NVIDIA-Nemotron-3-Super-120B-A12B GGUF size and VRAM requirements
unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF is a very large language model with 120.67 billion parameters, built on the nemotron_h_moe architecture. It is released under the other license and has been downloaded 11,731 times.
To run unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF locally at a 4,096-token context, its quantized versions need between 50.2 GB (IQ1_M, lowest quality) and 226.08 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included.
Available GGUF quantizations for unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF include IQ1_M, IQ2_XXS, IQ2_M, Q2_K_XL, IQ3_S, IQ3_XXS, Q3_K_M, Q3_K_S, Q3_K_XL, IQ4_NL, IQ4_XS, Q4_K_S, GGUF, Q4_K_M, Q4_K_XL, Q5_K_S, Q5_K_M, Q5_K_XL, Q6_K, Q6_K_XL, Q8_0, Q8_K_XL, BF16. The model supports a native context length of up to 1,048,576 tokens; a longer context grows the KV cache and the memory needed.
GGUF file size and memory by quantization
Compare real GGUF weight sizes, estimated KV cache and total memory for Q4, Q5, Q8 and every quantization published in this repository.
| Quant. | Bits | Quality | Weights | KV | Total | Speed~ | Verdict |
|---|---|---|---|---|---|---|---|
| IQ1_M | 3.49 | Fair | 49.06 GB | 0.34 GB | 50.2 GB | — | Insufficient |
| IQ2_XXS | 3.49 | Fair | 49.06 GB | 0.34 GB | 50.2 GB | — | Insufficient |
| IQ2_M | 3.5 | Fair | 49.12 GB | 0.34 GB | 50.27 GB | — | Insufficient |
| Q2_K_XL | 3.62 | Fair | 50.9 GB | 0.34 GB | 52.05 GB | — | Insufficient |
| IQ3_S | 3.75 | Fair | 52.74 GB | 0.34 GB | 53.89 GB | — | Insufficient |
| IQ3_XXS | 3.75 | Fair | 52.74 GB | 0.34 GB | 53.89 GB | — | Insufficient |
| Q3_K_M | 4.09 | Fair | 57.47 GB | 0.34 GB | 58.62 GB | — | Insufficient |
| Q3_K_S | 4.09 | Fair | 57.47 GB | 0.34 GB | 58.62 GB | — | Insufficient |
| Q3_K_XL | 4.15 | Fair | 58.33 GB | 0.34 GB | 59.47 GB | — | Insufficient |
| IQ4_NL | 4.28 | Good | 60.06 GB | 0.34 GB | 61.2 GB | — | Insufficient |
| IQ4_XS | 4.28 | Good | 60.06 GB | 0.34 GB | 61.2 GB | — | Insufficient |
| Q4_K_S | 5.24 | Very good | 73.59 GB | 0.34 GB | 74.73 GB | — | Insufficient |
| GGUF | 5.44 | Very good | 76.42 GB | 0.34 GB | 77.57 GB | — | Insufficient |
| Q4_K_M | 5.47 | Very good | 76.87 GB | 0.34 GB | 78.02 GB | — | Insufficient |
| Q4_K_XL | 5.55 | Very good | 78.02 GB | 0.34 GB | 79.17 GB | — | Insufficient |
| Q5_K_S | 5.95 | Very good | 83.56 GB | 0.34 GB | 84.7 GB | — | Insufficient |
| Q5_K_M | 7.12 | Excellent | 99.96 GB | 0.34 GB | 101.11 GB | — | Insufficient |
| Q5_K_XL | 7.13 | Excellent | 100.18 GB | 0.34 GB | 101.33 GB | — | Insufficient |
| Q6_K | 7.61 | Excellent | 106.87 GB | 0.34 GB | 108.01 GB | — | Insufficient |
| Q6_K_XL | 7.81 | Excellent | 109.76 GB | 0.34 GB | 110.9 GB | — | Insufficient |
| Q8_0 | 8.52 | Excellent | 119.65 GB | 0.34 GB | 120.79 GB | — | Insufficient |
| Q8_K_XL | 8.78 | Excellent | 123.39 GB | 0.34 GB | 124.53 GB | — | Insufficient |
| BF16 | 16.01 | Excellent | 224.93 GB | 0.34 GB | 226.08 GB | — | Insufficient |
KV cache computed from the model's exact architecture. Speed is a rough estimate bounded by memory bandwidth.
Frequently asked questions
What kind of model is unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF?
unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF is a language model with 120.67 billion parameters, based on the nemotron_h_moe architecture. It is released under the other license and distributed as GGUF files for local inference.
Can I run unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF on an 8 GB GPU?
No. unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF does not fit on an 8 GB GPU, even with the smallest quantization and system RAM offloading.
Can I run unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF on a 16 GB GPU?
No. unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF does not fit on a 16 GB GPU, even with the smallest quantization and system RAM offloading.
Can I run unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF on a 24 GB GPU?
Partially. unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF only fits on a 24 GB GPU by offloading part of it to system RAM (with IQ4_NL), which runs but is slower.
What context length does unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF support?
unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF supports a native context length of up to 1,048,576 tokens. A longer context grows the KV cache, so it increases the memory needed to run the model.
What is the best quantization for unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF?
For unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF, a strong default is Q4_K_M, which needs about 78.02 GB and keeps most of the quality while roughly halving the memory versus 8-bit. With VRAM to spare, Q5_K_M or Q6_K add a little more quality; if you are tight on memory, a smaller quantization still runs. Pick the highest quantization that fits your VRAM.