NVIDIA L40 meets the minimum VRAM requirement for Q4 inference of moonshotai/Kimi-Linear-48B-A3B-Instruct. Review the quantization breakdown below to see how higher precision settings impact VRAM and throughput.
NVIDIA L40 can run moonshotai/Kimi-Linear-48B-A3B-Instruct with Q4 quantization. At approximately 63 tokens/second, you can expect Good speed - acceptable for interactive use.
You have 23GB headroom, which is sufficient for system overhead and smooth operation.
| Quantization | VRAM needed | VRAM available | Estimated speed | Verdict |
|---|---|---|---|---|
| Q4 | 25GB | 48GB | 62.58 tok/s | ✅ Fits comfortably |
| Q8 | 50GB | 48GB | 44.43 tok/s | ❌ Not recommended |
| FP16 | 101GB | 48GB | 20.92 tok/s | ❌ Not recommended |