This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Read our privacy policy>

Reminder

To have a better experience, please upgrade your IE browser.

upgrade
If you need help, please click here:
MENU

Kevin Tolly

Founder if The Tolly Group

Kevin Tolly

Founder of The Tolly Group

Huawei AI Fabric Has Passed Tolly Test and Delivers Industry-leading Performance

Huawei commissioned Tolly to evaluate Huawei AI Fabric solution’s performance in three application scenarios (high performance computing - HPC, distributed AI training, and distributed storage), and compared it with Cisco Nexus switches’ performance. Huawei AI Fabric and Cisco’s solution are both based on RDMA over Converged Ethernet (RoCEv2). In all three scenarios, Huawei AI Fabric’s performance was better than Cisco Nexus switches.

Huawei AI Fabric Solution Takes the Lead in Three Scenarios

  • High-Performance Computing (HPC)

    The Job Completion Time of Huawei AI Fabric Solution is 30% less than that of Cisco.

    In scenarios such as life science and meteorology, the communication delay is shortened.In scenarios such as life science and meteorology, the job completion efficiency is improved.

    AI Fabric
    Cisco
    29.9%

    Job Completion Efficiency

    View Test Details
  • Distributed AI Training

    The AI training efficiency of Huawei AI Fabric Solution is 27% higher than that of Cisco solution.

    In scenarios such as voice recognition and autonomous driving, the computing efficiency improvement can reduce the investment on expensive GPU servers.

    AI Fabric
    Cisco
    27.5%

    AI Training Efficiency

    View Test Details
  • Distributed Storage

    The storage performance of Huawei AI Fabric Solution is better than that of Cisco solution by over 30%.

    The IOPS performance is improved in public cloud scenarios, increasing the revenue of cloud disks.

    AI Fabric
    Cisco
    30.5%

    IOPS

    View Test Details

Detailed Test Data in the HPC Scenario

HPC Scenario Intel MPI Benchmarks (IMB) Test Results (MPI_Allreduce)

Detailed Test Data in the Distributed AI Training Scenario

TensorFlow-GPU Distributed AI Training (Parameter Server Architecture) Test Results

Detailed Test Data in the Distributed Storage Scenario

Distributed Storage FIO Tool Test Results