Reference Guide

30 Dell EMC Ready Solutions for AI Deep Learning with NVIDIA | v1.0
ResNet-152 74.90% 92.21% 74.84% 92.16% 0.06% 0.05%
VGG-16 68.35% 88.45% 68.30% 88.42% 0.05% 0.03%
VGG-19 68.47% 88.46% 68.38% 88.42% 0.09% 0.03%
GoogLeNet 68.95% 89.12% 68.77% 89.00% 0.18% 0.12%
AlexNet 56.82% 79.99% 56.79% 79.94% 0.03% 0.06%
Figure 16: Resnet50 inference performance on V100 vs P100 (V100 is 3.7x faster than P100)
To demonstrate the inference advantage of V100 over P100, tests in FP16 mode were performed on both V100
and P100 and the result is shown in Figure 16. Batch size of 39 was used for V100 and 10 for P100. Different
batch sizes were chosen for the two GPU generations such that the inference latencies were almost identical
(~7ms in the figure). The result shows that when inference latency is held constant, the inference throughput of
the V100 is 3.7x faster when compared to the P100.
3.3 NVIDIA DIGITS Tool and the Deep Learning Solution
DIGITS is a web frontend to Caffe, Torch and TensorFlow, developed by NVIDIA. The user can use the DIGITS
graphical user interface for Deep Learning training and inference. The Deep Learning Solution presented in this
paper integrates the DIGITS software with the cluster software making it trivial for the system administrator to
deploy DIGITS for the users, and easy for the user to use DIGITS.
In order to use DIGITS, the user first needs to allocate the resource, load the DIGITS module and start DIGITS
Figure 17.