MLPerf is growing at an extended rate. Different market analysts including Hyperion Research, IDC, and so on all agree that this growth validates that MLPerf is a substantially helpful benchmark to make fair comparisons and purchasing decisions. These comparisons are especially helpful to see that different vendors and OEMs collaborate to produce a high-performance system. Owing to its fast growth, the number of results submitted to MLPerf has grown approximately 150 percent in this round.

Furthermore, some customers use MLPerf as an entry point to assess the performance that they can expect from a new system that they plan to acquire. The system might be a new NVIDIA accelerator or a new Dell server, or an overall system including the server and the accelerator.

All Dell submissions to MLPerf Training v2.0 include NVIDIA accelerators. These accelerators enable a full spectrum of use cases that are seen in the MLPerf training benchmarks to run at faster time to convergence. For example, in MLPerf v1.1, we had the full spectrum of the results for all the benchmarks. Due to time constraints, for this round of submission to MLPerf v2.0, we did not have the full spectrum of results.

The following figure shows the results with an exponentially scaled y axis:

Figure 1.   Full spectrum of results across all benchmarks in the MLPerf Training suite v1.1

The preceding figure shows all the results that were submitted with the XE8545x4A100-SXM-80GB to MLPerf Training v1.1 and all the results converged to the target accuracy. These results show that the Dell PowerEdge XE8545 server with A100-SXM-80GB cards can handle different classes of workloads.

The following figure shows the percentage of submissions derived from the number of closed results. Our results were extensive, followed by H3C and NVIDIA. It is encouraging to see new submitters provide a substantial number of submissions. This addition fosters more collective results and opportunities to obtain a specific result, if needed. For example, with a larger number of results, it is possible to obtain a specific datapoint about ResNet with a specific accelerator.

Figure 2.   For closed division submissions, Dell Technologies submitted the greatest number of results compared to other submitters