NVIDIA Nemotron 3 Super 120B A12B BF16 heretic v2 needs about 80.9 GB of memory at MXFP4_MOE (a 68.6 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 2 quants, from 68.6 GB (MXFP4_MOE) to 241.5 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 47.2 GB.
NVIDIA Nemotron 3 Super 120B A12B BF16 heretic v2 – 120B parameter LLM with MoE
This model contains 121 billion parameters, with 12 billion active parameters in the mixture of experts. It supports reasoning and tool‑calling abilities. The model is available in two GGUF quantizations: MXFP4_MOE and BF16.
Good at
- Reasoning tasks
- Tool‑calling
- Large context handling
Limits
- Requires large VRAM – 80.9 GB for MXFP4_MOE at 8 k context
- BF16 quant needs 253.8 GB VRAM at 8 k context
- No vision support
Best for
- Complex reasoning workflows
- Automated tool integration
- Long‑form content generation
Who it’s for. Engineers and researchers needing a high‑capacity, expert‑mixing language model for advanced inference. Suitable for environments with ample GPU memory and specialized deployment pipelines.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| MXFP4_MOE | 68.6 GB | 80.9 GB | Too big | Get file → | |
| BF16 | 241.5 GB | 253.8 GB | Too big | Get file → |
For 8 GB VRAM use is not feasible. With 80 GB VRAM the MXFP4_MOE quant is recommended. If you have 250 GB VRAM, the BF16 quant can be used. Memory estimates are approximate.
What GPU do you need for NVIDIA Nemotron 3 Super 120B A12B BF16 heretic v2?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: no quant fits at an 8K context.
- 24 GB VRAM: no quant fits at an 8K context.
- 32 GB VRAM: no quant fits at an 8K context.
- 48 GB VRAM: no quant fits at an 8K context.
Which NVIDIA Nemotron 3 Super 120B A12B BF16 heretic v2 quant should you choose?
MXFP4_MOE (68.6 GB) is the middle option in this repository. Lower bit counts mean smaller files and more quality loss.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Frequently asked questions about NVIDIA Nemotron 3 Super 120B A12B BF16 heretic v2
How much VRAM does NVIDIA Nemotron 3 Super 120B A12B BF16 heretic v2 need?
About 80.9 GB at MXFP4_MOE with an 8K context window. That is the 68.6 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run NVIDIA Nemotron 3 Super 120B A12B BF16 heretic v2 on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 80.9 GB, so you would need a larger GPU or CPU offloading.
What is the file size of NVIDIA Nemotron 3 Super 120B A12B BF16 heretic v2 GGUF?
It depends on the quant: MXFP4_MOE is 68.6 GB, BF16 is 241.5 GB.
What context length does NVIDIA Nemotron 3 Super 120B A12B BF16 heretic v2 support?
The GGUF header lists a trained context length of 1,048,576 tokens (1024K). The KV cache grows with the context you use: about 11.8 GB at 8K and 47.2 GB at 32K.
Where can I download NVIDIA Nemotron 3 Super 120B A12B BF16 heretic v2 GGUF?
The files are on Hugging Face in the Sabomako/NVIDIA-Nemotron-3-Super-120B-A12B-BF16-heretic-v2-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does NVIDIA Nemotron 3 Super 120B A12B BF16 heretic v2 use?
Hugging Face lists the license as other. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run this model on 8 GB VRAM?
No, the smallest quant requires at least 80.9 GB VRAM at 8 k context.
Does the model support vision tasks?
No, the model has no vision capabilities.
What is the context length?
The model was trained with a context of 1,048,576 tokens.