NVIDIA-Nemotron-3-Nano-30B-A3B-FP8
by nvidia
31.6B params · text-generation · 295 likes · 1.4M downloads
NVIDIA-Nemotron-3-Nano-30B-A3B-FP8 is a 31.6B parameter model. At Q4 quantization it requires 16GB of VRAM. It runs comfortably on GeForce RTX 4090 (41 tok/s), GeForce RTX 5090 (62 tok/s), A100 PCIe 80 GB (76 tok/s).
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 41 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 62 tok/s | comfortable |
| M4 Max 128GB | 128GB | 15 tok/s | tight |
| M4 Pro 48GB | 48GB | 7 tok/s | tight |
| M4 Pro 24GB | 24GB | 7 tok/s | tight |
| A100 PCIe 80 GB | 80GB | 76 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 153 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 36 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 30 tok/s | comfortable |
| M4 Max 64GB | 64GB | 15 tok/s | tight |