Center of Excellence

The NVIDIA DGX A100 is a groundbreaking system designed to revolutionize AI and high-performance computing (HPC). Launched in 2020, it is equipped with up to 8 NVIDIA A100 Tensor Core GPUs, each capable of delivering up to 312 TFLOPS of deep learning performance. The A100 GPU, based on NVIDIA's Ampere architecture, enhances both performance and efficiency for AI and HPC workloads. A key feature of the DGX A100 is its Multi-Instance GPU (MIG) technology, which allows each A100 GPU to be partitioned into as many as seven instances, enabling efficient resource sharing among multiple users and workloads. This flexibility ensures optimal utilization for both large-scale AI models and smaller parallel tasks.

The DGX A100 includes high-speed interconnects with NVIDIA NVLink, ensuring seamless data transfer and synchronization across all GPUs. Additionally, NVIDIA NVSwitch technology allows all GPUs to communicate at full bandwidth, eliminating bottlenecks in multi-GPU configurations. The system boasts 320 GB of HBM2e GPU memory and 15 TB of NVMe SSD storage, providing massive capacity and speed for data-intensive applications. Complementing its hardware, NVIDIA offers a comprehensive software stack with optimized libraries, pre-trained models, and development tools, simplifying the deployment and management of AI workflows. Compatibility with leading AI frameworks like TensorFlow, PyTorch, and MXNet ensures seamless integration into existing workflows.