Can NVIDIA GeForce RTX 5090 run Codestral 22B?

22B parameter Code model on 32GB GDDR7

Yes — runs at 8-bit quantization
~50-62 tok/sFast
SpeedFast inference, near-native speed
QualityNear-lossless — virtually identical to FP16

VRAM Requirements

Codestral 22B is a 22B parameter model. At full precision (FP16), it requires 44GB of VRAM. Your NVIDIA GeForce RTX 5090 has 32GB, so you'll need to quantize it to 8-bit (Q8) to fit.

FP16 (Full Precision)44GB (need 12GB more)

Maximum quality, no quantization

Q8 (8-bit)22GB (10GB free)

Near-lossless, ~50% size reduction

Q4 (4-bit)13GB (19GB free)

Good quality, ~75% size reduction

Your GPU VRAM: 32GB GDDR7 at 1792 GB/s bandwidth
Recommended system RAM: 64GB DDR5 (2x GPU VRAM minimum for model overflow)

What This Means in Practice

Codestral 22B at this precision on NVIDIA GeForce RTX 5090 gives excellent code completion and generation. Fast enough for real-time IDE integration. Handles complex refactoring, multi-file edits, and long-context code understanding.

How to Set It Up

Step 1: Install Ollama

curl -fsSL https://ollama.com/install.sh | sh

Ollama is the easiest way to run local LLMs. Works on Linux, macOS, and Windows.

Step 2: Download and run Codestral 22B

ollama run codestral:22b

This downloads the model (~22GB). First run takes a few minutes.

Step 3: Verify GPU is being used

nvidia-smi

Check that VRAM usage increases when the model loads. You should see ~22GB used.

NVIDIA GeForce RTX 5090 Specs

VRAM32GB GDDR7
Memory Bandwidth1792 GB/s
TDP575W
CUDA Cores21,760
Street Price~$2800
AI Rating10/10

Other Code Models on NVIDIA GeForce RTX 5090

About Codestral 22B

Top code completion model. Q4 fits on 16GB GPUs.

Category: Code · Parameters: 22B · CUDA required: No (runs via llama.cpp/GGUF)