Nvidia’s Vera CPU team says the rise of agentic AI is reversing a long-standing trend in CPU design. For years, the industry focused on chiplet designs to increase core counts, often at the expense of single-threaded performance.
Agentic AI workloads often involve sequential tasks, such as tool calls and code execution. These tasks can be limited by single-core speed, not just the number of parallel cores. Faster individual cores and lower latency can improve system efficiency and help prevent expensive GPUs from sitting idle.
Nvidia says its Vera CPU, with a monolithic design and custom Olympus cores, is built for this demand. The company aims to deliver what it calls “max single-threaded performance at scale.”