Confirmed by AMD & Intel - Rivals Intel and AMD Team Up on PC Chips to Battle NVidia

[senseamp]

Do people really care about 1mm package height savings on a gaming laptop? Also, with CPU, stacked memory, and GPU being so close to each other, would that not create a thermal hot spot compared to spreading them out, which would necessitate a thicker heat spreader, undoing the height savings?

The machine they showed has a hinge that lifts it off the lap, and moves keyboard back so that the palms don't rest on the computer. Ostensibly it's to accommodate a drawing pad up front, but could it also be because it's thermally uncomfortable to be in closer proximity to a person's lap or hands when under load?

 

[coercitiv]

Do people really care about 1mm package height savings on a gaming laptop?

Gaming laptops?! Unless Intel contra reveues the heck out of this product, it will be high cost and only make sense in machines which need to be slim and powerful at the same time. That smells like high end ultra mobile notebooks with workstation pedigree. People who buy laptops for gaming will still go for Nvidia based solutions in 2018, which will be either cheaper, faster, or both.

Also, with CPU, stacked memory, and GPU being so close to each other, would that not create a thermal hot spot compared to spreading them out, which would necessitate a thicker heat spreader, undoing the height savings?

In mobile you're limited by cooling capacity first (heatsink surface area, airflow), not how close chips are. If anything, higher integration has the potential to free up to enough space inside the chassis to allow a bigger fan/heatsink assembly, or go for a smaller dual fan setup. Either is a win in terms of thermal performance over the classic setup.

 

[IRobot23]

So we can expect ryzen + vega also in 2018. 8 core + 200mm^2 die gpu with 4gb of hbm?

So midrange hbm +vega + pinnacle + insane amount of glue = insane laptop performance

 

[Yotsugi]

So we can expect ryzen + vega also in 2018.

Not in such package.

8 core + 200mm^2 die gpu with 4gb of hbm?

4core + GPU though, APU die is getting refreshed late 2018.

 

[IRobot23]

Not in such package.

4core + GPU though, APU die is getting refreshed late 2018.

Why would they put IF in VEGA then?

 

[Yotsugi]

Why would they put IF in VEGA then?

IF is a protocol.
PHYs can be whatever.
GMI, PCI-E, on-die crossbar, whatever you want.

Goddamit, the "Infinity Fabric" is almost nVidia levels of marketing.
AMD learned well, it seems.

 

[IRobot23]

IF is a protocol.
PHYs can be whatever.
GMI, PCI-E, on-die crossbar, whatever you want.

Goddamit, the "Infinity Fabric" is almost nVidia levels of marketing.
AMD learned well, it seems.

Okay, I guess you didn't understand me.

 

[Yotsugi]

Okay, I guess you didn't understand me.

IF in Vega is layered on top of on-chip crossbar.
Did you get the memo?
It's not some magic.
Just unified protocol.

 

[IRobot23]

IF in Vega is layered on top of on-chip crossbar.
Did you get the memo?
It's not some magic.
Just unified protocol.

I see. Whats its purpose for a GPU, if not MCM?

 

[Yotsugi]

I see. Whats its purpose for a GPU, if not MCM?

To simplify on-die resource management.
The data layer is on-die crossbar.
The management layer is Infinity Control Fabric.

if not MCM?

MCM is not exclusive to IF.
Magny-Cours and Interlagos Opterons were a things.

 

[IRobot23]

To simplify on-die resource management.
Infinity Control Fabric is one hell of a flexible thing.

MCM is not exclusive to IF.
Magny-Cours and Interlagos Opterons were a things.

I got that. Well I got excited when I saw that HPC APU thing.
https://www.dvhardware.net/article62212.html

 

[Yotsugi]

I got that. Well I got excited when I saw that HPC APU thing.
https://www.dvhardware.net/article62212.html

That's an ancient roadmap, the Greenland super-APU is no longer a thing.
Greenland, or GFX9, or Vega, call it whatever, lacks GMI.

