Intel is expected to launch its first dedicated GPU lineup (since Larabee) in mid-2020, codenamed Arctic Sound with the Xe branding. The announcement should come at Computex 2020, with the first wave of Xe cards based on Intel’s 10nm node.
Intel hasn’t stopped there though. The company has already planned the second and maybe even third generation of Xe graphics cards. According to Videocardz, the second generation Xe cards will include a part called “Ponte Vecchio” for the HPC space. Basically imagine hundreds if not thousands of GPUs working in tandem to create a Skynet like scene.
Exascale computing refers to computing systems capable of at least one exaFLOPS, or a billion billion (i.e. a quintillion) calculations per second. Such capacity represents a thousandfold increase over the first petascale computer that came into operation in 2008.Wikipedia
“Ponte Vecchio” is an old stone bridge in Florence, Italy. Like AMD, Intel seems to have started codenaming its exascale products after Italian historical monuments or cities. In case you’re wondering why these GPUs are named after a bridge: well Ponte Vecchio will leverage Intel’s CXL (Compute Express Link) to synchronize all the graphics processors.
Keep in mind that CXL and most likely Ponte Vecchio won’t be used in gaming GPUs and will largely be limited to Data Centers and workstations. For now, DX12’s multi-GPU API will have to do regardless of how efficient or inefficient it may be.
According to reports, PV will be a part of a secret project at Intel called “Aurora”. The first details are expected to come on the 17th of November. It will supposedly include:
- Sapphire Rapids Scalable Xeon CPUs (7nm Goldencove core)
- Ponte Vecchio GPUs (also based on the 7nm node)
- Intel’s OneAPI (for GPUs?)
Aurora is expected to see the light of the day sometime in 2021. Furthermore, Vecchio is rumored to use the Foveros 3D packaging technology in addition to the CXL interconnect. And more importantly, it has also been confirmed that the Xe graphics cards will use HBM memory, something Raja Koduri inadvertently disclosed while talking to a Russian channel.