Earlier this week, Intel confirmed it would work with AMD to develop a new GPU for use in its NUC (Next Unit of Computing) systems. The news sent waves through the tech community, both because it had been previously rumored (and specifically denied), by Intel and because it’s the first such collaborative product effort between AMD and Intel in… well, basically ever. The two companies may work together in joint efforts to write standards or as members of other tech organizations, but they haven’t jointly announced collaborative products in decades.
Thanks to an earlier leaked roadmap and an image from Chiphell, we can now start putting things together on what the new NUC will look like and be capable of. First, here’s the leaked photo from Chiphell:
A few thoughts, in no particular order: The CPU is going to be the top left block (the dark patch could be burn damage), while the GPU and a single stack of HBM2 is at the other end of the package. What kind of performance can we expect from that kind of configuration?
This slide is a few years old, but it illustrates the HBM2 product stack fairly well. HBM2 supports up to 8GB of memory per stack, at capacities ranging from 2-8GB and 128GB/s – 256GB/s of bandwidth per stack. A comparable dual-channel APU or Intel on-die GPU using DDR4-3200 would offer 51.2GB/s of memory bandwidth per stack, which means this new GPU will decisively outpace the old–the only question is by how much.
According to an earlier leaked roadmap courtesy of PC Perspective, Intel is planning three separate Hades Canyon SKUs with a 46W, 66W, and 96W TDP:
One of the two Hades Canyon devices shown on the left roadmap is labeled as “Hades Canyon VR,” but the three potential SKUs on the right only differ in their TDPs. If we assume that the one labeled chip is the already-launched Core i7-6770HQ, that leaves a 66W TDP and a 96W TDP still to populate, and the difference is likely to come down to the GPU. And a GPU in a 96W combined form factor may well be capable of at least basic VR, though we’ll have to see how performance looks to determine that.
It’s not hard to see how this would play. Intel can offer different Vega configurations to hit different performance targets by adjusting the GPU and HBM clock speeds. It’s been several years since we’ve speculated on how HBM2 could deliver the APU performance AMD has long promised–I admit, when we started covering the topic we didn’t think it would arrive in an Intel SKU, but this program should yield dividends for both companies.