NVIDIA L40S meets the minimum VRAM requirement for Q4 inference of RedHatAI/Llama-3.2-90B-Vision-Instruct-FP8-dynamic. Review the quantization breakdown below to see how higher precision settings impact VRAM and throughput.
NVIDIA L40S can run RedHatAI/Llama-3.2-90B-Vision-Instruct-FP8-dynamic with Q4 quantization. At approximately 30 tokens/second, you can expect Moderate speed - useful for batch processing.
You have 4GB headroom, which is sufficient for system overhead and smooth operation.
| Quantization | VRAM needed | VRAM available | Estimated speed | Verdict |
|---|---|---|---|---|
| Q4 | 44GB | 48GB | 30.00 tok/s | ✅ Fits comfortably |
| Q8 | 88GB | 48GB | 23.83 tok/s | ❌ Not recommended |
| FP16 | 176GB | 48GB | 11.71 tok/s | ❌ Not recommended |