White Papers

Deep Learning Performance: Scale-up vs Scale-out
Architectures & Technologies Dell EMC | Infrastructure Solutions Group
22
7 Performance Results
7.1 Single Node – Throughput (images/sec)
The charts below show the results for different servers running the short tests to extract
throughput images/second using ResNet50 with batch size 128 and number of steps =100.
The results for single node are with maximum number of GPUs supported within that node.
7.1.1 PowerEdge R740xd
Figure 15: PowerEdge R740-P40 server with up to 3 GPUs
The PowerEdge R740 with P40 GPU is tested with different pre-trained neural models. The results
show how different models use the amount of available memory e.g. ResNet50 uses more
memory than GoogLeNet or AlexNet and therefore we see lower throughput.