How much VRAM does Nemotron 3 Ultra 550B A55B need?
Nemotron 3 Ultra 550B A55B needs 316 GB for its weights at 4-bit, 318 GB with an 8K context and 318 GB with 32K. At 8-bit, 556 GB.
- Parameters
- 561B
- Active per token
- 55B
- Longest context
- 256K
- Released
- Jun 2026
- License
- OpenMDW 1.1
- Strongest at
- Chat, Coding, Docs
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| 16-bit | 1,044 GB | 1,045 GB | 1,046 GB | 1,047 GB |
| 8-bit | 555 GB | 556 GB | 556 GB | 557 GB |
| 6-bit | 428 GB | 429 GB | 430 GB | 431 GB |
| 5-bit | 371 GB | 373 GB | 373 GB | 374 GB |
| 4-bit | 316 GB | 318 GB | 318 GB | 319 GB |
| 3-bit | 255 GB | 257 GB | 257 GB | 258 GB |
| 2-bit | 196 GB | 197 GB | 197 GB | 199 GB |
| 1.6-bit | 108 GB | 109 GB | 109 GB | 110 GB |
One conversation, 16-bit context cache, runtime overhead included.
What runs it
| Machine | Best fit | Context | Speed | First reply |
|---|---|---|---|---|
| RTX 3060 12 GB | Too big | |||
| RTX 4060 Ti 16 GB | Too big | |||
| RTX 3090 | Too big | |||
| RTX 4090 | Too big | |||
| RTX 5090 | Too big | |||
| 2× RTX 3090 | Too big | |||
| RTX PRO 6000 Blackwell | Too big | |||
| M4 Max 128 GB | Too big | |||
| M5 Max 128 GB | Too big | |||
| Ryzen AI Max+ 395 128 GB | Too big | |||
| DGX Spark 128 GB | 1.6-bit | 256K | 15 tok/s | 13 s |
| M3 Ultra 256 GB | 1.6-bit | 256K | 27 tok/s | 35 s |
| M5 Ultra 512 GB | 4-bit | 256K | 17 tok/s | 11 s |
| H100 SXM | Too big | |||
| DGX Station GB300 | 2-bit | 256K | 93 tok/s | 1.0 s |
| 8× H200 NVL | 8-bit | 256K | 42 tok/s | 0.5 s |
Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.
How to run it
- llama.cpp
llama-server -hf unsloth/NVIDIA-Nemotron-3-Ultra-550B-A55B-GGUF- vLLM
vllm serve nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16- MLX (Mac)
mlx_lm.chat --model mlx-community/Nemotron-3-Ultra-550B-A55B-4bit
Questions
How much VRAM does Nemotron 3 Ultra 550B A55B need?
Nemotron 3 Ultra 550B A55B needs 316 GB for its weights at 4-bit, 318 GB with an 8K context and 318 GB with 32K. At 8-bit, 556 GB.
Can Nemotron 3 Ultra 550B A55B run on an RTX 4090?
Not on one card. The smallest machine here that runs it is the DGX Spark 128 GB.