Making compute power directly into Tokens! Hygon debuts an open computing architecture: CPU handles scheduling, DCU handles production.

On August 28, Hygon Information Technology unveiled its "Agent to Token Open Computing Architecture" yesterday, leveraging the synergy of CPU and DCU dual-chip collaboration and open interconnect to provide a new computing framework from chips to applications for the large-scale development of the token economy.

Currently, large models are rapidly penetrating industrial scenarios, with agents becoming the core driver for AI deployment in real-world businesses. Tokens, as the key medium connecting business needs, computing investment, and service output, have their production efficiency and unit cost directly determining the depth of agent applications.

This architecture integrates agent scheduling and token generation into a unified computing framework through the "CPU+DCU" dual-chip design and open interconnect system. Hygon CPUs serve as the system hub for agent business flows, covering core modules such as data preprocessing, task orchestration and decision-making, hierarchical memory storage, and vector database retrieval.

DCUs focus on parallel computing acceleration for token generation, handling tasks including batch decoding, KV cache optimization, preprocessing acceleration, and long-sequence support. The dual-chip collaboration dynamically allocates resources based on agent concurrency scale and task complexity, precisely converting computing power into token productivity.

Hygon's "Agent to Token Open Computing Architecture" is fully open across four dimensions: computing, interconnect, security, and software stack. Computing openness supports native dual-chip combinations; interconnect openness breaks bandwidth bottlenecks through the self-developed HSL protocol; security openness relies on chip-level intrinsic security for end-to-end trustworthiness; and software stack openness adapts to over 100 AI frameworks to lower entry barriers.

This release represents Hygon's systematic computing upgrade for the token economy, promoting the coordinated deployment and application of domestic CPUs, DCUs, and various accelerator cards within a unified computing system.