"Nearly a year ago, news broke that AMD and Intel might collaborate on a GPU. Both companies wasted no time denying the story, with Intel firmly slamming its foot down on the “No, we aren’t,” side of the business. And, of course, that means they absolutely are.
Intel has announced the new GPU “brings together our high-performing Intel Core H-series processor, second generation High Bandwidth Memory (HBM2) and a custom-to-Intel third-party discrete graphics chip from AMD’s Radeon Technologies Group – all in a single processor package.”
This is significant for several reasons. First, to the best of our knowledge, Ryzen APUs are DDR4 parts, with no support for HBM2 in first-generation hardware. Launching an APU with Vega on Intel graphics and HBM2 virtually guarantees this new part will outperform AMD’s own APUs in-market. It’s a significant change in the relationship between Intel and AMD.
But therein lies the rub. We’ve talked before about how integrated graphics are fundamentally bottlenecked by bandwidth and need a solution like HBM2 to drive better desktop experiences. AMD, however, hasn’t been able to afford to bring these capabilities to market in its own APUs. Instead, first-generation Raven Ridge (seen in the slideshow below) is a conventional DDR4 solution."
"Granted, AMD could still claim a significant performance improvement if it certifies DDR4-3200 as opposed to older DDR4-2133, but this kind of oomph may not work well in mobile, where the goal is to reduce power consumption, not increase it. Intel is going to have the faster solution, thanks to HBM2. These new hybrid Intel-AMD cores will use Intel’s new multi-die connection technology, dubbed EMIB, as opposed to the interposer used in conventional HBM2 stacks that we covered EMIB earlier this year."
"There are specifics we don’t know yet, like how the technologies from AMD (which uses Infinity Fabric) mesh together with EMIB, or what the power consumption figures will look like. This isn’t a deal to drive performance in smaller form factors; H-class CPUs from Intel are typically rated for up to 45W of performance. But this does suggest Intel wants to launch a set of SKUs that don’t need external graphics at all, possibly replacing its own lineup of Crystal Well-derived graphics SKUs.
As for AMD, it may well be the company stands to make more money on its Intel collaboration than on its own APUs on a sale-for-sale basis; AMD may own the entire APU “stack,” but it also has to sell these chips for less money. Alternately, AMD may view this as an excellent way to boost its mindshare and later launch products to take advantage of the same capability. There’s a lot we don’t know yet, but it’s interesting to see the rumor mill was right on this one."