Recent rumours have indicated that following Ampere, Nvidia's next GPU architecture will be named Lovelace. Now, more details about the upcoming flagship AD102 GPU are starting to surface, with new rumours indicating some of the specifications.
Leaker @kopite7kimi shared the expected structure of the AD102 GPU, which will allegedly come with a “12×6” structure. Based on this structure, 3DCenter calculated that this GPU might come with 12x GPCs (graphics processing clusters), 72x (12×6) TPCs (texture processing units), and 144x SMs (streaming processors). This translates to a maximum of 18432 CUDA cores and 66.4 TFLOPs at 1.8GHz, 71% more than the GA102 GPU used on the Nvidia Tesla A40 graphics card. The memory configuration and process node details are still unknown.
So, nVidia's AD102 chip maybe is like:
18'432 FP32 units
~66 TFlops FP32 power (on 1.8 GHz) https://t.co/A8OnUktE1s
— 3DCenter.org (@3DCenter_org) December 28, 2020
For now, we know that Nvidia is working on increasing the number of SKUs on the RTX 30 series by launching the RTX 3080Ti, the RTX 3070Ti, and the RTX 3060 12GB/6GB in early 2021. The Lovelace GPU architecture is expected to come at a later date, after a possible ‘Super' refresh of the RTX 30 series.
Following the nomenclature that Nvidia has been using in the latest generations of graphics cards, the Lovelace GPUs will most likely go on to be known as the RTX 40 series. There are no details about a release date for these graphics cards.
KitGuru says: Next-gen Nvidia GPU architecture is likely still a long way from launch, but these early, rumoured specifications are interesting to look at. Are any of you thinking about skipping the current generation and waiting for the next big GPU launch? Or are you still set on trying to obtain an RTX 30 series or RX 6000 series GPU as soon as the stock situation clears up?