GPUs/Intel Arc B580/Codestral 22B

Can Intel Arc B580 run Codestral 22B?

22B parameter Code model on 12GB GDDR6

Barely — requires CPU/RAM offloading
~1-3 tok/s (offload)
SpeedVery slow — expect 1-3 tokens/sec
QualityQuality is fine but the speed makes it impractical for interactive use
Intel GPUs lack CUDA. While Codestral 22B can technically run via llama.cpp/GGUF, the setup is more complex and less optimized than on NVIDIA hardware.

VRAM Requirements

Codestral 22B is a 22B parameter model. At full precision (FP16), it requires 44GB of VRAM. Your Intel Arc B580 only has 12GB — not enough even at maximum compression.

FP16 (Full Precision)44GB (need 32GB more)

Maximum quality, no quantization

Q8 (8-bit)22GB (need 10GB more)

Near-lossless, ~50% size reduction

Q4 (4-bit)13GB (need 1GB more)

Good quality, ~75% size reduction

Your GPU VRAM: 12GB GDDR6 at 456 GB/s bandwidth
Recommended system RAM: 32GB DDR5 (2x GPU VRAM minimum for model overflow)

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:q4_K_M

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

Step 3: Verify GPU is being used

rocm-smi

Check that VRAM usage increases when the model loads. You should see ~13GB used.

Intel Arc B580 Specs

VRAM12GB GDDR6
Memory Bandwidth456 GB/s
TDP150W
CUDA CoresN/A
Street Price~$230
AI Rating2/10

Other Code Models on Intel Arc B580

About Codestral 22B

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

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