nvidia
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16
text generationother1.1M downloadsHugging Face
No server in stock right now has the 1253 GB this model needs for inference.
GPU memory it needs (estimate)
Rule of thumb from the model's size. Real needs depend on batch size, context length and framework.
Inference (full precision)
1253 GB
Inference (8-bit)
626 GB
Inference (4-bit)
313 GB
LoRA fine-tuning
1357 GB
QLoRA (8-bit base)
679 GB
Full fine-tuning
5011 GB