5080 16GB16 GB4 GPUs2 GPUs1 GPU1 GPU4 GPUs
NVIDIA GeForce RTX 4090 24GB24 GB4 GPUs2 GPUs1 GPU1 GPU4 GPUs
NVIDIA L4 24GB24 GB4 GPUs2 GPUs1 GPU1 GPU4 GPUs
NVIDIA GeForce RTX 5090 32GB32 GB2 GPUs1 GPU1 GPU1 GPU2 GPUs
NVIDIA Tesla V100 32GB32 GB2 GPUs1 GPU1 GPU1 GPU2 GPUs
NVIDIA A100 PCIe 40GB40 GB2 GPUs1 GPU1 GPU1 GPU2 GPUs
NVIDIA L40S 48GB48 GB2 GPUs1 GPU1 GPU1 GPU2 GPUs
NVIDIA Quadro RTX 8000 48GB48 GB2 GPUs1 GPU1 GPU1 GPU2 GPUs
NVIDIA RTX 6000 Ada 48GB48 GB2 GPUs1 GPU1 GPU1 GPU2 GPUs
NVIDIA RTX A6000 48GB48 GB2 GPUs1 GPU1 GPU1 GPU2 GPUs
NVIDIA A100 PCIe 80GB80 GB1 GPU1 GPU1 GPU1 GPU1 GPU
NVIDIA A100 SXM 80GB80 GB1 GPU1 GPU1 GPU1 GPU1 GPU
NVIDIA H100 PCIe 80GB80 GB1 GPU1 GPU1 GPU1 GPU1 GPU
NVIDIA H100 SXM 80GB80 GB1 GPU1 GPU1 GPU1 GPU1 GPU
NVIDIA RTX PRO 6000 Blackwell 96GB96 GB1 GPU1 GPU1 GPU1 GPU1 GPU
NVIDIA H200 141GB141 GB1 GPU1 GPU1 GPU1 GPU1 GPU
NVIDIA B200 180GB180 GB1 GPU1 GPU1 GPU1 GPU1 GPU
NVIDIA B300 288GB288 GB1 GPU1 GPU1 GPU1 GPU1 GPU
NVIDIA GB300 NVL288 GB1 GPU1 GPU1 GPU1 GPU1 GPU

What is Codestral 22B?

Codestral 22B is Mistral AI's first code model, trained on more than 80 programming languages including Python, Java, C, C++, JavaScript and Bash. It answers questions about code and fills in the middle of existing code, which suits completion in code editors.

Developer
Mistral AI
Released
May 2024
Parameters
22.2B
Context window
32,768 tokens
License
Mistral AI Non-Production License

The Mistral AI Non-Production License allows research and testing; production use needs a commercial license from Mistral AI.

Source: Codestral 22B v0.1 model card (as of November 17, 2026).

Frequently asked questions

How much VRAM does Codestral 22B need?
For inference with an 8,192-token context and one request, Codestral 22B needs about 50.9 GB in BF16, 26.5 GB in FP8, 14.3 GB in INT4, including a 10 % runtime margin. Real usage depends on the inference engine and its settings.
Can Codestral 22B run on a single GPU?
Yes. In BF16 it needs about 50.9 GB, which fits one NVIDIA A100 PCIe 80GB (80 GB).
How much GPU memory does fine-tuning Codestral 22B take?
QLoRA (4-bit base): about 14.4 GB; LoRA (BF16 base): about 57.7 GB; Full fine-tuning (BF16, AdamW): about 390.7 GB. Longer training sequences and larger batches add activation memory on top.
What is the best GPU for Codestral 22B?
Best for inference: 1 GPU × NVIDIA A100 PCIe 40GB (FP8 or INT8 weights, 26.5 GB at 8,192 tokens). Best for fine-tuning: 1 GPU × NVIDIA A100 PCIe 80GB (LoRA on BF16 weights, 57.7 GB). Budget pick: 1 GPU × NVIDIA GeForce RTX 5080 16GB (Smallest card class that holds a 4-bit build, 14.3 GB). Check live offers before booking, as availability changes.
Can Codestral 22B run on an RTX 4090?
Yes. In 4-bit it needs about 14.3 GB, which fits one RTX 4090 (24 GB). FP16 needs about 50.9 GB, 4 of these cards.
Can Codestral 22B run on an RTX 5090?
Yes. In FP8 it needs about 26.5 GB, which fits one RTX 5090 (32 GB). FP16 needs about 50.9 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.