Introducing Ponte Vecchio, the first commercial product based on the architectureXe The innovative - which will be adapted to all levels from server arrays to ultra-economical mobile products
A few years have passed since Intel announced that it plans to try its luck, and not for the first time, in parallel processing, or graphical processing if desired, thus providing a first significant competition for the Duopoly of NVIDIA and-AMD For the first time in a long time - and now this vision feels much more real than ever with the unveiling of the technical details of a first product based on New Age architecture. ships and-Turing Get ready to meet the Xe.
Ponte Vecchio is the code name for parallel home processing accelerator Intel Which will hit the market in 2021 and will fit the world of servers and especially the world of powerful computing (HPC) of the formidable supercomputers - based on the future production lithograph of Intel In 7 nm, for the intriguing Foveros technology that will stack several different types and uses of chips over the chassis chip Single, on fast internal EMIB links between chassis chips and a new Xe Link connectivity technology for linking different accelerators based on the CXL standard and infrastructure PCI-Express 5.0.
From the official slides in the company's presentations, the chipmaker seems to be aiming for a more resilient architecture than we know NVIDIA and-AMD Today - there will not be a basic processing cluster that is the same in all products with a defined number of processing units, cache and timing units, but different chiplets that are adapted to different performance and connected to separate cache chips as required in each field. More similar to the Zen 2 general processor architecture of AMD, Than other products in the field.
Ponte Vecchio probably won't be the first product based on the architectureXe In the market (we are supposed to see its applications in the home market as early as next year), but it is the first to get some technical detail right now - and also a testament to the power and designed flexibility of the technology that can also fit breakthrough Exascale systems.
Intel has not released data on expected performance from Ponte Vecchio units, so it's hard to compare the venture to products we now know from competitors - but data it presented on the expected array for the mighty Aurora supercomputer to be operated in the U.S. Department of Energy's National Laboratory in 2021 allowed some Very crude calculations which give us a particularly promising anchor point.
With the promise of groundbreaking performance at the ExaFLOPs level (quintillion floating point calculations per second) based on approximately 15,000 Ponte Vecchio accelerators, an estimate of performance on the order of about 60 TeraFLOPS (approximately 60 trillion floating point calculations per second) is obtained for each unit of response. Such a standard 32-bit accuracy - this is compared to the performance that reaches a level of about 15 TeraFLOPS from similar leading accelerators of NVIDIA and-AMD Right now.
It's a huge leap to be treated with healthy skepticism (the computer in question hasn't been built after all, and a lot of things can change in a year and a half or two in the world of computing and hardware), but this is a first testament to the fact that Xe's X Intel Are supposed to be competitive in relation to the competitors' products already in the first generation - seemingly without the need for several years of trial and error until they manage to catch up with years of gaps.
Expect more details and data about Xe And various products that will be based on it in the near future!