In another 10 days, on May 14, local time, NVIDIA will release Huang Renxun’s original keynote speech at the GTC 2020 conference, with the highlight of the announcement of the next generation of Ampere GPU architecture, which has been strongly implied in NVIDIA’s official press release. Now, the iron evidence of the mountain has finally appeared, and it is from NVIDIA itself.
According to justia, a trademark search site, NVIDIA quietly registered the “DGX A100” on March 30 this year, corresponding to the next-generation deep learning workstation and AI supercomputer based on the Ampere architecture, ga100 core Tesla A100 computing card.
The NVIDIA DGX system has gone through several generations of products, initially the Tesla GP100 based on pascal Pascal architecture, considered NVIDIA’s first true computing processor, and later the Tesla V100 with the Volta Volt architecture, a system that can contain 4-16 GPU processors.
NVIDIA previously used DGX-1, DGX-2 and other product models, this time did not renew the DGX-3, but the DGX A100 name directly exposed its identity.
Details of the DGX A100 system are unclear, but there is speculation that it will contain at least eight Tesla A100 computing cards of unprecedented scale and performance.
It is important to note, however, that the previous Volta architecture was limited to computing cards, while the architecture for game cards was Turing, a generation that could also continue the dual architecture strategy, with Ampere for computing and Hope for games.