Run Nemotron 3.5 Lightning 30B-A3B locally: memory, speed and cost
Nemotron 3.5 Lightning 30B-A3B has 30 billion parameters, 3 billion active per token (mixture of experts). At Q4_K_M it needs about 20 GB with 32K tokens of context, so the cheapest hardware here that runs it is the Intel Arc Pro B70 (about 114 tokens a second). Through an API it costs $0.025 / $0.12 per 1M tokens.
Memory needed
Weights plus context cache
| Quantisation | 8K context | 32K context | 128K context |
|---|---|---|---|
| Q4_K_Musual choice | 20 GB | 20 GB | 21 GB |
| Q5_K_M | 23 GB | 23 GB | 24 GB |
| Q8_0 | 34 GB | 34 GB | 35 GB |
| BF16 | 62 GB | 62 GB | 63 GB |
Estimates: weights at the stated bits per weight, plus the context (KV) cache, plus about 1 GB of overhead. Lower-bit quantisations save memory at some cost in quality.
Nemotron 3.5 Lightning 30B-A3B on each device
Q4_K_M, 32K context
| Hardware | Memory | Fits? | Speed (estimate) |
|---|---|---|---|
| RTX 5090 | 32 GB | Fits | 240 tok/s |
| RTX 5080 | 16 GB | Partly, with system RAM | 70 tok/s |
| RTX 5070 Ti | 16 GB | Partly, with system RAM | 70 tok/s |
| RTX 5060 Ti 16 GB | 16 GB | Partly, with system RAM | 61 tok/s |
| RTX 4090 (used) | 24 GB | Fits | 191 tok/s |
| RTX 3090 (used) | 24 GB | Fits | 184 tok/s |
| RTX 3060 12 GB | 12 GB | Partly, with system RAM | 44 tok/s |
| RTX PRO 6000 Blackwell | 96 GB | Fits | 240 tok/s |
| Radeon RX 9070 XT | 16 GB | Partly, with system RAM | 66 tok/s |
| Radeon RX 7900 XTX | 24 GB | Fits | 187 tok/s |
| Radeon AI PRO R9700 | 32 GB | Fits | 164 tok/s |
| Intel Arc Pro B70 | 32 GB | Fits | 114 tok/s |
| Mac Studio M5 Max 128 GB | 128 GB | Fits | 143 tok/s |
| Mac Studio M5 Ultra 512 GB | 512 GB | Fits | 162 tok/s |
| Ryzen AI Max+ 395 128 GB | 128 GB | Fits | 76 tok/s |
| NVIDIA DGX Spark | 128 GB | Fits | 80 tok/s |
Generation speed estimated from memory bandwidth, calibrated against published llama.cpp results. Fastest here: RTX 5090, about 240 tokens a second. Try other settings in the calculator.
How capable is it?
Epoch AI hasn’t scored Nemotron 3.5 Lightning 30B-A3B yet, so we don’t show a capability figure.
Or use it through an API
$0.025 / $0.12 per 1M input / output tokens through OpenRouter (Nemotron 3.5 Lightning hosted prices). A customer chat assistant of 100M input and 20M output tokens a month costs about $4.90, so local hardware pays for itself only with heavy use.
Questions
How much VRAM does Nemotron 3.5 Lightning 30B-A3B need?
About 20 GB at Q4_K_M with 8K tokens of context, 20 GB with 32K and 21 GB with 128K. At 8-bit (Q8_0) it needs about 34 GB with 8K context.
Can Nemotron 3.5 Lightning 30B-A3B run on a Mac?
Yes. A Mac Studio M5 Max with 128GB runs it with 32K context at an estimated 143 tokens a second.
How fast is Nemotron 3.5 Lightning 30B-A3B on an RTX 4090?
About 191 tokens a second with 32K context, an estimate from the card’s memory bandwidth.
What does Nemotron 3.5 Lightning 30B-A3B cost through an API?
$0.025 / $0.12 per 1M input / output tokens through OpenRouter, the cheapest host we track. A customer chat assistant of 100M input and 20M output tokens a month would cost about $4.90.