Huang Renxun explains Vera CPU in detail: the world's first processor designed for agents rather than humans According to Beating monitoring, Huang Renxun extensively introduced the design philosophy of Vera CPU in his GTC Taipei 2026 speech. He pointed out that in the past, all CPUs were designed for humans, and the world of human interaction is measured in seconds. Cloud CPUs are rented out based on the number of cores. But the world of agents is measured in nanoseconds: every time a tool is called or a database is accessed, the agent demands the fastest response, and any waiting will block the next step of reasoning. In the past, we built CPUs for 1 billion humans, and in the future, we will build CPUs for billions of agents. He summarized the four design pillars of Vera CPU. Firstly, the single threaded performance (IPC) is the highest in the world: it can fetch, decode, and execute 10 instructions per clock cycle, as agents require extremely low latency rather than traditional throughput. Secondly, each core has world-class bandwidth. Thirdly, the total on-chip bandwidth has exceeded its limit: the new interconnect architecture connects all CPU cores at the speed of light, with a cross-sectional bandwidth of 3.6 TB/s, no chiplet boundaries, and no cross chip overhead. All cores work together instead of being rented out by core. Fourth, ultimate energy efficiency: Deploy as many CPUs as possible without squeezing the power generated by GPU tokens. Vera CPU simultaneously achieved multiple industry firsts: the world's first CPU to support PCIe Gen 6, and the first server processor to be equipped with LPDDR5 memory and achieve 1.2 TB/s bandwidth (2 to 3 times the current highest performance x86 CPU). Huang Renxun stated that achieving a 5% improvement in the CPU industry is already astonishing, and 10% is even rarer. However, the performance gap between Vera CPUs and the strongest x86 is an unprecedented magnitude. He also revealed that Nvidia has sold millions of Grace CPUs (Grace Blackwell series), making it one of the world's largest CPU manufacturers. The Vera CPU plays three roles in the system: orchestrating GPUs and managing KV cache within the Vera Rubin NVL72 rack; As the harness layer of the Agent, it runs model orchestration, tool invocation, and database access; And drive the world's fastest AI storage server in the Vera BlueField storage system. [Original link]