Intel and AMD partner up on their first-ever APU

Hell has frozen over, ladies and gents. Intel and AMD have gotten in bed together and announced a collaboration. They’ll work on a combined Intel CPU and AMD GPU package that’ll be sold to notebook vendors who want to spend less time sourcing quality components. There’s a lot of confusion surrounding this new chip, so let’s get to demystifying it.

The announcement took pretty much everyone by surprise. After months of leaks on Geekbench and other hardware leaderboards, no-one was quite prepared for the reality of what we saw in those leaks – that Intel had replaced their own internal graphics chip with something much more powerful from AMD. For years, Intel had a deadline looming ahead of them concerning a deal made with NVIDIA back in 2011 – that both companies would cross-license technology for integrated graphics and interconnect, all so that Intel could make integrated graphics chips on their own without stepping on NVIDIA’s toes legally. That agreement ended on 1 April 2017.

Following that date, Intel is unable to make a new IGP design without getting themselves into a quagmire they can’t wriggle out of. Processors still to come like Cannonlake and beyond were in the design stages before the agreement ended, but anything newer than that might face some problems. In fact, it’s likely that the bulk of the problems lie with their high-performance GT2 graphics, codenamed Iris/Iris Pro. Those showed serious potential against NVIDIA’s entry-level GPUs, but it was probably based on more cross-licensed technology than Intel would like to admit. Here’s what Intel says about their new “processors”:

“The new product, which will be part of our 8th Gen Intel Core family, 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.”

Notice that they specifically mention their processor from the H-series quad-core family, which are high-end, high-performing CPUs with integrated graphics and a 45W thermal limit. Nothing is mentioned about their integrated Iris graphics. That’s replaced by a custom-designed AMD Radeon RX graphics card hooked up to a single stack of high-bandwidth memory (HBM). It’s the holy grail of all-in-one solutions. It just isn’t being sold by AMD.

I suppose calling it an APU isn’t fair to AMD though. This is just a multi-chip solution on a substrate, almost like a miniature daughterboard that connects to the motherboard on the laptop itself. Intel gets to purchase the GPU from AMD and the HBM elsewhere, working around any supply issues that might limit them buying everything from AMD. The “custom” part of the GPU is interesting, though. Because Intel’s chips don’t use AMD’s Infinity Fabric, and aren’t compatible with it, AMD had to make alterations to their design to have the GPU connected through PCI Express, in addition to using Intel’s power management solutions to get everything to stay under a reasonable thermal limit.

The switch from using GDDR5 memory or shared DDR4 also goes out the window thanks to the inclusion of HBM. That memory is effectively on-die, but is not accessible to the CPU, which means that Intel can’t use it as a fourth-level cache, which they used to do in the past. This means there’s a cost saving (in addition to the power savings) compared to GDDR5. so laptops which have this “chip” will also see longer battery life than a version with everything spread out and using older RAM. It might be easier to cool as well.

Who would want this? The obvious answer would be Apple, who’d be the biggest customer on the block that might want an integrated solution like this. It could be integrated into their Macbook Pro lineup, and if there’s enough SKUs available to include it, they might put this kind of thing into their Macbook Air family as well. Apple’s preference for AMD’s GPUs might have even prompted this product into existence, solely because their business is important to Intel.

Other laptop vendors who previously would’ve looked at integrating a Core-H series processor and NVIDIA’s GeForce MX150 GPU might also have to pause and think for a bit. This solution would be easier and cheaper to design a cooler for, and since the CPU and GPU share the same power budget, it’s easier to plan for as well.

There’s also something to be said for extra battery life. Because the motherboard can shrink thanks to reducing the number of components inside, there’s more space for a bigger battery. We don’t use DVD drives anymore, so that’s even more space available inside thin-and-light designs that are just outside of what would qualify as an Ultrabook. Specs-wise, this will be a big bump for notebooks in this category. Not only are you getting up to four cores with hyper-threading, we’re potentially also getting 4GB of HBM in a single stack alongside up to 1,408 shaders. That’s about twice the power of an Xbox One in a package that’s only a little bigger.

Make no mistake, this is going to be a game-changer. Intel and AMD say that products including this multi-chip solution should make their debut at CES 2018.

Source: Intel