On January 6, Marvell Technology, Inc. announced a custom AI accelerator (XPU) architecture with integrated co-packaged optics (CPO) technology. The company reports the solution allows customers to seamlessly scale the size of their AI servers.
Marvell reports that its custom AI accelerator architecture combines its high-bandwidth memory (HBM) compute architecture, XPU compute silicon, and other chiplets with 3D silicon photonics engines on the same substrate.
The integrated optics, Marvell reports, allow connections between XPUs to achieve faster data transfer rates and distances 100 times longer than electrical cabling. The company further reports that CPO technology allows for increased XPU density, going from tens of XPUs within a rack to hundreds across multiple racks.
Will Chu, senior vice president and general manager of Marvell’s custom, compute, and storage group, reported that the custom XPUs can significantly increase the performance of AI servers.
“Integrating optics directly into XPUs takes custom accelerated infrastructure to the next level of scale and optimization that hyperscalers must deliver to satisfy the growing demands of AI applications,” said Chu in a press release.
