NVIDIA Vera CPU is here: Designed from scratch for AI agents, fast 1

NVIDIA Vera CPU is here: designed from scratch for AI agents, 1.5 times faster, both OpenAI and Anthropic will be used AI search startup Perplexity announced plans to use NVIDIA's new CPU chip Vera. This chip is not an extension of NVIDIA's traditional GPU product line, but a new architecture product designed from scratch to welcome the AI ​​agent era. It targets the deep pain points of traditional CPUs being "acclimated" in AI agent scenarios. Intel AMD CPUs are a "legacy from the laptop era." AI agents act as autonomous programs, capable of continuously writing code, executing tasks, and coordinating complex workflows without pausing or taking breaks like human users. However, Intel and AMD CPUs were born in the era of laptop computers. Their architecture was originally designed to serve intermittent human-computer interaction. When dealing with the continuous loop calculations of AI agents, it is often difficult to achieve ideal results. Nvidia further analyzed the bottleneck of traditional CPUs in its official blog: slower execution speed will reduce the number of effective evaluations in each cycle of reinforcement learning, thereby extending training time and increasing single-user service latency. In addition, insufficient speed may also cause KV cache loss, pushing up computing power costs and forming a vicious cycle. Nate Kupp, vice president of infrastructure at Perplexity, said that in actual tests, Nvidia Vera CPU performed AI agent encoding tasks about 1.5 times faster than traditional CPUs, which is highly consistent with its core architecture. The core speed is increased by 1. 8 times, and the training cycle is greatly shortened. Vera's design directly addresses the above pain points. Its core speed is increased by 1. 8 times, and up to 85% of evaluation tasks can be completed within the same time window. At the same time, faster execution speed can provide richer reinforcement learning feedback, generate higher-quality training signals, significantly shorten the overall training cycle, and reduce the negative impact related to computing power. Perplexity did not disclose the specific purchase quantity, but NVIDIA has previously confirmed that companies such as OpenAI, Anthropic and Oracle also plan to use this CPU. As top AI laboratories have launched self-developed chip plans, Nvidia is actively broadening its hardware layout, and Vera is a key step in its transformation from "GPU overlord" to "full-stack computing power provider". When AI agents become the core form of next-generation software, the underlying hardware tailored for them may redefine the competitive landscape of the entire computing power market. via AI News (author: AI Base)