AceleMax DGS-1216AS

Purpose-Built for the Convergence of Simulation, Data Analytics, and AI

Massive datasets, exploding model sizes, and complex simulations require multiple GPUs with extremely fast interconnections. The NVIDIA HGX™ platform brings together the full power of NVIDIA GPUs, NVIDIA® NVLink®, NVIDIA Mellanox® InfiniBand® networking, and a fully optimized NVIDIA AI and HPC software stack from NGC™ to provide highest application performance. With its end-to-end performance and flexibility, NVIDIA HGX enables researchers and scientists to combine simulation, data analytics, and AI to advance scientific progress. With a new generation of A100 80GB GPUs, a single HGX A100 now has up to 1.3 terabytes (TB) of GPU memory and a world’sfirst 2 terabytes second (TB/s) of memory bandwidth, delivering unprecedented acceleration for emerging workloads, fueled by exploding model sizes and massive data-sets.

Third-Generation NVIDIA NVLink Creates a Single Super GPU

Scaling applications across multiple GPUs requires extremely fast movement of data. The third generation of NVIDIA NVLink in the NVIDIA A100 Tensor Core GPU doubles the GPU-to-GPU direct bandwidth to 600 gigabytes per second (GB/s), almost 10X higher than PCIe Gen4. Third-generation NVLink is available in four-GPU and eight-GPU HGX A100 servers from leading computer makers.

Multi-Instance GPU (MIG) Delivers Seven Accelerators in a Single GPU

Every AI and HPC application can benefit from acceleration, but not every application needs the performance of a full A100 Tensor Core GPU. With MIG, each A100 can be partitioned into as many as seven GPU instances, fully isolated at the hardware level with their own high-bandwidth
memory, cache, and compute cores. This allows HGX A100 systems to offer up to 112 GPU instances, giving developers access to breakthrough speed for every application, big and small, with guaranteed quality of service.

With A100 80GB, seven MIGs can be configured with 10 GB each (double the size of A100 40GB MIGs), making it now possible to perform inference on batch-size constrained models like BERT-LARGE (a natural language processing model with superhuman understanding) at much higher
batch sizes, delivering up to a 1.3X increase in throughput.

Design Versatility to Suit Any Workload

NVIDIA HGX™ A100 delivers a best-in-class server platform through GPU baseboards and a design guide that provides different configuration options. This allows unmatched versatility, enabling server manufacturers to build a range of CPU and GPU systems or cloud instances ideal for different workloads.

Third-Generation Tensor Cores Redefine the Future of AI and HPC

First introduced in the NVIDIA Volta™ architecture, NVIDIA Tensor Core technology has brought AI training times down from weeks to hours and provided massive acceleration to inference operations. The third generation of Tensor Cores in the NVIDIA Ampere architecture builds upon these innovations by providing up to 20X more floating operations per second (FLOPS) for AI applications and up to 2.5X more FLOPS for FP64 HPC applications.

NVIDIA HGX A100 4-GPU delivers nearly 80 teraFLOPS of FP64performance for the most demanding HPC workloads. NVIDIA HGX A100 8-GPU provides 5 petaFLOPS of FP16 deep learning compute. Abd the HGX A100 16-GPU configuration achieves a staggering 10 petaFLOPS, creating the world’s most powerful accelerated server platform for AI and HPC.

Applications:

AI, HPC, VDI, machine intelligence, deep learning, machine learning, artificial intelligence, Neural Network, advanced rendering and compute.