text-gen 120B params
nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16
nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16
736,923 downloads
344 likes
Hardware Recommendations
Three tiers based on VRAM requirements: minimum for basic inference, comfortable for longer contexts, headroom for future model sizes.
Minimum $7,999
Apple Mac Studio (M4 Max, 128GB)
128GB unified 85W TDP
Runs 120B at Q4_K_M quantization with 4K context.
Comfortable $7,999
Apple Mac Studio (M4 Max, 128GB)
128GB unified 85W TDP
Runs 120B at Q5_K_M quantization with 8K+ context.
VRAM Requirements
| Quantization | VRAM | Use case |
|---|---|---|
| Q4_K_M | 74 GB | Basic inference, 4K context |
| Q5_K_M | 92 GB | Good quality, 8K+ context |
| Q8 | 146 GB | Near-lossless, 16K+ context |
| FP16 | 290 GB | Full precision, max quality |
Formula: params × bytes_per_param × 1.2 overhead + 2 GB base. Actual requirements vary by framework and context length.