At the end of last year, Intel announced that it will launch its new high-performance GPUs “Ponte Vecchio” based on Xe2 at 7 nm in 2021, and it seems that they are already starting to test them according to a new EEC registration.
According to this registration, Intel is testing three “Ponte Vecchio RVP Pre-Alpha” cards, where RVP stands for the Reference Validation Platform. This means that this is a very early prototype that is only trying to validate the operation of the GPU, although neither the PCB nor the cooling system uses the final design.
Unfortunately, the listing of these three models doesn’t tell us much because nobody knows how to crack their GAPV3KI2TC, GAPV4KI2TC and GAPV5KE2TC codes yet. However, this shows us that Intel is working on not just one, but at least three variants for release.
Things we know about Ponte Vecchio so far
These GPUs will have a multi-die design so that up to 16 DIEs per GPU can be connected via Xe Link, a CXL-based technology. They will also have what Intel calls “Rambo Cache”, a huge cache block that all GPUs can share for better performance, and what Intel calls XEMF (Xe Memory Fabric), which uses the same concept but with VRAM.
These GPUs will be used for the first time on the Aurora supercomputer, which will be equipped with 6 of these GPUs up to 16 DIEs, as well as two next-generation Xeon “Sapphire Rapids”, which will achieve maximum performance through special CPU-GPU connections.