Can I run Nemotron 3 Nano 30B-A3B on an RTX 5090?
Yes: Nemotron 3 Nano 30B-A3B needs about 20.3 GB at Q4_K_M with 32K tokens of context, which fits the 32 GB RTX 5090 with 11.7 GB to spare. At 32K the RTX 5090 holds up to Q6_K (27.2 GB), and Q4_K_M runs up to 256K (full) tokens.
Yes Q4_K_M with 32K tokens of context
- Q4_K_M, 32K
- 20.3 GB
- RTX 5090
- 32 GB, 1,792 GB/s
- To spare
- 11.7 GB
- Tokens/s
- 172–324 tokens/s
At Q4_K_M with 32K tokens of context it writes about 172–324 tokens/s for one request on an RTX 5090.
Best precision for Nemotron 3 Nano 30B-A3B on an RTX 5090
The most precise setting that leaves at least 0.5 GB free; one that fits with less is marked tight.
| Context | Best fit | Memory | Free | Tokens/s |
|---|---|---|---|---|
| 8K | Q6_K | 27.1 GB | 4.92 GB | 144–266 |
| 32K | Q6_K | 27.2 GB | 4.77 GB | 139–255 |
| 128K | Q6_K | 27.9 GB | 4.15 GB | 120–217 |
| 256K (full) | Q6_K | 28.7 GB | 3.32 GB | 102–182 |
Nemotron 3 Nano 30B-A3B on the RTX 5090 as the context fills
One request, FP16 KV cache, 0.5 GB plus 10% overhead; a minus sign is memory missing, and tight is less than 0.5 GB free.
| Context | Q4_K_M | Free | Tokens/s | Q8_0 | Free | Tokens/s |
|---|---|---|---|---|---|---|
| 4K | 20.1 GB | 11.9 GB | 182–346 | 34.9 GB | −2.90 GB | — |
| 8K | 20.1 GB | 11.9 GB | 181–343 | 34.9 GB | −2.92 GB | — |
| 16K | 20.2 GB | 11.8 GB | 178–336 | 35.0 GB | −2.98 GB | — |
| 32K | 20.3 GB | 11.7 GB | 172–324 | 35.1 GB | −3.08 GB | — |
| 64K | 20.5 GB | 11.5 GB | 162–302 | 35.3 GB | −3.28 GB | — |
| 128K | 20.9 GB | 11.1 GB | 144–266 | 35.7 GB | −3.70 GB | — |
| 256K (full) | 21.7 GB | 10.3 GB | 119–215 | 36.5 GB | −4.52 GB | — |
--n-cpu-moe for Nemotron 3 Nano 30B-A3B on the RTX 5090
The measured Q4_K_M GGUF fits the RTX 5090 whole at 32K (23.8 GB), so --n-cpu-moe is not needed there. Measured Q4_K_M file, 1 GB of buffers; tokens/s by system RAM speed.
Nemotron 3 Nano 30B-A3B on more than one RTX 5090
| Cards | Best at 32K | Tokens/s | Q4_K_M longest | Q4_K_M tokens/s | Rent per hour |
|---|---|---|---|---|---|
| 2× (64 GB) | Q8_0 | 140–287 | 256K (full) | 177–386 | $1.38 |
| 4× (128 GB) | BF16 | 146–302 | 256K (full) | 218–514 | $2.76 |
Tensor parallel with the combined bandwidth, as in vLLM; llama.cpp splits layers by default and runs at about one card's speed.
Other options
- The next smaller setting, Q3_K_M, takes 16.5 GB at 32K, 15.5 GB under the RTX 5090; it fits with 0.5 GB to spare up to 256K (full) tokens.
- Renting an RTX 5090 costs about $0.69 an hour (median on getdeploying.com, 2026-09-29): $0.59–1.1 per million tokens at 172–324 tokens/s.
- Smallest setup for Q4_K_M at 32K: RTX 3090 (24 GB).
Run Nemotron 3 Nano 30B-A3B on the RTX 5090 with llama-server
llama-server -hf ggml-org/NVIDIA-Nemotron-3-Nano-30B-A3B-GGUF:Q4_K_M -c 262144NVIDIA-Nemotron-3-Nano-30B-A3B-Q4_K_M.gguf, 22.4 GB, from ggml-org/
Questions
Can I run Nemotron 3 Nano 30B-A3B on an RTX 5090?
Yes: Nemotron 3 Nano 30B-A3B needs about 20.3 GB at Q4_K_M with 32K tokens of context, which fits the 32 GB RTX 5090 with 11.7 GB to spare. At 32K the RTX 5090 holds up to Q6_K (27.2 GB), and Q4_K_M runs up to 256K (full) tokens.
How fast is Nemotron 3 Nano 30B-A3B on an RTX 5090?
At Q4_K_M with 32K tokens of context it writes about 172–324 tokens/s for one request on an RTX 5090.
Does Nemotron 3 Nano 30B-A3B need --n-cpu-moe on an RTX 5090?
The measured Q4_K_M GGUF fits the RTX 5090 whole at 32K (23.8 GB), so --n-cpu-moe is not needed there.
What does a second RTX 5090 change for Nemotron 3 Nano 30B-A3B?
Two RTX 5090 cards (64 GB in one tensor-parallel group) hold Nemotron 3 Nano 30B-A3B at Q8_0 with 32K, and Q4_K_M up to 256K (full) tokens, at about 177–386 tokens/s.
Try other settings in the VRAM calculator, the speed calculator or the MoE offload planner. See also Nemotron 3 Nano 30B-A3B VRAM requirements, what LLMs an RTX 5090 can run and every pair, or detect your own GPU. Model data checked .