Intel Kaby Lake + AMD Radeon product *spin-off*

  • Thread starter Deleted member 13524
  • Start date
It sure as hell would be risky though.
Bold as hell idea! :p

Risky would not even get close, tho. Such a setup would risk having a lot of wonky performance outliers and gotchas, and wear on the flash could also end up being considerable. With AMD's traditionally weak software, its platform drivers would be a crapshoot where performance and stability is concerned, not to mention any *ahem* sort-of indirectly promised yet loudly tooted performance-enhancing and/or power-saving features which may never turn out to be real... ;)
 
Bold as hell idea! :p

Risky would not even get close, tho. Such a setup would risk having a lot of wonky performance outliers and gotchas, and wear on the flash could also end up being considerable. With AMD's traditionally weak software, its platform drivers would be a crapshoot where performance and stability is concerned, not to mention any *ahem* sort-of indirectly promised yet loudly tooted performance-enhancing and/or power-saving features which may never turn out to be real... ;)
It wouldn't be any different from the consoles with all system memory GDDR based. Wear on SSDs shouldn't be any different than any other platform with the exception of running out of memory and paging. If anything I'd expect better performance from CPUs having all that bandwidth readily available. Even if limited in capacity, most desktops don't require that much memory. NUCs commonly ship with 4-8GB and a single stack of HBM can manage that while providing what is likely an order of magnitude more bandwidth.

With all the always connected mobile designs it seems obvious where they are going here.
 
Except software for consoles are carefully tailored for the amount of memory available; PCs don't work that way at all as you're well aware yourself. :p
Same concern exists with any system that runs out of memory though. Even 8GB would be more than many desktops require for simple gaming and currently contain. Even if paging to NVMe, the HBM should take the brunt of the ware. It's also the writing that would be an issue, which will exclude nearly all typical assets for a game.
 
Hello 2018 Mac Mini.
One can only hope, Mac Minis have been very anemic for a long time now. Cooling well over 100W (including VRMs, PSU, plus all the other junk in there) in a package as small as the Mac Mini would be hella challenging though, unless they throttle the processors that is...
 
It could easily be done in a smaller new Mac Pro like form factor.

The Mac Mini desperately needs to return to quad-core (with SMT) processors again.
 
Hello 2018 Mac Mini.

Apple is using 28W CPUs in the Mac Mini. They're not going to upgrade their $500 NUC with a GPU+CPU package that consumes at least 2x as much and probably costs as much as the whole box..

RR Ryzen 7 2700U for the next Mac Mini ftw!

But it won’t happen.

Well apple does seem to be pushing Radeon GPUs everywhere they can, and the lack of Raven Ridge solutions in PC OEMs could mean something.
Raven Ridge's cores do have a bit higher IPC than desktop Ryzen and at least in synthetics they seem to be practically on par with Skylake solutions, meaning it would always be a massive improvement over what they're using right now, which are 28W Haswell dual-cores with DDR3 and Iris GPUs without EDRAM.
 
Apple is using 28W CPUs in the Mac Mini. They're not going to upgrade their $500 NUC with a GPU+CPU package that consumes at least 2x as much and probably costs as much as the whole box..



Well apple does seem to be pushing Radeon GPUs everywhere they can, and the lack of Raven Ridge solutions in PC OEMs could mean something.
Raven Ridge's cores do have a bit higher IPC than desktop Ryzen and at least in synthetics they seem to be practically on par with Skylake solutions, meaning it would always be a massive improvement over what they're using right now, which are 28W Haswell dual-cores with DDR3 and Iris GPUs without EDRAM.
Would the Ryzen 2200G and 2400G be a possibility in their 35W variant spec?
If the slides were not fake about 65W and 35W, raises what is AMD's intention with the "mobile" 35W Vega APU and when available.

Separately worth remembering Apple is doing its best to ignore and exclude any official solution/product involving Nvidia, so they would push Radeon everywhere they can.
 
Apple is using 28W CPUs in the Mac Mini. They're not going to upgrade their $500 NUC with a GPU+CPU package that consumes at least 2x as much and probably costs as much as the whole box..



Well apple does seem to be pushing Radeon GPUs everywhere they can, and the lack of Raven Ridge solutions in PC OEMs could mean something.
Raven Ridge's cores do have a bit higher IPC than desktop Ryzen and at least in synthetics they seem to be practically on par with Skylake solutions, meaning it would always be a massive improvement over what they're using right now, which are 28W Haswell dual-cores with DDR3 and Iris GPUs without EDRAM.

They did use 45 W Quad-Core processors before. Heck the pre-2010 Mac Mini's even had a 110 W power supply. A small change in form factor could do a lot.
 
Well apple does seem to be pushing Radeon GPUs everywhere they can, and the lack of Raven Ridge solutions in PC OEMs could mean something.
Raven Ridge's cores do have a bit higher IPC than desktop Ryzen and at least in synthetics they seem to be practically on par with Skylake solutions, meaning it would always be a massive improvement over what they're using right now, which are 28W Haswell dual-cores with DDR3 and Iris GPUs without EDRAM.
But is RR's VCN better than Intel's QuickSync? I don't think Apple would really care about hw accellerated videoencoding, but users will. On the other hand macOS could use the extra GPU power for face recognition, AR etc.
 
Is it possible that the CPU could use the 4gb of hmb2 as a "cache" before the main ram ?

I highly doubt, it looks like a very standard CPU (with the regular intel IGP and all) apart from the package with the Radeon,
it's just connected to the radeon via pcie x8, so using the Radeon memory as a cache is not really interesting I don't think,
 
Last edited:
This design appears to have 64 ROPS for 24 CU's. Full fat back end of Vega 64 given to Vega 24.

Will be very interesting to see die size and efficiency.
 
Back
Top