Over the weekend, Intel unveiled a new GPU architecture called Ponte Vecchio for high performance computing (HPC) and artificial intelligence (AI) workloads. That means it’s under the same Xe umbrella that Intel’s individual gaming GPU will occupy, but it’s tied to a data center rather than a home computer.
I will not take a closer look at Ponte Vecchio. That’s because data center GPUs are out of the way of PC gaming. However, there are some noteworthy bits that I want to pass by.
One of them is that Ponte Vecchio is being built with a 7 nanometer manufacturing process. Intel’s struggle for 10nm against CPU is well documented and still waiting for the arrival of 10nm desktop CPUs today (the only 10nm component that arrives at volume from Intel so far was the Ice Lake lineup for ultraportable laptops)
“Ponte Vecchio will be manufactured with Intel’s 7nm technology and will be Intel’s first Xe-based GPU optimized for HPC and AI workloads. Ponte Vecchio leverages Intel’s Foveros 3D and EMIB packaging innovations and includes high bandwidth memory It offers features within several technology packages, Intel said, “Compute Express link interconnect and other intellectual property.”
That’s a fairly advanced GPU architecture. Foveros is a 3D chip packaging technology Intel introduced at the time, as you will remember in a December press release. The reason Intel’s implementation is unique to other 3D stacking technologies is that it applies to logic chips that are first applied to logic chips.
Another interesting point here is how Intel classifies individual GPUs. In a keynote on Sunday, Intel’s graphic boss Raja Koduri said that Intel, as reported by our friends, will have one Xenon (Xe) structure spanning the various product segments under which the various substructures become AnandTech.
In this sense, Ponte Vecchio will inevitably share characteristics with everything that the individual game GPU Intel ultimately reveals. At this stage, I don’t know how much two DNA can share, but 3D chip stacking is possible.
This approach also provides some flexibility when fine-tuning different GPUs depending on the market. Instead of activating or deactivating certain parts to release the amount of a single GPU in other markets, Intel seems to be based on a unique design for other markets based on the generic architecture (Xe).
Ponte Vecchio will debut on the Aurora Supercomputer at Argonne National Laboratory in 2021. Intel’s first discrete Xe GPU will be released in 2020, but anyone can guess if it’s for games, servers, and so on. Either way, Black Friday doesn’t expect a deal on Intel GPUs this year.