NVIDIA Star Elastic: 360× Savings on Neural Network Training
NVIDIA has unveiled Star Elastic, a method that trains three models of different sizes (30B, 23B and 12B parameters) in a single 160B-token run. The compute savings: 360×. At inference, elastic budget control improves accuracy by 16% and cuts latency by a factor of 1.9. After quantization, all three models fit on a single RTX GPU.
AI-processed from MarkTechPost; edited by Hamidun News
NVIDIA has unveiled Star Elastic, a neural network training method that fits three models of different sizes (30B, 23B and 12B parameters) into one weights file, retraining all of them from scratch in a single training run instead of three separate ones.
How much does
Star Elastic save when training neural networks? Star Elastic trains three NVIDIA Nemotron models (30B, 23B, 12B parameters) in a single 160B-token run, which saves 360 times the compute compared with training each model separately.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.