Our results are based on single-node Dell PowerEdge XE9680, XE8640, and R760xa servers with a combination of four NVIDIA GPUs:

• NVIDIA H100 PCIE 80 GB
• NVIDIA H100 SXM5 80 GB
• NVIDIA A100 PCIE 80 GB
• NVIDIA A100 SXM5 80 GB

We also submitted with previous generation PowerEdge servers, such as the PowerEdge XE8545 and R750xa servers with NVIDIA A100 GPUs, to help compare versions.

New benchmark

We submitted the DLRMv2 results on the Dell PowerEdge XE9680 and XE8640 servers with NVIDIA H100 and NVIDIA A100 GPUs.

Different operating systems

We ran tests on different operating systems to show the performance difference. The testing shows that Dell PowerEdge servers and NVIDIA accelerators work well for different workloads and models.

Submitted configurations

Our MLPerf Training v3.0 submission includes multinode and single-node systems. This whitepaper describes the performance of the latest generation single-node systems.

Note: A future whitepaper will describe the performance of multinode systems.

The following general syntax is used for the system name:

<Number of servers> x < Dell server name> x <number of accelerators> x <NVIDIA accelerator name>

To identify a single-node system, note that there are no entries for the number of servers or the number of servers is equal to one. For example, R750xax4A100-PCIE-80GB is a single-node system.

The following table lists the single-node systems to show performance improvement rendered by the newer generation servers.

Note: All systems include NVMe drives.

Table 3.      Dell single-node systems