INTEL ENTERS DISCRETE GPU MARKET

THE WORLD OF CONSUMER GPUS has been a contest between AMD and Nvidia for so long, it’s hard to remember that the market used to be crowded with contenders. That will change—Intel has been working on its own discrete graphics card tech, the Xe Graphics Architecture, for some time. Its first iteration, code-named Ponte Vecchio, is a 7nm chip (Intel’s first). It uses a highspeed Compute Express Link (CXL) and Intel’s 3D Foveros system, which enables chips to be built in layers. It has also announced its first card featuring Ponte Vecchio, which Intel claims is an exascale GPU (capable of a billion, billion FLOPs).

Before we get carried away, Intel is aiming its initial efforts at the very top of the market; this is a server card for heavy HPC and AI workloads. And to make good on that exascale promise requires multiple cards working together. There are some mouth-watering high-end technologies at play. The CXL interface requires PCIe 5.0, for example; we’ve barely got PCIe 4.0 on desktops. Intel has released some baffling material on Xe, but what is clear is that it has been engineered to be highly scalable, and interconnected at all levels.

The consumer version of the Xe Graphics Architecture won’t arrive until next year, and will probably be built on a 10nm process. Performance will be in the more realistic teraFLOP department. The Xe family will eventually come in three main brands: the high-end Xe-HPC, Xe-HP for data centers and professional workstations, and Xe-LP for consumers. Intel has made plans for virtually every market segment, right down to laptops and integrated graphics. This is a massive project, in time and money, and it’s clear Intel is serious about making a concerted grab for the GPU market. It’s about time we had another horse in the race.