Abstract

With CloudEngine (CE) series switches, Huawei developed the AI Fabric Intelligent Lossless Data Center Network Solution which faces artificial intelligence (AI) era’s data centers.

Huawei commissioned Tolly to evaluate Huawei AI Fabric solution’s performance and compared it with Mellanox switches’ in the high performance computing - HPC scenario and the distributed AI training scenario. Huawei AI Fabric is based on RDMA over Converged Ethernet (RoCEv2) while Mellanox Technology’s solution is based on InfiniBand. In both test scenarios, Huawei AI Fabric’s performance was on par with the performance of Mellanox.