November 18th, 2019 by Lyle Smith
NVIDIA Announces Scalable GPU-Accelerated Supercomputer in the Microsoft Azure Cloud
NVIDIA has announced the availability of a new cloud-based, GPU-accelerated supercomputer available on Microsoft Azure. Built to handle demanding AI, machine learning and high-performance computing applications, NVIDIA indicates that their new offering will provide significant performance and cost advantages compared to traditional CPU-based computing. For example, AI researchers will be able to spin up multiple NDv2 instances and train complex conversational AI models in hours, says the company.
To build this new scalable GPU-accelerated supercomputer, Microsoft and NVIDIA engineers used 64 NDv2 instances on a pre-release version of the cluster to train BERT in approximately three hours. This was possible due to the multi-GPU optimizations provided by NCCL, an NVIDIA CUDA X library, and high-speed Mellanox interconnects.
NVIDIA adds that those using multiple NDv2 instances will also notice a range of benefits when running complex HPC workloads. Even a single NDv2 instance will deliver much faster results compared to a traditional HPC node without GPU acceleration for specific types of applications, such as deep learning. This performance also can scale linearly to a hundred instances for large-scale simulations.
NVIDIA also claims that all NDv2 instances will benefit from the GPU-optimized HPC applications, machine learning software and deep learning frameworks from the NVIDIA NGC container registry and Azure Marketplace.
NDv2 is available now in preview.