November 18, 2019 | 13:25
At the Supercomputing 2019 conference in Colorado, Intel has named a new GPU based on the forthcoming Xe architecture: Ponte Vecchio - but don't expect to see results at home any time soon.
Named after a bridge in Florence, Ponte Vecchio is the first Xe GPU for HPC and AI. It won't actually launch until 2021, when it will debut as part of the Aurora Supercomputer for the US Department of Energy, but it's still fascinating for now as an insight into what's to come.
That's because it's part of Intel's first attempt at discrete graphics computing since the Larrabee project over 10 years ago. Ponte Vecchio will be manufactured on Intel's own 7nm process and is aimed at general-purpose computing workloads with this model specifically designed with High Performance Computing (HPC) and AI in mind. Intel referred to it as its first exascale GPU.
Ari Rauch, VP of architecture, graphics and software at Intel announced that the firm wasn't disclosing all the details of the construction of Ponte Vecchio just yet, but some information was provided. It'll utilise Intel's 3D packaging technology, Foveros, along with a new interconnect called Xe Link. That's based off the standards for CPU-to-Device and CPU-to-Memory interconnects laid out by the Intel-founded CXL (Compute eXpress Link) consortium.
While that might not make sense to all, it's the same group that both Nvidia and AMD have since joined, and that's a pivotal consortium to be a part of. It seems highly likely that this interconnect is going to be crucial when it comes to HPC performance required by supercomputers in the coming years.
Rauch summed it up simply for the average user as 'you can assume that this device takes advantage of all the latest and greatest technology from Intel,' although obviously he's in the business of making sure that you're overly excited about something that's not even that near to release yet.
For now, Ponte Vecchio is solidly aimed at the HPC market; Intel's plans for the consumer GPU market remain largely a mystery. That said, Intel has confirmed its full graphics portfolio will be utilising a single architecture (Xe with different market-specific configurations) and a single API (oneAPI), which means the underlying tech behind Ponte Vecchio will be the same as those intended for gaming systems. That is, once it's been sufficiently tweaked for such needs.
Sadly, the question of when we'll be seeing a third name in the gaming GPU market remains entirely unanswered. Competition is always a useful thing for consumers, though, and even this HPC-focussed reveal brings Intel a step closer to competing with Nvidia and AMD in the world of frame rates. For now though, anything that means more powerful supercomputers has to be good news for the future of computing.
February 27 2020 | 11:00