NVIDIA-Nemotron-3.5-Lightning-30B-A3B GGUF size and VRAM requirements
unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF is a very large language model with 31.58 billion parameters, built on the nemotron_h_moe architecture. It is released under the other license and has been downloaded 56,457 times.
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF is a Mixture-of-Experts model with 128 experts, of which 6 are active on each token. Routing only 6 of 128 experts makes it noticeably faster than a dense model of the same size — but every expert still has to be held in memory, so the numbers below are set by the full parameter count, not the active one. It is built from 52 transformer layers, a hidden size of 2,688, 32 attention heads. It uses grouped-query attention (32 query heads sharing 2 key/value heads), which already trims KV-cache memory compared with full multi-head attention.
To run unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF locally at a 4,096-token context, its quantized versions need between 19.1 GB (IQ1_M, lowest quality) and 62.33 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included. Context length drives that memory directly: at 4,096 tokens the KV cache for NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF is about 0.2 GB, rising to roughly 52.0 GB at its full 1,048,576-token context. Shorter prompts free up memory for a higher-quality quantization.
For most users the best balance is Q4_K_S, needing about 23.79 GB. That means unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF fits entirely in the VRAM of a 24 GB GPU or larger, running fully on the GPU.
Available GGUF quantizations for unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF include IQ1_M, IQ2_XXS, IQ2_M, IQ3_XXS, IQ3_S, IQ4_NL, Q3_K_XL, MXFP4, Q4_K_S, Q4_K_M, Q4_K_XL, Q5_K_S, Q5_K_M, Q5_K_XL, Q8_0, Q6_K_XL, 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 | 4.92 | Good | 18.09 GB | 0.2 GB | 19.1 GB | 2.8 t/s | Offload |
| IQ2_XXS | 4.92 | Good | 18.09 GB | 0.2 GB | 19.1 GB | 2.8 t/s | Offload |
| IQ2_M | 4.92 | Good | 18.1 GB | 0.2 GB | 19.1 GB | 2.8 t/s | Offload |
| IQ3_XXS | 5.01 | Very good | 18.4 GB | 0.2 GB | 19.4 GB | 2.7 t/s | Offload |
| IQ3_S | 5.36 | Very good | 19.7 GB | 0.2 GB | 20.71 GB | 2.5 t/s | Offload |
| IQ4_NL | 5.38 | Very good | 19.78 GB | 0.2 GB | 20.78 GB | 2.5 t/s | Offload |
| Q3_K_XL | 5.38 | Very good | 19.78 GB | 0.2 GB | 20.78 GB | 2.5 t/s | Offload |
| MXFP4 | 5.88 | Very good | 21.62 GB | 0.2 GB | 22.62 GB | 2.3 t/s | Offload |
| Q4_K_S | 6.2 | Very good | 22.79 GB | 0.2 GB | 23.79 GB | 2.2 t/s | Offload |
| Q4_K_M | 6.4 | Very good | 23.53 GB | 0.2 GB | 24.53 GB | — | Insufficient |
| Q4_K_XL | 6.46 | Very good | 23.75 GB | 0.2 GB | 24.76 GB | — | Insufficient |
| Q5_K_S | 6.64 | Excellent | 24.42 GB | 0.2 GB | 25.43 GB | — | Insufficient |
| Q5_K_M | 7.65 | Excellent | 28.14 GB | 0.2 GB | 29.14 GB | — | Insufficient |
| Q5_K_XL | 7.71 | Excellent | 28.33 GB | 0.2 GB | 29.33 GB | — | Insufficient |
| Q8_0 | 8.87 | Excellent | 32.6 GB | 0.2 GB | 33.6 GB | — | Insufficient |
| Q6_K_XL | 8.87 | Excellent | 32.6 GB | 0.2 GB | 33.6 GB | — | Insufficient |
| Q8_K_XL | 9.78 | Excellent | 35.96 GB | 0.2 GB | 36.97 GB | — | Insufficient |
| BF16 | 16.68 | Excellent | 61.33 GB | 0.2 GB | 62.33 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-3.5-Lightning-30B-A3B-GGUF?
unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF is a language model with 31.58 billion parameters, based on the nemotron_h_moe architecture. It is released under the other license and distributed as GGUF files for local inference.
Is unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF a Mixture-of-Experts (MoE) model?
Yes. unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF is a Mixture-of-Experts model with 128 experts, of which 6 are activated per token. That makes it faster than a dense model of the same size, but all 128 experts must be loaded into memory, so the VRAM/RAM it needs is driven by the total parameter count, not the active one.
How much VRAM do you need to run unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF?
You need about 23.79 GB of VRAM to run unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF entirely on the GPU using the Q4_K_S quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.
Can I run unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF on an 8 GB GPU?
Partially. unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF only fits on an 8 GB GPU by offloading part of it to system RAM (with Q4_K_S), which runs but is slower.
Can I run unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF on a 16 GB GPU?
Partially. unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF only fits on a 16 GB GPU by offloading part of it to system RAM (with Q8_K_XL), which runs but is slower.
Can I run unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF on a 24 GB GPU?
Yes. With 24 GB of VRAM you can run unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF fully on the GPU using Q4_K_S (about 23.79 GB).
What context length does unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF support?
unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-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-3.5-Lightning-30B-A3B-GGUF?
For unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF, a strong default is Q4_K_M, which needs about 24.53 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.