Huawei scales Ascend for larger models

Huawei is expanding its Ascend platform with larger systems designed to handle AI models containing up to ten trillion parameters.

Atlas 960 SuperPoD reaches 15,488 processors

Huawei's Atlas 960 SuperPoD roadmap calls for systems containing as many as fifteen thousand four hundred eighty-eight Ascend 960 processors, up from its earlier eight thousand one hundred ninety-two processor configuration.

New system delivers massive computing power

The system can deliver thirty EFLOPS of FP eight computing and sixty EFLOPS using FP four, alongside four thousand four hundred sixty terabytes of memory capacity and thirty four petabytes per second interconnect bandwidth.

Huawei accelerates chip release schedule

Huawei has accelerated development of the Ascend 960 family, with the 960DT scheduled for Q one twenty twenty-seven and the 960PR planned for Q three twenty twenty-seven, arriving earlier than previously expected.

David Wang outlines multi-year roadmap

David Wang, deputy chairman and rotating chairman of Huawei, said the company will roll out the Ascend 970 and 980 chips in twenty twenty-eight and twenty twenty-nine respectively, with compute specifications continuing to double.

“In 2028 and 2029, we will roll out the Ascend 970 and 980 chips, respectively.”

UnifiedBus reduces interconnection bottlenecks

UnifiedBus is Huawei's interconnect architecture for linking processors, memory, and other components within large AI computing systems, combining optical technology to extend high-speed connections deeper into the system.

Atlas 960E cuts power consumption dramatically

The Atlas 960E ties up to four thousand ninety-six NPUs and uses approximately five thousand five hundred Hi-ONE units instead of about forty-eight thousand conventional eight hundred G optical modules, reducing power consumption by more than five hundred fifty kilowatts.