Nvidia recently announced that its new Vera Rubin micro-architecture is in development and is scheduled to be launched in 2026. The Rubin CPX variant of this architecture will focus on meeting the needs of artificial intelligence workloads that require processing massive context windows. At a press conference, Nvidia CEO Jensen Huang stated: "The Vera Rubin platform will mark a new leap in AI computing, introducing the next-generation Rubin GPU and a new class of processor called CPX."

Rubin CPX is particularly suitable for applications that require processing over one million tokens, such as complex software development and high-definition video generation. According to Nvidia's plan, the Vera Rubin NDL144CPX GPU will be available by the end of 2026. The CPX model is specifically designed for applications requiring long context windows, offering 8 exaflops of AI performance, 30 PF NVFP4 context computing capability, and three times the exponential computation performance compared to the Nvidia GB300NVL72 system. In addition, the CPX model is equipped with 128GB GDDR7 memory, 4 encoders, and 4 decoders, designed specifically for video generation, and provides 100TB of fast memory.

Nvidia executives said that the Vera Rubin NDL144CPX can be considered part of a large artificial intelligence factory. To support the construction of large-scale data centers, Nvidia also plans to launch terascale reference designs. This means that Nvidia will closely collaborate with infrastructure companies to redesign data centers from a computing perspective, providing reference designs covering aspects such as building, design, simulation, and operation.

Before this release, Nvidia also announced the latest MLPerf inference test results, where the Blackwell GPU set a new record, surpassing the baseline of the Llama3.1405B interactive model. This innovative technology is called "disaggregated service," which allows the same hardware to achieve improved performance, providing additional revenue opportunities for enterprises that have already deployed solutions.

Key Points:  

🔍 **Nvidia releases Rubin CPX GPU, aimed at supporting large-context AI applications.**  

🚀 **This GPU will be available by the end of 2026, featuring powerful AI performance and memory configuration.**  

🏢 **Nvidia plans to launch terascale reference designs for data centers, helping to build AI factories.**