The new chips are expected to be more expensive than previous generations, but the efficiency gains will accelerate the development of massive AI models. Some observers have criticized the performance data charts provided by the company as misleading.
Story
Nvidia Vera Rubin NVL72 Cuts GPU Needs for Large AI Model Training by 75%
AI17 hours ago
The Vera Rubin NVL72 is capable of training a 10T-parameter mixture-of-experts AI model on 100T tokens in one month using approximately 16,000 GPUs. The same process requires 64,000 GB200 GPUs under the current Grace Blackwell architecture.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.