On January 17, a Twitter user 'KittyCorgi' shares initial information
about two GPUs: Ampere GA103 and GA104. The former GPU is supposedly associated with a higher-end GPU while the latter is likely for RTX 3070 series.
According to the rumor, GA103 GPU would supposedly feature 60 SMs (Streaming Multiprocessors 'Ampere') while GA104 would end up with 48. The weird naming convention (103) is something we do no fully understand, NVIDIA does not use uneven numbers for high-end GPUs.
The original source
also shared die diagrams for GA103 and GA104, they look professional but they are not as detailed as Turing diagrams, hence we put a strong doubt on their credibility.
If this data is to be believed then RTX 3080 could get GA103 GPU, while GA104 would end up in RTX 3070. This is the same approach as we had for the original Turing series where RTX 2080 received the TU104 processor and 2070 was based on smaller TU106 silicon. Of course, this has changed for the SUPER series where both SKUs feature TU104. The rumor also mentions a 320-bit bus for RTX 3080, which would allow 10/20GB memory configuration. This is also very uncommon for NVIDIA and probably just as hard to believe.
NVIDIA has never confirmed Ampere as their next-generation architecture for the GeForce series. It is only present in multiple publications and leaks from foundries. For now, all we know is that the next-architecture could also be called Hopper. We actually expect NVIDIA CEO Jensen Huang to present a new roadmap at GTC 2020, which starts on March 22nd.