DUG is thrilled and honoured to be a NVIDIA pre-release partner for their brand-new A100 80 GB PCIe GPU, announced last month at the International Supercomputing Conference (ISC) 2021!
Our HPC experts had their hands on the product and they were very impressed with it:
“NVIDIA’s A100 80 GB PCIe will deliver a step change in raw performance for our research and production GPU solutions while its robust tooling aligns perfectly with our agile, responsive approach to emerging client needs,” said Dr Stuart Midgley, the Chief Information Officer of DUG Technology.
Major performance enhancements for AI and HPC.
The latest update to the A100 accelerator delivers unprecedented acceleration to power the world’s high-performing data centres for artificial intelligence (AI), data analytics, model training, simulation challenges and high-performance computing (HPC) applications.
Based on NVIDIA’s Ampere GA100 architecture, the NVIDIA A100 80 GB card – manufactured with TSMC’s 7 nm process – houses an insane 54 billion transistors. The card is a dual-slot 10.5 inch PCIe Gen4 card and debuts the world’s highest PCIe card memory bandwidth up to 1.94 terabytes per second (a 25 per cent increase from its 40 GB counterpart), with 80 GB of HBM2e high-bandwidth memory.
The A100 80 GB PCIe GPU’s gigantic memory capacity and high-memory bandwidth enable massive amounts of data and more advanced neural networks (NN) to be held in memory, reducing inter-node communication and overall energy consumption.
With NVIDIA’s new kit, users who depend on HPC for their work will be able to achieve higher throughput and faster results, speeding time to solution for the largest models and most massive data sets.