Trademark entry reveals the NVIDIA DGX A100: The first GA100 Ampere-based deep learning workstation
NVIDIA had to cancel its GTC 2020 event owing to the COVID-19 pandemic, but the company would be releasing a pre-recorded keynote on May 14 in which Jensen Huang is expected to delve deep into the upcoming Ampere GPU architecture.
So far, we've been seeing several leaked references to the upcoming Ampere GPUs such as the purported RTX 3080. Now, an entry in the Justia trademark database points to a new NVIDIA deep learning workstation powered by Ampere called the DGX A100.
This is the first time we are seeing a reference to a DGX A100 based on a GA100 GPU indicating that this is indeed an Ampere workstation. The trademark entry does not offer any details pertaining to the workstation or the GPU itself, but Videocardz expects it to feature at least eight Tesla A100 computing units similar to the DGX-1 Volta. A DGX-2-esque system with up to 16 Tesla A100 units could be in the offing as well.
There are several NVIDIA deep learning workstations already available. These include the DGX-2, DGX Station, and DGX-1 Volta based on the Volta architecture alongside the Pascal-based DGX-1 Pascal. The DGX-2 was heralded by NVIDIA as the world's largest GPU given that it features a whopping 16 Tesla V100 units that add up to 81,920 CUDA cores.
From what we gather so far, GA100 would be the monster GPU in the Ampere lineup with up to 8,192 CUDA cores, 256 RT cores, 1,024 tensor cores, and a boost clock of 1,750 MHz. We have also seen evidences for Ampere parts with 7,936 and 7,552 CUDA cores on Geekbench. While the GA100 is expected to be the full deal and would likely feature in the RTX 3080 Ti, other GPUs such as the GA103 and GA104 are expected to power the RTX 3080 and RTX 3060, respectively.