Nemotron 3.5 Lightning 30B-A3B
31.6B total, 3B active · 256k context · openmdw-1.1 · released 2026-08-01
chatcodingagenticreasoning
What to run it on in Canada
Fully in memory at 32k context, priced at retailers today.
Something we price holds it fully in memory, but none reaches 15 tok/s, our bar for recommending it. The table below has the speeds.
All hardware that runs it
Everything we price, best value first. Two cards split the model by layers.
| Hardware | Quant | Fits | Decode | Price | tok/s per 1,000 |
|---|---|---|---|---|---|
| Mac Studio M4 Max 128GB | UD-Q6_K_XL | fits | 60 tok/s | – | – |
| Mac mini M4 Pro 64GB | UD-Q6_K_XL | fits | 30 tok/s | – | – |
| Framework Desktop Ryzen AI Max+ 395 128GB | UD-Q6_K_XL | fits | 24 tok/s | – | – |
| Mac Studio M3 Ultra 512GB | UD-Q6_K_XL | fits | 89 tok/s | – | – |
| NVIDIA DGX Spark 128GB | UD-Q6_K_XL | fits | 28 tok/s | – | – |
| GMKtec EVO-X2 Ryzen AI Max+ 395 128GB | UD-Q6_K_XL | fits | 24 tok/s | – | – |
| Mac Studio M3 Ultra 256GB | UD-Q6_K_XL | fits | 89 tok/s | – | – |
| Mac mini M4 Pro 48GB | UD-Q6_K_XL | fits | 30 tok/s | – | – |
| Mac Studio M4 Max 64GB | UD-Q6_K_XL | fits | 60 tok/s | – | – |
| Mac Studio M3 Ultra 96GB | UD-Q6_K_XL | fits | 89 tok/s | – | – |
| NVIDIA RTX PRO 6000 Blackwell | UD-Q6_K_XL | fits | 187 tok/s | – | – |
| ASUS Ascent GX10 128GB | UD-Q6_K_XL | fits | 28 tok/s | – | – |
| Framework Desktop Ryzen AI Max+ 395 64GB | UD-Q6_K_XL | fits | 24 tok/s | – | – |
| NVIDIA GeForce RTX 3090 | UD-Q6_K_XL | offload | 24 tok/s | – | – |
| NVIDIA GeForce RTX 5090 | UD-Q6_K_XL | offload | 78 tok/s | – | – |
| NVIDIA GeForce RTX 4090 | UD-Q6_K_XL | offload | 24 tok/s | – | – |
| NVIDIA GeForce RTX 5070 Ti | UD-Q6_K_XL | offload | 15 tok/s | – | – |
| NVIDIA GeForce RTX 5060 Ti 16GB | UD-Q6_K_XL | offload | 14 tok/s | – | – |
| AMD Radeon RX 7900 XTX | UD-Q6_K_XL | offload | 23 tok/s | – | – |
| NVIDIA GeForce RTX 3060 12GB | UD-Q6_K_XL | offload | 12 tok/s | – | – |
| AMD Radeon RX 9070 XT | UD-Q6_K_XL | offload | 14 tok/s | – | – |
| NVIDIA GeForce RTX 4060 Ti 16GB | UD-Q6_K_XL | offload | 13 tok/s | – | – |
| NVIDIA GeForce RTX 5080 | UD-Q6_K_XL | offload | 15 tok/s | – | – |
| NVIDIA GeForce RTX 4070 Ti Super | UD-Q6_K_XL | offload | 15 tok/s | – | – |
| NVIDIA GeForce RTX 5070 | UD-Q6_K_XL | offload | 13 tok/s | – | – |
| AMD Radeon RX 9060 XT 16GB | UD-Q6_K_XL | offload | 13 tok/s | – | – |
| NVIDIA GeForce RTX 4070 Super | UD-Q6_K_XL | offload | 12 tok/s | – | – |
| NVIDIA GeForce RTX 4080 Super | UD-Q6_K_XL | offload | 15 tok/s | – | – |
| AMD Radeon AI PRO R9700 | UD-Q6_K_XL | offload | 41 tok/s | – | – |
| AMD Radeon RX 7800 XT | UD-Q6_K_XL | offload | 14 tok/s | – | – |
| AMD Radeon RX 7900 XT | UD-Q6_K_XL | offload | 18 tok/s | – | – |
| Intel Arc B580 | UD-Q6_K_XL | offload | 11 tok/s | – | – |
| NVIDIA GeForce RTX 3090 Ti | UD-Q6_K_XL | offload | 24 tok/s | – | – |
| AMD Radeon RX 7600 XT | UD-Q6_K_XL | offload | 13 tok/s | – | – |
| NVIDIA GeForce RTX 3080 10GB | UD-Q6_K_XL | offload | 12 tok/s | – | – |
| Intel Arc Pro B60 | UD-Q6_K_XL | offload | 17 tok/s | – | – |
| NVIDIA Tesla P40 | UD-Q6_K_XL | offload | 19 tok/s | – | – |
| Intel Arc A770 16GB | UD-Q6_K_XL | offload | 13 tok/s | – | – |
2 other cards and machines we price cannot hold it at this context.
Memory needed
Weights, KV cache and runtime buffers at each context length (KV cache at fp16).
| Quant | File | @8k | @32k | @64k | @128k |
|---|---|---|---|---|---|
| UD-Q6_K_XLgguf | 32.6 GB | 33.4 GB | 33.7 GB | 34.1 GB | 34.8 GB |
| Q8_0gguf | 32.6 GB | 33.4 GB | 33.7 GB | 34.1 GB | 34.8 GB |
Run it
Copy a command for the runtime you use.
- UD-Q6_K_XL
- Ollama
ollama run hf.co/unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF:UD-Q6_K_XL - LM Studio
lms get unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF@ud-q6_k_xl - llama.cpp
llama-server -hf unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF:UD-Q6_K_XL
- Ollama
- Q8_0
- Ollama
ollama run hf.co/unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF:Q8_0 - LM Studio
lms get unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF@q8_0 - llama.cpp
llama-server -hf unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF:Q8_0
- Ollama