 

[majord]

I think the focus was more on having a higher volume product using EMIB than the FPGA. Once the CPU supports EMIB, I imagine you could see a follow up product with the CPU/PCH/GPU/HBM2 all fused together.



You would have to include the interposer with the Vega+HBM2 though.

This is true, but I guess the point is, the advantage of EMIB is being offset (somewhat) by the fact it's still a two package, 4 die solution.

Strange to think If AMD had EMIB, All these APU dreams would have very likely come true.. I.e a larger HBM2 equiped variant of Raven Ridge.

 

[Yotsugi]

Strange to think If AMD had EMIB

Maybe Amkor will deliver on S-SLIM.
Someday.
Possibly.
Maybe.
AMD should start poking at OSAT partners right about now.

 

[IntelUser2000]

This is true, but I guess the point is, the advantage of EMIB is being offset (somewhat) by the fact it's still a two package, 4 die solution.

I don't get what you are saying. Even if the CPU used EMIB, you'd still end up with 3 dies in one package(CPU/HBM2/GPU). The point of using EMIB is when its hard/takes long to do monolithic die.

If you want it all on "one die" that's called 3D stacking. EMIB and Interposer is a 2.5D solution.

And, its one package, 3 dies. The leaked motherboard doesn't show a second chip.

 

[LightningZ71]

Interposer/EMIB wouldn't have helped AMD as much as people think. Big GPUs take up a lot of die space. Things would have been massive. At 7nm, you should see something interesting.

 

[majord]

I don't get what you are saying. Even if the CPU used EMIB, you'd still end up with 3 dies in one package(CPU/HBM2/GPU). The point of using EMIB is when its hard/takes long to do monolithic die.

If you want it all on "one die" that's called 3D stacking. EMIB and Interposer is a 2.5D solution.

And, its one package, 3 dies. The leaked motherboard doesn't show a second chip.

Ok, maybe forget the EMIB and die count part for a sec.. That's really only a factor for the chip manufacture (cost ) . I'm looking at it from the notebook OEM's perspective, and what Intel are actually offering - Which from what I can tell, is a two package solution , the first of which is a quite large, ungainly package, with the GPU and CPU seperated by quite some distance. Therefore It seems a RR or Zeppelin + Vega mobile would be just as compelling from a space point of view, perhaps even more compact, despite the fact the Vega chip would be more expensive due to the interposer.

The OEM can also mix and match CPU/GPU tier as they see fit.

 

[ksec]

Looks like a lot of people dont follow much about Apple.

1. This is likely not Apple Specific, much like the Iris Pro, but it is definitely made to Apple's requirement.

2. Apple has always been using AMD graphics, the only times when they do Nvidia was to tell AMD to play nice.

3. Nvidia never worked well for Apple, Apple needs full control, that is something Jensen wont give up on.

4. Apple also write their own Graphics drivers, or at least in deep collaboration with AMD. This actually goes further back to the ATI era.

5. In GPU, hardware is cheap, software is expensive. It is highly likely Apple dont want another GPU drivers stack to work with ( Nvidia ).

6. Those who continue to say Apple has little to no volume, despite its pricing and focus on higher end segment, Apple is the 5th largest PC vendor in the world, and this year they might enter Top 3 Vendor in market revenue.

7. In an recent interview, the number given out for Mac were 80% Notebook and 20% Desktop. I dont think there is any argument that Intel has always had the better Mobile CPU, and 10nm version coming soon. Hence why AMD APU didn't get the chance.

8. To an earlier post has mentioned this, The biggest threat to Intel is ARM, i would argue the biggest thread to both Intel and AMD is ARM. Microsoft has already announced their plan for Windows on ARM. Intel knows Microsoft has always tried but failed, their execution has never been good, and with a few moves and tricks they could get Microsoft tamed. On the other hand they know if Apple decide to switch to ARM, they will move earth and mountains to get it done, and there will be no turning back. Then Intel would have themselves kickstarted the biggest avalanche to x86.

