How much VRAM does Nemotron 3 Super 120B A12B need?
Nemotron 3 Super 120B A12B needs 69.8 GB for its weights at 4-bit, 70.8 GB with an 8K context and 71.0 GB with 32K. At 8-bit, 123 GB.
- Parameters
- 124B
- Active per token
- 12B
- Longest context
- 256K
- Released
- Mar 2026
- License
- NVIDIA Nemotron Open Model
- Strongest at
- Chat, Coding, Agents, Docs
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| 16-bit | 230 GB | 231 GB | 231 GB | 232 GB |
| 8-bit | 122 GB | 123 GB | 124 GB | 124 GB |
| 6-bit | 94.4 GB | 95.4 GB | 95.6 GB | 96.4 GB |
| 5-bit | 81.9 GB | 82.9 GB | 83.1 GB | 83.9 GB |
| 4-bit | 69.8 GB | 70.8 GB | 71.0 GB | 71.8 GB |
| 3-bit | 56.3 GB | 57.3 GB | 57.5 GB | 58.2 GB |
| 2-bit | 43.2 GB | 44.2 GB | 44.4 GB | 45.2 GB |
| 1.6-bit | 23.7 GB | 24.8 GB | 25.0 GB | 25.7 GB |
One conversation, 16-bit context cache, runtime overhead included.
What runs it
| Machine | Best fit | Context | Speed | First reply |
|---|---|---|---|---|
| RTX 3060 12 GB | Too big | |||
| RTX 4060 Ti 16 GB | Too big | |||
| RTX 3090 | Too big | |||
| RTX 4090 | Too big | |||
| RTX 5090 | Too big | |||
| 2× RTX 3090 | 2-bit | 96K | 81 tok/s | 1.3 s |
| RTX PRO 6000 Blackwell | 5-bit | 256K | 82 tok/s | 0.6 s |
| M4 Max 128 GB | 5-bit | 256K | 27 tok/s | 13 s |
| M5 Max 128 GB | 5-bit | 256K | 29 tok/s | 3.9 s |
| Ryzen AI Max+ 395 128 GB | 5-bit | 256K | 15 tok/s | 7.7 s |
| DGX Spark 128 GB | 6-bit | 256K | 17 tok/s | 2.7 s |
| M3 Ultra 256 GB | 8-bit | 256K | 27 tok/s | 7.5 s |
| M5 Ultra 512 GB | 8-bit | 256K | 35 tok/s | 2.3 s |
| H100 SXM | 4-bit | 256K | 118 tok/s | 0.5 s |
| DGX Station GB300 | 8-bit | 256K | 127 tok/s | 0.2 s |
| 8× H200 NVL | 8-bit | 256K | 110 tok/s | 0.1 s |
Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.
How to run it
- Ollama
ollama run nemotron-3-super:120b-a12b- llama.cpp
llama-server -hf unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF- vLLM
vllm serve nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16- MLX (Mac)
mlx_lm.chat --model mlx-community/NVIDIA-Nemotron-3-Super-120B-A12B-4bit
Questions
How much VRAM does Nemotron 3 Super 120B A12B need?
Nemotron 3 Super 120B A12B needs 69.8 GB for its weights at 4-bit, 70.8 GB with an 8K context and 71.0 GB with 32K. At 8-bit, 123 GB.
Can Nemotron 3 Super 120B A12B run on an RTX 4090?
Not on one card. The smallest machine here that runs it is the 2× RTX 3090.