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
System configuration
Benchmark disclaimer
GPU building block MLPerf performance Dense compute building block performance Conclusion References

System configuration

Home > Workload Solutions > High Performance Computing > White Papers > Dell Validated Design for Government > System configuration

System configuration

  • Performance benchmarking was conducted in the Dell HPC and AI Innovation Lab using the system configurations listed in Table 3.

    Table 3.      Benchmark system configurations

    Building block

    AI/DL/ML/HPC GPU Compute - 1

    PowerEdge XE8545

    Dual AMD EPYC 7763

    512GB RAM 16x 32GB 3200 MTps DIMMs

    4x NVIDIA A100 80GB SXM4

    NVIDIA ConnectX-6 HDR adapter

    NVIDIA QM8790 InfiniBand Switch

    AI/DL/ML/HPC GPU Compute - 2

    PowerEdge R750xa

    Dual Intel Xeon Gold 6338

    512GB RAM 16x 32GB 3200 MTps DIMMs

    4x NVIDIA A100 80GB PCIe

    NVIDIA ConnectX-6 HDR adapter

    NVIDIA QM8790 InfiniBand Switch

    Dense Compute
    PowerEdge C6520
    Dual Intel Xeon Platinum 8352Y
    512GB RAM 16x 32GB 3200 MTps DIMMs
    NVIDIA ConnectX-6 HDR adapter

    NVIDIA QM8790 InfiniBand Switch

    The BIOS configurations used for the reference systems are described in Table 4 and Table 5.

    Table 4.      BIOS configuration – GPU Building Block 1

    BIOS Option

    Setting

    Logical Processor

    Disabled

    Virtualization Technology

    Disabled

    System Profile

    Performance

    NUMA Per Socket

    4

    Table 5.      BIOS configuration – GPU Building Block 2 and Dense Compute

    BIOS Option

    Setting

    Logical Processor

    Disabled

    Virtualization Technology

    Disabled

    System Profile

    Performance

    Sub NUMA Clustering

    Enabled

    The software versions for the MLPerf benchmarking of the GPU Compute building blocks are described in Table 6.

    Table 6.      Software versions for MLPerf Benchmarking

    Component

    Version

    Platform

    XE8545 (GPU Building Block1)

    R750xa (GPU Building Block 2)

    Operating System

    CentOS 8.2

    CentOS 8.2

    Kernel

    4.18.0-193.el8.x86_64

    4.18.0-193.el8.x86_64

    OFED

    5.4-1.0.3.0

    5.4-1.0.3.0

    Bright Cluster Manager

    9.1

    9.1

    MLPerf Training

    1.1

    1.1

    The software versions used for the Dense Compute reference system are listed in Table 7.

    Table 7.      Software versions for Dense Compute benchmarks

    Component

    Version

    Operating System

    Red Hat Enterprise Linux 8.3

    Kernel

    4.18.0-240.22.1.el8_3.x86_64

    OFED

    NVIDIA Mellanox 5.4-3.0.3.0

    Bright Cluster Manager

    9.0

    HPL

    2.3 (Intel oneAPI 2021.4.0)

    HPCG

    3.1 (Intel oneAPI 2021.4.0)

    STREAM

    5.10

    OSU Micro-Benchmarks

    5.9