Reference Guide

24 Dell EMC Ready Solutions for AI Deep Learning with NVIDIA | v1.0
(c) VGG16
Figure 10: Neural network training performance with different storages systems and image database
options. The batch size is 256, 256 and 128 for AlexNet, Resnet50 and VGG16, respectively. All training
are in FP16 mode.
The training performance of the tested neural networks are affected by different flags in the benchmark. It was
found that the training performance of Resnet50 -
- four. There was no obvious performance improvement for VGG16 and
AlexNet. The performance comparison between the baseline version and the tuned version for Resnet50 is
shown in Figure 11. The baseline performance is the performance presented in Figure 10(b). With one node,
the performance was improved from 2,657 images/sec to 2,940 images/sec. With two nodes, the performance
was improved from 4,560 images/sec to 5,590 images/sec.
Figure 11: The performance comparison of Resnet50 between the baseline and tuned versions