CPU for the AI agent, which supports tool use and code execution, has entered the shipment phase.NVIDIA's Vera CPU is part of the 7 chip configuration, which combines with the GPU and the inference accelerator, connection and network components.
CPU for agents enters product supply phase
type of agent.AIDesigned for it.NVIDIAThe firstCPUbyNVIDIAVera isCPUThe shipment has begun.NVIDIAThe stages indicated are not limited to the publication of design and adoption plans,CPUThis is the transition to supply as a product.
In the agent type AI, the processing of the CPU side involves the model inference operation as well as the use of external tools and code execution.Vera CPU is positioned as a processor that incorporates this processing path into an agent base.
Assign tool use and code execution to CPU
SpaceXAI is the future.AIArchitecture by Vera.CPUAdopted and demonstrated plans to speed up orchestration, tool use, code execution, data processing, and simulation.CPUTreatment of Vera.CPUAdmission plan to be assigned.
NVIDIAby VeraCPUWith high core performance and memory bandwidth, it offers predictable performance even when loaded.This design reduces agent processing time,GPUIt supports the maintenance of the base.
The 7 type of chip containing CPU.
Agent for the platform.NVIDIAVera Rubin has different uses.7Combine types of chips. Component elements are VeraCPUby RubinGPUand Growth3byLPUby NVLink6Switch and BlueField4byDPUThe Spectrum-6bySPXand ConnectX-9It is super nic.
This configuration includes not only CPU and GPU, but also the accelerator, chip interconnection, data processing and networking components.With the launch of Vera CPU, the CPU for agents has become the product of the supply phase that shapes this differential chip configuration.
Separate the contribution of CPU from the measurement value of 30.
NVIDIA used the SemiAnalysis AgentX workload to evaluate the rax scale inference system NVIDIA Vera Rubin NVL72. This workload preserves the increased context, tool calls, and sub-agent generation generated by actual agent-type coding sessions.
In the initial measurement of NVIDIA, the Vera Rubin NVL72 powered by the DeepSeek V4 Pro showed a maximum output of GB300 NVL72 to 1 per megawatt of 30. However, this result is waiting for confirmation by SemiAnalysis and does not include the performance of the tool call by the Vera CPU, so it is not a value that shows the acceleration of the CPU single.