Home > Storage > PowerFlex > White Papers > Dell Validated Design for Virtual GPU with VMware and NVIDIA on PowerFlex > Test methodology
The ResNet50 model performs image classification training on labeled images and inference on new images. The system learns whether an image contains a cat, dog, car, or train. The ResNet50 v1.5 script operates on ImageNet 1K, a popular image classification dataset from the ILSVRC challenge. All images are grouped into 1000 categories or classes. DL researchers commonly use this dataset for performance validation and comparison studies. In our tests, the following ResNet50 TF features are used.
• Multi-GPU training with Horovod: The model in this solution uses Horovod to implement efficient multi-GPU training with NCCL. For more information, see the example sources in this repository or the TensorFlow tutorial.
• NVIDIA DALI: DALI is a library that accelerates the data preparation pipeline. To accelerate the input pipeline, define your data loader with the DALI library. For more information, see the example sources in this repository or the DALI documentation.
• Automatic mixed precision (AMP): You can modify a computation graph by TensorFlow on runtime to support mixed precision training.
For this paper, the most common use cases of ResNet50 training, performance, and inference workloads were chosen to demonstrate that the PowerFlex family is well suited for NVIDIA A100 GPUs on VMware ESXi hosts.