Nvidia has announced more details about the DGX A100 system and its new generation of Ampere GPUs. It appears that the company has already registered the brand for its new HPC systems according to recent sources.
Nvidia DGX A100: the next generation of Ampere GPUs for HPC
The specific name for the system is DGX A100, which gives some hints about its basics. Designed specifically for the advanced and HPC user community, this model provides supercomputing capabilities in a desktop workstation format.
Nvidia had already released similar DGX formats based on its Pascal and Volta GPUs. The Volta DGX production line was a simplified design that offered more options to HPC (high-performance computing) users. Several variants were designed with a total variability from four Tesla V100 GPUs to models with sixteen units. The new Nvidia Ampere DGX A100 system (clearly based on GA100 GPU chips) would feature one of the 128 SM flagship configurations. Initially, units with 8-tesla GPUs would be introduced, while more complex formats would be introduced gradually.
The current filtered data of the Nvidia Ampere based DGX A100 system indicates 30 processing TFLOPs. As a reference, the current DGX 2 system has Intel’s Xeon platinum CPU (14 nm) and Skylake architecture, 1.5 TB memory, and NV switches.
It is not yet known whether Nvidia will offer users alternatives between an Intel Xeon or AMD EPYC system. If so, this would be an opportunity for the company to offer high-performance computer systems with both Intel and AMD processors.
Via: Videocardz