The first trace of NVIDIA’s upcoming next-generation Big GPU.
NVIDIA Ampere GA100
NVIDIA deep learning workstations, known as DGX systems, have gone through few iterations. The first system launched with Pascal GP100 GPU considered the first true computing processor from NVIDIA. The company later introduced four Volta systems featuring from 4 to 16 Tesla V100 processors.
On May 14th, NVIDIA will release a keynote, originally meant to be shown during GTC 2020. However, the situation caused by the novel coronavirus outbreak has forced NVIDIA to postpone, shift and finally cancel the physical GTC event. The keynote to be released will be a prerecorded video, likely without any audience.
The keynote is expected to bring a lot of news regarding the upcoming Ampere architecture. Thanks to the latest leak, we are confident that Huang will talk about Ampere.
A new entry has been discovered in Justia trademark database, where NVIDIA has filed an application for the DGX A100 trademark. NVIDIA has not used such a product name before, so the DGX-1, and DGX-2 will likely not succeeded by DGX-3 but DGX A100 instead. Luckily for us, the new name format basically reveals the name of the processor.
The “A100” directly refers to Tesla A100, the computing processor based on GA100 GPU. None of these Ampere devices have been confirmed by NVIDIA. Details on the new DGX are scarce, but we expect at least 8 Tesla A100 computing processors in the system. The transistor count per GPU will likely increase due to a smaller fabrication node (expected to be either 8nm or 7nm).