Skip to content

Nemotron 3 Ultra 550B A55B VRAM requirements

How much VRAM Nemotron 3 Ultra 550B A55B needs
4-bit weights: 316 GBWith 8K context: 318 GBWith 32K context: 318 GB8-bit, 8K context: 556 GB
Parameters
561B
Active per token
55B
Longest context
256K
Released
Jun 2026
License
OpenMDW 1.1
Strongest at
Chat, Coding, Docs

How much VRAM Nemotron 3 Ultra 550B A55B needs

Memory Nemotron 3 Ultra 550B A55B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit1,044 GB1,045 GB1,046 GB1,047 GB
8-bit555 GB556 GB556 GB557 GB
6-bit428 GB429 GB430 GB431 GB
5-bit371 GB373 GB373 GB374 GB
4-bit316 GB318 GB318 GB319 GB
3-bit255 GB257 GB257 GB258 GB
2-bit196 GB197 GB197 GB199 GB
1.6-bit108 GB109 GB109 GB110 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Nemotron 3 Ultra 550B A55B

The best way Nemotron 3 Ultra 550B A55B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Nemotron 3 Ultra 550B A55B on RTX 4060 report
RTX 5060Too bigPDF: Nemotron 3 Ultra 550B A55B on RTX 5060 report
RTX 3070Too bigPDF: Nemotron 3 Ultra 550B A55B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on RTX 3060 12 GB report
Arc B580Too bigPDF: Nemotron 3 Ultra 550B A55B on Arc B580 report
RTX 4070Too bigPDF: Nemotron 3 Ultra 550B A55B on RTX 4070 report
RTX 4070 SuperToo bigPDF: Nemotron 3 Ultra 550B A55B on RTX 4070 Super report
RTX 5070Too bigPDF: Nemotron 3 Ultra 550B A55B on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Nemotron 3 Ultra 550B A55B on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Nemotron 3 Ultra 550B A55B on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Nemotron 3 Ultra 550B A55B on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Nemotron 3 Ultra 550B A55B on RTX 5070 Ti report
RTX 5080Too bigPDF: Nemotron 3 Ultra 550B A55B on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Nemotron 3 Ultra 550B A55B on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Nemotron 3 Ultra 550B A55B on RTX 3090 report
RTX 4090Too bigPDF: Nemotron 3 Ultra 550B A55B on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Nemotron 3 Ultra 550B A55B on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Nemotron 3 Ultra 550B A55B on RTX 5090 report
2× RTX 3090Too bigPDF: Nemotron 3 Ultra 550B A55B on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Nemotron 3 Ultra 550B A55B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: Nemotron 3 Ultra 550B A55B on RTX 6000 Ada report
L40SToo bigPDF: Nemotron 3 Ultra 550B A55B on L40S report
2× RTX 5090Too bigPDF: Nemotron 3 Ultra 550B A55B on 2× RTX 5090 report
4× RTX 3090Too bigPDF: Nemotron 3 Ultra 550B A55B on 4× RTX 3090 report
RTX PRO 6000 BlackwellToo bigPDF: Nemotron 3 Ultra 550B A55B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on Mac mini M4 Pro 64 GB report
M4 Max 128 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on M4 Max 128 GB report
M5 Max 128 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on M3 Ultra 96 GB report
M3 Ultra 256 GB1.6-bit256K27 tok/s35 sPDF: Nemotron 3 Ultra 550B A55B on M3 Ultra 256 GB report
M3 Ultra 512 GB4-bit256K13 tok/s35 sPDF: Nemotron 3 Ultra 550B A55B on M3 Ultra 512 GB report
M5 Ultra 512 GB4-bit256K17 tok/s11 sPDF: Nemotron 3 Ultra 550B A55B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GBToo bigPDF: Nemotron 3 Ultra 550B A55B on Ryzen AI Max+ 395 128 GB report
RTX Spark1.6-bit256K16 tok/s13 sPDF: Nemotron 3 Ultra 550B A55B on RTX Spark report
DGX Spark1.6-bit256K15 tok/s13 sPDF: Nemotron 3 Ultra 550B A55B on DGX Spark report
Dual DGX Spark2-bit256K8.6 tok/s7.2 sPDF: Nemotron 3 Ultra 550B A55B on Dual DGX Spark report
DGX Station GB3002-bit256K93 tok/s1.0 sPDF: Nemotron 3 Ultra 550B A55B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell1.6-bit256K63 tok/s1.7 sPDF: Nemotron 3 Ultra 550B A55B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell4-bit256K30 tok/s0.9 sPDF: Nemotron 3 Ultra 550B A55B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K19 tok/s0.5 sPDF: Nemotron 3 Ultra 550B A55B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: Nemotron 3 Ultra 550B A55B on A100 80 GB SXM report
H100 SXMToo bigPDF: Nemotron 3 Ultra 550B A55B on H100 SXM report
4× H200 NVL6-bit256K50 tok/s0.9 sPDF: Nemotron 3 Ultra 550B A55B on 4× H200 NVL report
8× H200 NVL8-bit256K42 tok/s0.5 sPDF: Nemotron 3 Ultra 550B A55B on 8× H200 NVL report

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.

What is the minimum VRAM for Nemotron 3 Ultra 550B A55B?

About 109 GB at 1.6-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the M3 Ultra 256 GB.

How big is Nemotron 3 Ultra 550B A55B?

561B parameters (55B active per token): 1,044 GB of weights at 16-bit, 316 GB at 4-bit.

Can Nemotron 3 Ultra 550B A55B run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the M3 Ultra 256 GB.

More models