Run NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 locally: hardware requirements
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 needs at least 1068 GB of GPU memory for the model alone. Its repository does not publish what each extra person adds, so plan for more.
NVIDIA, United States. From its files on Hugging Face, checked 2026-11-05.
Context per person: more context lets the model work with longer conversations and more files, and needs more GPU memory for every person. As a guide: 8K for short chat, 32K for chat over documents, 64K for AI agents, whose instructions, tools and files all count.
80 users means about 16 people working at the same moment. That share is a planning assumption; if you know yours, choose the total that matches it.
Work it out for your team: nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16
On Hugging FaceNVIDIA, United States · 561B parameters · License: other
What you need for 16 active users
At least 7 × B200
That holds the model alone. What each person adds is not public, so plan for more.
The server
3 TB RAM
64 CPU cores
4 TB NVMe
Rules of thumb
- GPU memory in total
- at least 1068 GB
- Weights
- 1121 GB (published files)
- Conversation cache, all active
- Not published
- Engine (vLLM)
- 3.0 GB
Ways to need less
- Use a 4-bit build2 × B200
- Use a smaller model: NVIDIA-Nemotron-3-Nano-30B-A3B-BF161 × H200 NVL
- Use a smaller model: NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF161 × H200 NVL
Your options by card
80 users means about 16 using the model at any moment. The others are not waiting; they are simply doing something else. If more become active than a build handles, they queue. 15 cards too small for this model are left out.
| Card | Memory | Active users, one server | For your 16 active |
|---|---|---|---|
| B200 | 180 GB | Not public for this model | At least 7 × B200 |
| H200 NVL | 141 GB | Not public for this model | At least 8 × H200 NVL |
These are memory limits. A card can hold more people than it can answer quickly, so test your own workload before you buy. More than one card in a server needs an engine that splits the model.
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 by team size
AI agents, 1 user in 5 active at a time, 64K context, as published.
| Total users | At once | GPU memory | What you need |
|---|---|---|---|
| 20 | 4 | at least 1068 GB | At least 7 × B200 |
| 80 | 16 | at least 1068 GB | At least 7 × B200 |
| 120 | 24 | at least 1068 GB | At least 7 × B200 |
| 500 | 100 | at least 1068 GB | At least 7 × B200 |
Similar models
- NVIDIA-Nemotron-3-Super-120B-A12B-BF16NVIDIA · 124B · 326 GB for 16 at once
- NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4NVIDIA · 66B · at least 74.2 GB for 16 at once
- NVIDIA-Nemotron-3-Nano-30B-A3B-BF16NVIDIA · 32B · 115 GB for 16 at once
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16NVIDIA · 32B · 118 GB for 16 at once
- NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16NVIDIA · 4.0B · 299 GB for 16 at once
- NVIDIA-Nemotron-3-Nano-4B-BF16NVIDIA · 4.0B · 179 GB for 16 at once
How this is worked out
GPU memory = weights + a conversation cache for each person + the serving engine, from this model's files on Hugging Face. These are memory limits, not speed. The full method and its limits.
Questions
How much GPU memory does NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 need?
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 needs at least 1068 GB of GPU memory for the model alone. Its repository does not publish what each extra person adds, so plan for more.
Can NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 run on a 24 GB GPU such as the RTX 4090?
Not as published for the model alone: it needs at least 1068 GB. A 4-bit or 8-bit build may fit, if the model has one.
What license does NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 use?
Hugging Face lists its license as other. Check the license on the model's page before you use it.
Running a model on your own servers?
NeueCode is private autonomous intelligence for enterprise: it takes on your organization's tasks on your own GPUs, with nothing leaving your network, a person approving every action, and all of it on record. It runs each model on one card, so read the single-card figures above.
What you need to run NeueCode