Benchmarking Neural Networks on Oracle Cloud Infrastructure with Mapr
High-speed data access and using this data to quickly train machine-learning (ML) and deep learning (DL) models is a priority for enterprises building next-gen data-intensive applications. Building data-driven use cases usually entails gathering data assets, building complex ML and DL models, training them using the data assets, and finally deploying them in production. Data scientists and IT operations personnel spend 30-65% of their time obtaining the data and making it available for ML models.
The MapR® Data Platform provides industry-leading high-speed data access performance. Coupled with the flexibility of Oracle Cloud Infrastructure (OCI) and its use of NVIDIA Tesla GPUs, MapR delivers the industry’s best performing solutions for model training for enterprises.
This paper presents benchmarking conducted by MapR with help from Oracle and the resulting performance numbers with ResNet models on the OCI. The test results show that using this combination of technologies from MapR, Oracle, and NVIDIA allows nearly 100% utilization of the GPU processing power.