Huawei unveils Peerium architecture for million-processor scale systems

Huawei announced the Peerium Computing Architecture, combining Nested BSP, unified memory addressing and the UnifiedBus interconnect so systems at roughly million-processor scale can act as one computer. First products are the Atlas 950 SuperPoD and SuperCluster with 256,000 cards, already being deployed, while an Atlas 960 system with near-packaged optics is under testing.

• Peerium uses Nested BSP, unified addressing and the UnifiedBus peer interconnect
• Atlas 950 SuperCluster with 256,000 cards is deploying; Atlas 960 is in testing
• Huawei says it extends, not replaces, von Neumann single-machine design
• No independent latency or bandwidth benchmarks were attached