Nemotron 3 Ultra (BF16)
By NVIDIA · United States
Updated 2026-08-28
Overview
NVIDIA's aligned Nemotron 3 Ultra: a 561B MoE model with ~55B active parameters, 128K context, and multilingual reasoning and code capability. Datacenter-only hardware requirements.
When to pick this model
- Frontier-scale multilingual reasoning or coding workloads on datacenter infrastructure
- You need a chat-ready MoE model rather than a base checkpoint
- Multi-GPU H100/MI300 clusters are available for deployment
- Commercial use under NVIDIA's Open Model License
VRAM requirements by quantization
| Quantization | VRAM required |
|---|---|
| Q4_K_M (recommended) | 325 GB |
| Q5_K_M | 398 GB |
| Q8_0 | 600 GB |
| FP16 (no quantization) | 1122 GB |
VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.
In practice, Nemotron 3 Ultra (BF16) is server-class even at Q4_K_M (325 GB). Stepping up to Q8_0 nearly doubles the footprint to 600 GB, and unquantized FP16 weights take 1122 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.
Without a GPU, Nemotron 3 Ultra (BF16) needs roughly 729 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 1.5 tokens/sec on entry-level GPUs, on the order of 2.5 tokens/sec on a mid-range card, and up to 5 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.
What hardware do you need
The table below matches Nemotron 3 Ultra (BF16) to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.
| GPU memory | Example cards | Best fit for Nemotron 3 Ultra (BF16) |
|---|---|---|
| 8 GB | RTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GB | Does not fit — needs 325 GB at Q4_K_M |
| 12 GB | RTX 5070, RTX 5070 Ti Laptop, RTX 4080 Laptop | Does not fit — needs 325 GB at Q4_K_M |
| 16 GB | RTX 5080, RTX 4080 Super, Radeon RX 9070 XT | Does not fit — needs 325 GB at Q4_K_M |
| 24 GB | RTX 4090, Radeon RX 7900 XTX, RTX 5090 Laptop | Does not fit — needs 325 GB at Q4_K_M |
| 32 GB | RTX 5090 | Does not fit — needs 325 GB at Q4_K_M |
Which GPU should you buy to run Nemotron 3 Ultra (BF16)?
To run Nemotron 3 Ultra (BF16) locally at Q4, you need ~325 GB of VRAM. The best value for this is a Apple Mac Studio (64+ GB unified memory).
As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.
Strengths
- Frontier-scale MoE: 561B total / ~55B active parameters in BF16
- Multilingual coverage across 11 languages, including French
- 128K native context
- NVIDIA Open Model License permits commercial use
Limitations
- ~325GB VRAM at Q4, ~1.1TB in BF16 — requires multi-GPU H100/MI300-class hardware
- No Ollama tag; HuggingFace-only distribution
- License terms should be reviewed carefully for commercial deployment
Typical workloads
In our catalog grid, Nemotron 3 Ultra (BF16) is filed under Datacenter Reasoning, Multilingual Code, High-Capacity MoE — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: code generation and review (pair it with an editor integration like Continue.dev or Cline); multi-step reasoning and math-flavoured tasks; multilingual workloads.
The 125k-token context window covers long chats and mid-sized documents, though very large retrieval workloads will need chunking. It ships under the NVIDIA Open Model License license — commercial use is generally possible but read the specific terms before embedding it in a product.
Architecture & training
Architecture: Mixture-of-Experts · 561B total parameters · ~55B active per token · 128K context · BF16 weights
Training: BF16 checkpoint (post-Base variant) from the Nemotron 3 Ultra family. Multilingual pretraining covering 11 languages (en, fr, es, it, de, pt, ja, ko, hi, ar, zh).
A frontier-scale multilingual reasoning and coding MoE, but strictly a datacenter deployment.
Quick start
# HuggingFace : nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.
Similar models worth comparing
Frequently asked questions
How much VRAM does Nemotron 3 Ultra (BF16) need?
At the recommended Q4_K_M quantization, Nemotron 3 Ultra (BF16) needs about 325 GB of VRAM. Q8_0 takes 600 GB, and unquantized FP16 weights take 1122 GB.
Can Nemotron 3 Ultra (BF16) run without a GPU?
Yes — with roughly 729 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.
What context window does Nemotron 3 Ultra (BF16) support?
Nemotron 3 Ultra (BF16) supports a 125k-token context window (128,000 tokens).
Can I use Nemotron 3 Ultra (BF16) commercially?
Nemotron 3 Ultra (BF16) ships under the NVIDIA Open Model License license. Commercial use is generally permitted subject to its terms — review the license text before shipping a product.
How fast is Nemotron 3 Ultra (BF16) on consumer hardware?
Our compatibility engine estimates on the order of 2.5 tokens/sec on a mid-range GPU and up to 5 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.
Which quantization of Nemotron 3 Ultra (BF16) should I download first?
Start with Q4_K_M (325 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It does not fit a single 24 GB consumer card — plan for multi-GPU or server hardware.