On August 29, Kuai Technology reported significant developments from NVIDIA, following the release of an impressive financial report. NVIDIA announced the successful production of its next-generation data center GPU, Rubin, and CPU, Vera, at TSMC. These advancements are set for release next year as planned.
According to Collette Cres, NVIDIA’s CFO, the Rubin platform chips, which include the Vera CPU, Rubin GPU, CX9 Super NIC network card chip, NVLink144/Spectrum X switch chip, and a silicon optical chip, are currently in production. Cres stated, “The Rubin GPU will be put into mass production as scheduled next year.”
Drilling, a critical phase in chip development, has been completed successfully, meeting expected design standards. The next steps involve producing samples for verification, testing, and optimization.
Rubin and Vera were initially announced last year. Rubin is designed to replace the existing Blackwell series and its future iteration, Blackwell Ultra. Named after American astronomer Vera Rubin, the GPU will feature the next generation of HBM4 high bandwidth memory with eight stacks, with the first model being the R100. It will utilize TSMC’s 3nm EUV process. An upgraded version, Rubin Ultra, is planned for 2027, boasting 12 stacks of HBM4 memory for enhanced capacity and performance.
Rubin is aimed at both data center and consumer markets, similar to Blackwell, and will serve as the foundation for RTX 60 series graphics cards. Together, the Vera CPU and Rubin GPU will create a new generation of super chips with an upgraded sixth-generation NVLink interconnect bus, offering a bandwidth of up to 3.6TB/s.
The enhanced CX9 NIC network card for data centers will offer a remarkable bandwidth of 1600Gbps, equivalent to 1.6 million Gigabits.
For more information, remember to credit Kuai Technology if you need to reference this article.
Source link
