Nemotron 3 Ultra 550B — VRAM, speed & local setup
Written by Jakub Rusinowski · Last updated
550B parameters, 55B active per token (512 latent experts, 22 routed), in a 108-layer stack where only 12 layers are attention, so the KV cache stays small. Includes a multi-token-prediction module for faster generation. Text only, 262,144-token context, OpenMDW-1.1. NVIDIA documents vLLM and SGLang serving on multi-GPU nodes.
Nemotron 3 Ultra 550B needs about 339 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
VRAM and speed by quantization
Quoted against NVIDIA RTX 4090 (24 GB). Includes the KV cache at 8K context, so it reads higher than the headline figure.
| Quant | Memory | VRAM | Speed (est.) | Fit |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 185.2 GB | — | Too big | |
| Q3_K_M 3.41 bpw | 239.8 GB | — | Too big | |
| Q4_K_M 4.83 bpw | 339.3 GB | — | Too big | |
| Q5_K_M 5.67 bpw | 398.2 GB | — | Too big | |
| Q6_K 6.56 bpw | 460.5 GB | — | Too big | |
| Q8_0 8.50 bpw | 596.4 GB | — | Too big | |
| F16 16.00 bpw | 1121.9 GB | — | Too big |
Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. Nemotron 3 Ultra 550B VRAM calculator →
Get Nemotron 3 Ultra 550B running
The cheapest catalogued GPU that runs Nemotron 3 Ultra 550B is the Apple M3 Ultra (512 GB).
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
How to run Nemotron 3 Ultra 550B
No local Ollama tag. Ollama lists nemotron-3-ultra only as a cloud-hosted ':cloud' tag, which runs on Ollama's servers, not your machine.
Specifications
Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.
- Parameters
- 550 Billion (55B active)
- Context window
- 262,144
- Architecture
- Hybrid Mamba-2 + LatentMoE + attention, with MTP
- Provider
- NVIDIA
- Licence
- OpenMDW-1.1
- Availability
- Weights downloadable
- Practical deployment
- Workstation (large unified memory or pro GPU)
- Specified at
- Q4_K_M
- System RAM
- 512 GB
- Record updated
- 2026-10-13
Commercial use permitted. No usage restrictions beyond attribution.
Quality and use cases
Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.
Independent evaluation — not vendor-claimed and not measured by us
| Benchmark | Score | Provenance |
|---|---|---|
| AA Intelligence Index v4.3.2 (Reasoning) | 23 | independent · Artificial Analysis · snapshot 2026-10-13 |
Nemotron 3 Ultra 550B — frequently asked questions
How much VRAM does Nemotron 3 Ultra 550B need?
About 339 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does Nemotron 3 Ultra 550B run on an RTX 4090 (24 GB)?
No. Nemotron 3 Ultra 550B needs about 339 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.
How do I run Nemotron 3 Ultra 550B locally?
No local Ollama tag. Ollama lists nemotron-3-ultra only as a cloud-hosted ':cloud' tag, which runs on Ollama's servers, not your machine. Running the published tag would send your prompts to a hosted GPU rather than your own machine.