NVIDIA recently disclosed detailed technical information about its next-generation data center CPU, Rosa. This processor will be released alongside a new GPU named Feynman, aiming to address rapidly evolving agentic AI workloads. The Rosa CPU targets further improvements in single-thread performance and energy efficiency within the same silicon area. Expected to launch in 2028, the Rosa CPU will utilize LPDDR5 memory and leverage the new BlueField-5 DPU to connect with the next-generation network card, ConnectX-10 SuperNIC. The Feynman GPU, manufactured using TSMC's 1.6nm process, delivers up to 50 PFLOPS of single-GPU performance, with inference capabilities five times that of the previous Blackwell architecture.
