"During a press briefing ahead of today's keynote, Dion Harris, Nvidia's senior director of HPC and AI infrastructure solutions, described Vera Rubin as "six chips that make one AI supercomputer." Those six chips include the Vera CPU, Rubin GPU, NVLink 6th-gen switch, Connect-X9 NIC, BlueField4 DPU, and Spectrum-X 102.4T CPO. The platform will support 3rd-generation confidential computing and, according to Nvidia, will be the first rack-scale trusted computing platform."
"Nvidia claims the Rubin GPU is capable of delivering five times as much AI training compute as Blackwell. The Vera Rubin architecture as whole can train a large "mixture of experts" (MOE) AI model in the same amount of time as Blackwell while using a quarter of the GPUs and at one-seventh the token cost. The Rubin launch was originally expected for late this year. Its early arrival today comes just a couple of months after Nvidia reported record high data center revenue, up 66 percent over the prior year."
Nvidia launched the Vera Rubin computing platform ahead of schedule, integrating six specialized chips into a single rack-scale trusted computing system that supports 3rd-generation confidential computing. The platform components are the Vera CPU, Rubin GPU, NVLink 6th-gen switch, Connect-X9 NIC, BlueField4 DPU, and Spectrum-X 102.4T CPO. Nvidia claims the Rubin GPU delivers five times the AI training compute of Blackwell. The Vera Rubin architecture can train a large mixture-of-experts (MOE) model in the same time while using one-quarter the GPUs and one-seventh the token cost. Partner-delivered Rubin products and services will appear in the second half of 2026.
Read at The Verge
Unable to calculate read time
Collection
[
|
...
]