Among the numerous bulletins in right now’s Intel Architecture Day, Intel can be providing a significant replace to their GPU roadmap over the following 24 months. The Xe household, already jam-packed with Xe-LP, Xe-HP, and Xe-HPC components, is now getting a fourth deliberate variant: Xe-HPG. Aimed immediately on the fanatic gamer market, this newest Xe variant might be Intel’s most gaming-focused half but, and the largest step but in Intel’s plans to be extra diversified in its foundry sources.
So what’s Xe-HPG? At a excessive degree, it’s meant to be the lacking piece of the puzzle in Intel’s product stack, providing a high-performance gaming and graphics-focused chip. This is versus Xe-HP, which is specializing in datacenter options like FP64 and multi-tile scalability, and Xe-HPC which is much more esoteric. In that respect, Xe-HPG may be regarded as all the things within the Xe household, distilled down right into a single design to push FLOPs, rays, pixels, and all the things else a robust video card would possibly want.
Like with the remainder of Intel’s forward-looking Xe bulletins, the corporate isn’t providing efficiency projections, options, or the like. But we do have some small particulars on what to anticipate.
First and foremost, past going after the fanatic efficiency area, Intel has confirmed that this half will assist ray tracing. A marquee function of high-end video playing cards, ray tracing will tackle even higher necessary over the approaching years because the soon-to-launch next-generation consoles head out the door with the function as nicely, ultimately reworking it right into a baseline function throughout all gaming platforms. Similarly, ray tracing is a vital part of Microsoft’s DirectX 12 Ultimate customary, which given the timing of this GPU and Intel’s intentions, I’d be shocked if Intel didn’t assist in full.
The chip might be constructed on the muse that’s Xe-LP. However it would additionally pull in applied sciences that Intel is pioneering for Xe-HP, and Xe-HPC. Not the least of which is uncooked scalability, which is with the ability to take the Xe-LP basis and scale it as much as a whole bunch (if not hundreds) of GPU execution models. But Intel can be pulling what they’re calling “compute frequency enhancements” from Xe-HPC, which presumably will enable them to maximise the chip’s total clockspeeds. All informed, I received’t be too stunned if it appears so much like Xe-HP generally, besides with server-driven options like quick FP64 assist and multi-tiling stripped out.
But Xe-HPG will even convey one thing new to the desk for your complete Xe household: GDDR6 assist. Intel is confirming that the chip – or somewhat, the microarchitecture the chip might be based mostly on – might be designed to work with GDRR6. This is versus Xe-HP(C), which as high-end server components use HBM, and Xe-LP, which is designed to be used with extra standard reminiscence varieties. GDDR6 compatibility is a novel want that displays it is a gaming targeted half: GDDR6 supplies the reminiscence bandwidth wanted for high-performance graphics, however with out the stratospheric prices of HBM reminiscence (an issue that has impacted another high-end GPUs over time). In an additional twist, Intel apparently licensed the GDDR controller IP from exterior the corporate, somewhat than growing it in-house; so Xe-HPG can have a really notable little bit of exterior IP in it.
But maybe most attention-grabbing of all for graphics insiders and Intel traders alike is the place Xe-HPG might be constructed: not at Intel. As a part of their Architecture Day roadmap, Intel has confirmed that the half might be made at an exterior fab. In truth it’s the one Xe half the place the GPU (or a minimum of the compute aspect) is being made completely at a third-party fab. Intel after all won’t reveal which fab that is – if it’s TSMC or Samsung –…
