Among the many announcements made during Intel Architecture Day today, Intel will also provide a major update to their GPU roadmap over the next 24 months. The Xe family, which is already packed with the Xe-LP, Xe-HP and Xe-HPC divisions, is currently having a planned fourth variant: Xe-HPG. Aimed directly at the enthusiast gamer market, this latest Vehicle variant will be Intel’s most gaming-focused division and the biggest step forward in Intel’s plan to diversify its foundry sources.
So what is Xe-HPG? At a high level, it means that the missing piece of the puzzle is in Intel’s product stack, providing a high-performance, graphics-focused gaming chip. This is in contrast to Xe-HP, which specializes in data center features like FP64 and multi-cell scalability, and the Xe-HPC is even more esoteric. In that respect, the Xe-HPG could be considered as everything in the Xe family, distilled into a single design to push FLOP, rays, pixels, and everything else a powerful video card might need.
As with the rest of Intel’s Forward-looking Vehicle announcements, the company makes no predictions about performance, features, or the like. But we have some small details about what to expect.
First and foremost, in addition to pursuing the performance space for enthusiasts, Intel has confirmed that this section will support ray tracing. A special feature of high-end video cards, ray tracing will be even more important in the years to come as upcoming next-generation consoles also have this feature, eventually turning it into a feature. Basic functionality on all game platforms. Likewise, ray tracing is a key component of Microsoft’s DirectX 12 Ultimate standard, given the timing of this GPU and Intel’s intentions I’d be shocked if Intel didn’t fully support it.
The chip will be built on the foundation of Xe-LP. However, it will also entail the technologies that Intel is pioneering for Xe-HP and Xe-HPC. Equally important among them is raw scalability, which can take the Xe-LP platform and scale up to hundreds (if not thousands) of GPU execution units. But Intel is also pulling what it calls “compute frequency enhancement” from Xe-HPC, which will probably allow them to maximize the overall clock speed of the chip. All told, I wouldn’t be too surprised if it looked a lot like Xe-HP in general, except that server-oriented features like fast and multi-layered FP64 support were removed.
But Xe-HPG will also bring something new to the entire Xe family: GDDR6 support. Intel is confirming that the chip – or more accurately, the microarchitecture the chip will be based on – will be designed to work with the GDRR6. This is in contrast to Xe-HP (C), as high-end server parts use HBM and Xe-LP, which are designed for use with a wider range of common memory types. GDDR6 compatibility is a unique need that reflects this as a gaming-focused part: GDDR6 provides the memory bandwidth needed for high-performance graphics but without the staggering overhead of HBM memory (an issue that has affected a number of other high-end GPUs over the years). On the other hand, Intel clearly licensed the GDDR controller IP from outside the company, instead of developing it internally; So the Xe-HPG will have a very noticeable bit of external IP in it.
But perhaps most interesting to those in the graphics industry as well as for Intel investors is where Xe-HPG will be built: not at Intel. As part of their Architecture Day roadmap, Intel has confirmed that this will be done at an external facility. In fact, it’s the only Vehicle part where the GPU (or at least the compute element) is made entirely of a third party. Intel of course won’t reveal which this is empty – if it’s TSMC or Samsung – but that means we’ll see a complete Intel GPU built in another company. If nothing else, this would make it a lot easier to compare the Xe-HPG with its rivals AMD and NVIDIA, since Intel will be using the same fab resources.
Looking at the same roadmap, it should be pointed out that Intel will not be using any of their advanced packaging technologies for this section. Since they don’t use HBM and they don’t do multilayer, there’s no need for things like EMIB, never mind Foveros. There’s still a lot left unknown with the cost side of Intel’s advanced packaging technologies, so removing it from Xe-HPG will probably help keep costs under control in a very competitive market.
And that’s the scoop about Xe-HPG. The newest and most game-focused member of Intel’s Vehicle GPU product stack is set to launch in 2021 – and when Intel wants to break into the broader GPU market, I have no doubts. That this won’t be the last time we’ll ever hear about it ever since.