"six chips that make one AI supercomputer."
"Vera Rubin is designed to address this fundamental challenge that we have: The amount of computation necessary for AI is skyrocketing,"
"Compared to the Blackwell model, Rubin marks a leap in performance, with more than triple the speed, could run inference five times faster, and can deliver significantly more inference compute per watt of energy."
Nvidia officially introduced the Vera Rubin architecture, which is in production and expected to ramp volume in the second half of the year. Vera Rubin combines six chips into a single system-level AI supercomputer design and targets sharply higher computational needs for AI workloads. The architecture claims more than triple the speed of Blackwell, up to five times faster inference, and significantly greater inference compute per watt. Rubin is designed to support more complex agent-style AI workloads and increased networking and data movement. Major cloud partners including AWS, OpenAI, and Anthropic plan deployments, and national lab systems will adopt Rubin. The launch follows record Nvidia data center revenue up 66% year-over-year.
Read at Business Insider
Unable to calculate read time
Collection
[
|
...
]