The next generation of GPU will be here, Rubin CPX will deduce millions of Token

The introduction of Rubin CPX GPU, designed for long-situation reasoning in British Weida, can be reasoned in millions of tokens at a time to support long-situ missions such as software development and video generation; the GPU will combine NVIDIA Vera CPU with Rubin GPU to form the Vera Rubin NVL 144 CPX platform, which provides 8 exaflops AI, 7.5 times the GB300 NVL72 system; the system is equipped with 100TB high-speed RAM and 1.7 PB/s RAM bandwidth, which is expected to be put on the market by the end of 2026, and will provide an unprecedented increase in performance and efficiency for long-sink missions。