Nemotron 3.5 Lightning 30B-A3B: VRAM requirements & speed
Nemotron 3.5 Lightning 30B-A3B (31.58B MoE, 3B active, openmdw-1.1). Native context 256k. Smallest GGUF: 25.27 GB.
Memory needed by quantization and context
GGUF file size plus f16 KV cache (and fixed overhead of ~1 GB per GPU not shown). Source: unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF.
| Quant | File | +8k ctx | +32k ctx | +128k ctx | +256k ctx |
|---|---|---|---|---|---|
| Q4_K_M UD-Q4_K_M | 25.27 GB | 25.4 GB | 25.5 GB | 26.1 GB | 26.9 GB |
| Q5_K_M UD-Q5_K_M | 30.22 GB | 30.3 GB | 30.5 GB | 31.1 GB | 31.9 GB |
| Q8_0 | 35 GB | 35.1 GB | 35.3 GB | 35.9 GB | 36.7 GB |
KV cache size by context length
Computed from the model's attention layout, including a fixed 0.05 GB linear-attention state.
| Context | f16 | q8_0 | q4_0 |
|---|---|---|---|
| 8k | 0.10 GB | 0.08 GB | 0.06 GB |
| 32k | 0.25 GB | 0.16 GB | 0.11 GB |
| 128k | 0.86 GB | 0.48 GB | 0.28 GB |
| 256k | 1.66 GB | 0.91 GB | 0.50 GB |
Which hardware runs Nemotron 3.5 Lightning 30B-A3B? · 16k context
| Hardware | Quant | Fit | Memory | Tok/s | Speed source |
|---|---|---|---|---|---|
| RTX 5090 32GB | UD-Q5_K_M | fits fully | 30.4 GB | 349 | measured |
| RTX PRO 6000 96GB | Q8_0 | fits fully | 35.2 GB | 291 | scaled from measured |
| Mac Studio M5 Ultra 256GB | Q8_0 | fits fully | 35.2 GB | 176 | estimated |
| Mac Studio M5 Ultra 512GB (ships late Oct 2026) | Q8_0 | fits fully | 35.2 GB | 176 | estimated |
| 2x RTX 3090 (48GB total) | Q8_0 | fits fully | 35.2 GB | 152 | estimated |
| RTX 4090 24GB | UD-Q4_K_M | MoE offload to CPU | 25.4 GB | 129 | estimated |
| RTX 3090 24GB | UD-Q4_K_M | MoE offload to CPU | 25.4 GB | 125 | estimated |
| AMD RX 7900 XTX 24GB | UD-Q4_K_M | MoE offload to CPU | 25.4 GB | 120 | estimated |
| Mac Studio (2025) M3 Ultra 96GB | Q8_0 | fits fully | 35.2 GB | 72 | scaled from measured |
| Mac Studio M5 Max 128GB | Q8_0 | fits fully | 35.2 GB | 68 | scaled from measured |
| MacBook Pro 14/16 M5 Max 128GB | Q8_0 | fits fully | 35.2 GB | 68 | scaled from measured |
| MacBook Pro 14/16 (2024) M4 Max 128GB | Q8_0 | fits fully | 35.2 GB | 61 | estimated |
| MacBook Pro 14/16 M5 Pro 48GB | UD-Q5_K_M | fits fully | 30.4 GB | 59 | estimated |
| AMD Ryzen AI Max+ PRO 495 (Ryzen AI Halo) 192GB | Q8_0 | fits fully | 35.2 GB | 52 | estimated |
| AMD Ryzen AI Max+ 395 (Strix Halo) 128GB | Q8_0 | fits fully | 35.2 GB | 50 | scaled from measured |
| RTX 5080 16GB | UD-Q4_K_M | MoE offload to CPU | 25.4 GB | 49 | estimated |
| RTX 5070 Ti 16GB | UD-Q4_K_M | MoE offload to CPU | 25.4 GB | 49 | estimated |
| NVIDIA DGX Spark 128GB | Q8_0 | fits fully | 35.2 GB | 47 | scaled from measured |
| RTX 5060 Ti 16GB | UD-Q4_K_M | MoE offload to CPU | 25.4 GB | 45 | estimated |
| 16GB VRAM class (RTX 4060 Ti 16GB, 32GB RAM) | UD-Q4_K_M | MoE offload to CPU | 25.4 GB | 42 | estimated |
| Mac mini (2024) M4 Pro 64GB | Q8_0 | fits fully | 35.2 GB | 37 | estimated |
| RTX 5070 12GB | UD-Q4_K_M | MoE offload to CPU | 25.4 GB | 36 | estimated |
| RTX 4070 12GB | UD-Q4_K_M | MoE offload to CPU | 25.4 GB | 36 | estimated |
| 8GB VRAM class (RTX 4060 8GB, 32GB RAM) | UD-Q4_K_M | MoE offload to CPU | 25.4 GB | 28 | estimated |
| RTX 3060 12GB (32GB RAM) | UD-Q4_K_M | MoE offload to CPU | 25.4 GB | 21 | estimated |
| Mac mini / iMac M4 16GB | - | does not fit | - | - | |
| Mac mini M6 32GB | - | does not fit | - | - | |
| MacBook Air M5 32GB | - | does not fit | - | - |
Discrete GPUs are modelled with 32 GB of system RAM available for offload. Does not fit even with offload: Mac mini / iMac M4 16GB, Mac mini M6 32GB, MacBook Air M5 32GB.
MLX conversions for Apple Silicon
No MLX conversion yet: we found no verified mlx-community 4-bit or 8-bit repo for Nemotron 3.5 Lightning 30B-A3B. Use the GGUF files with llama.cpp or ollama on a Mac. How we check.
Links
Model card · GGUF files · Open in the advisor · last verified 2026-10-20
FAQ
How much VRAM does Nemotron 3.5 Lightning 30B-A3B need?
The smallest quant (Q4_K_M) is 25.27 GB, so you need roughly 26 GB for an 8k context including KV cache and overhead. At 256k context the f16 KV cache alone is 1.7 GB.
Can I run Nemotron 3.5 Lightning 30B-A3B on a RTX 5090 32GB?
Yes: on RTX 5090 32GB it fits fully as UD-Q5_K_M at about 349 tok/s (measured). 14 of 28 hardware configurations fit it fully at 16k context.
What is the maximum context for Nemotron 3.5 Lightning 30B-A3B?
256k tokens natively. Memory, not the model, is usually the limit: see the KV cache table above.
All models: model index · by memory: 8 GB · 12 GB · 16 GB · 24 GB · 32 GB · 48 GB · 64 GB · 96 GB · 128 GB · 192 GB · 256 GB · 512 GB. Numbers use the same fit math as the advisor; sources and methods.