Dell PowerEdge C4140 Deep Learning Performance Comparison - Scale-up vs. Scale - Page 15

Deep Learning Performance: Scale-up vs Scale-out Performance Metrics Training Tests Dataset  Throughput images/second  Top-5 Accuracy on the training dataset  Training time  Short Tests to get throughput images/second  Long Tests to get accuracy convergence and training time ILSVRC2012 Table 1: Benchmark Setup 4.3 Neural Models & parameters Short Tests Long Tests Models  VGG-16  VGG-19  Inception-v3  Inception-v4  ResNet50  GoogLeNet  AlexNet  VGG-19  Inception-v4  ResNet50 Batch sizes  32  64  128  256  512  2014  2048  32  64  128  256 GPUs  1  3  2  4  3  8  4  5  6  7  8 Training Steps or Epochs 100 steps 90 epochs Table 2: Neural Models & Parameters Architectures & Technologies Dell EMC | Infrastructure Solutions Group 14