[Hardware] Intel Battlemage Arc GPUs Come In Two Variants: Xe2-HPG For Discrete & Xe2-LPG For Integrated, Raja Koduri Says “Making The Next Generation Will Be Much More Easier”

[BirSaNN]

Besides the two major updates to its Arc family, Intel also gave an update on the progress of its next-gen Xe2 GPUs codenamed Battlemage.

Intel Will Have A Smoother Time Making Xe2 Battlemage GPUs, Company Rep Confirms Xe2-HPG "Discrete" & Xe2-LPG "Integrated" In The Works
During a conversation with Hardwareluxx's editor, Andreas Schilling, Intel's Tom Peterson stated that the GPU division had a lot to learn from the first generation Arc GPU codenamed Alchemist. The first generation had a Xe architecture that scaled across the data center, discrete gaming, & integrated graphics solutions. Each sub-segment needs to be optimized according to the platform they are being aimed at and that's a complicated process.

With Intel's upcoming Xe2 architecture, a similar methodology is being applied with two different chip designs, the Xe2-HPG for discrete graphics and the Xe2-LPG for integrated solutions. But the company states that during its 2nd generation, Intel will have less trouble as they did with the 1st gen Alchemist GPUs and they are further planning to streamline down the path when Celestial or Xe3 GPUs hit shelves.

"I wish we had one design, like let's call it Xe 1 and all we had done is sprinkle that around unmodified. That is a winning strategy. What we have actually done unfortunately and we are Improving on this on future generations, we've done targeted micro architectures per vertical segment.So we have an architecture for high end datacenter, we have a slightly different architecture for desktop discrete and today we have a slightly different architecture for integrated.Now As we go forward in our roadmap, we realized this is a very, very expensive – the QA process and the segmantation.The thinking was we needed to differentiate our IP and customize it per each segment.

T think in hindsight we would have been better off have we rigorously said "you know what, we're gonna give up something", like give up some differentiation in the high end or we are going to have some overhead in the low end. We're gonna just have one thing and it goes everywhere unmodified. Thats more the strategy we are looking at going forward. And thats because, thats really the only way to get IP reused to really work.

There is a Xe and there is a Xe 2 and in that Xe 2 generation there is a Xe-LPG and there is a HPG (…) and there a slight variations (…) which is our big learning. The idea was we needed to optimize for each segment and build separate chips and do separate verifications. And I think now the real learning is we would be better off concentrating our focus and really thinking of it like a really solidly, hard IP business.

But it's a tough thing, because if you know that your are going into the data center, you know that you are going desktop discrete and you know you are going be integrated – they all have slightly different ways to optimize. And that's what we've done a much much better job going forward. We are learning to refrain from overly customizing IP because that proliferates QA and verification and really bloats the work to be done."

via HardwareLuxx

Furthermore, during a recent interview with Raja Koduri, the chief architect of Intel's GPU division confirmed that there were certain obstacles when making their first generation Arc GPUs and that those are launched now so making the next generation GPUs, aka Xe2 Battlemage and Xe3 Celestial, will be much easier.

link: https://wccftech.com/intel-battlemage-arc-gpus-come-in-two-variants-xe2-hpg-for-discrete-xe2-lpg-for-integrated-raja-koduri-says-making-the-next-generation-will-be-much-more-easier/