Your Browser is Out of Date

Nytro.ai uses technology that works best in other browsers.
For a full experience use one of the browsers below
Dell Validated Design for Government
Introduction Business challenges Solution overview Solution architecture Reference system GPU building block MLPerf performance
Introduction
Single server performance
Dense compute building block performance Conclusion References

Single server performance

Home > Workload Solutions > High Performance Computing > White Papers > Dell Validated Design for Government > Single server performance

Single server performance

  • The performance metrics for the four MLPerf Training benchmarks on a single server are described in Table 8.

    Table 8.      Select MLPerf Training results for GPU Building Blocks (in minutes)

    Component

    Image Classification (Resnet50)

    NLP (BERT-large)

    Object detection, heavy weight (Mask R-CNN)

    Image segmentation (3D U-Net)

    XE8545 (GPU Building Block 1)

    56.326

    38.855

    83.774

    55.999

    R750xa (GPU Building Block 2)

    62.949

    56.260

    93.134

    60.586

    The results in Table 8 are the training time in minutes for specific MLPerf Training benchmarks, where smaller values are better. The primary difference between the two building blocks is that the XE8545 building block has 400 W SXM4 GPUs and the R750xa has 300 W PCIe GPUs. The full MLPerf results are available at MLCommons results page.