Earlier this week, Intel confirmed it might work with AMD to develop a brand new GPU to be used in its NUC (Next Unit of Computing) programs. The information despatched waves by the tech neighborhood, each as a result of it had been beforehand rumored (and particularly denied), by Intel and since it’s the primary such collaborative product effort between AMD and Intel in… effectively, mainly ever. The two firms may go collectively in joint efforts to jot down requirements or as members of different tech organizations, however they haven’t collectively introduced collaborative merchandise in a long time.
Thanks to an earlier leaked roadmap and a picture from Chiphell, we will now begin placing issues collectively on what the brand new NUC will appear to be and be able to. First, right here’s the leaked picture from Chiphell:
A number of ideas, in no explicit order: The CPU goes to be the highest left block (the darkish patch may very well be burn injury), whereas the GPU and a single stack of HBM2 is on the different finish of the package deal. What sort of efficiency can we count on from that sort of configuration?
This slide is a couple of years outdated, however it illustrates the HBM2 product stack pretty effectively. HBM2 helps as much as 8GB of reminiscence per stack, at capacities starting from 2-8GB and 128GB/s – 256GB/s of bandwidth per stack. A comparable dual-channel APU or Intel on-die GPU utilizing DDR4-3200 would provide 51.2GB/s of reminiscence bandwidth per stack, which implies this new GPU will decisively outpace the outdated–the one query is by how a lot.
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 2 Hades Canyon units proven on the left roadmap is labeled as “Hades Canyon VR,” however the three potential SKUs on the fitting solely differ of their TDPs. If we badume that the one labeled chip is the already-launched Core i7-6770HQ, that leaves a 66W TDP and a 96W TDP nonetheless to populate, and the distinction is prone to come all the way down to the GPU. And a GPU in a 96W mixed kind issue might be able to at the very least fundamental VR, although we’ll should see how efficiency appears to find out that.
It’s not laborious to see how this could play. Intel can provide totally different Vega configurations to hit totally different efficiency targets by adjusting the GPU and HBM clock speeds. It’s been a number of years since we’ve speculated on how HBM2 may ship the APU efficiency AMD has lengthy promised–I admit, once we began masking the subject we didn’t suppose it might arrive in an Intel SKU, however this program ought to yield dividends for each firms.
*No photos of Hades Canyon have but been made accessible; our characteristic picture is an earlier mannequin.