9. Hence it is mutual benefits to both AMD and Intel, they both need to milk the x86 for as long as possible. GPU only makes you so much money, x86 is their real deal.

10. It is highly likely to be a custom Polaris, given Intel's time frame of 2018 1Q, i doubt a custom Vega were made and started 6 - 9 months ago.

11. This would give a single fan cooling design to use for Macbook Pro 13, MBP 15 will likely continue to have a dual fan design for CPU and GPU sperated.

12. I wonder what Macbook will use? Continue to use of iGPU from Intel? ( Hell no, please )

13. In case you are wondering, Apple is making way into AR and VR, hence the need for powerful GPU. Highly unlikely they have gaming in mind. ( PLEASE PROOF ME WRONG )

14. Given Intel will have to have their design fits to EMIB, i wonder if they took the chance to make custom die, no GPU included..... and might as well have an memory controller that support 32GB Memory...........

 

[jpiniero]

The regular Macbook uses the Y parts; so unless they are ready to dump Intel for ARM the only 2018 option really is Cannonlake-Y.

 

[Vattila]

12. I wonder what Macbook will use? Continue to use of iGPU from Intel? ( Hell no, please )

The latest MacBook Retina Mid 2017 model uses the weak Intel HD Graphics 615. If Apple should want the better performing AMD Radeon graphics across all their notebook products, AMD Ryzen Mobile should have a real chance here, assuming its performance and efficiency turn out to be as good as rumoured. Remember, AMD was allegedly really close to a win in MacBook Air with Llano.

With the cooperation between Intel and AMD on the Kaby Lake-G, Apple now has a chip to simplify its MacBook Pro line. Currently, as of the Mid 2017 refresh, the two 13 inch models, with and without the Touch Bar, comes with Intel Iris Plus Graphics 640 and 650, respectively, while the 15 inch model, also with Touch Bar, comes with switchable graphics between Intel HD Graphics 630 and AMD Radeon Pro 555. Using Kaby Lake-G across the line for the 2018 refresh looks like an obvious simplification, especially in terms of graphics drivers.

Kaby Lake-G has also been rumoured as an upgrade for the Mac Mini. As of the Late 2014 refresh, the two lower-end models use Intel HD Graphics 5000, while the top model uses Intel Iris Graphics 5100. Again, using Kaby Lake-G across the line looks like an obvious simplification, with the better performing AMD Radeon graphics.

The iMac 21.5 inch range currently comes with Intel quad-core processors ranging from 2.3 to 3.4 GHz base clock, and Intel Iris Graphics 640 for low-end 1080p model, and AMD Radeon Pro 555 and 560 for the two Retina 4K models, respectively. Using Kaby Lake-G for across the iMac 21.5 range should be a simplification. AMD Raven Ridge APU may be an option for the 1080p model. The three models in the iMac 27 inch Retina 5K range comes with higher-end AMD Radeon Pro 570, 575 and 580 graphics, respectively.

The new iMac Pro, coming in December, with an 18-core Intel processor, will have AMD Radeon Pro Vega discrete graphics.

So, in summary, with AMD Ryzen Mobile, the upcoming Raven Ridge APUs for desktop, and Intel's Kaby Lake-G, Apple now has a good opportunity to standardise on better performing AMD Radeon graphics across their desktop and notebook products.

 

[senseamp]

Apple wants its own stuff to be proprietary, but every one else's stuff to be generic. Which is what any company should aim for. To devalue everyone else's value add but their own. AMD is perfect for their needs, since they are content with being the generic version in every market they are in. The problem of course is without a proprietary value add, it's hard to make enough profit to invest for the long term. Also in an open model, the incentive is to wait for someone else to take the risk, and then quickly copy it once it is successful. Hence you get OpenCL. Apple will never go with a proprietary ecosystem that's not their own, no matter how superior. They ruined the trashcan Mac Pro with thermally throttled inefficient AMD GPUs, by their own admission, and yet they still signed up for more AMD GPUs for the new one. They are that afraid of their users latching on to a proprietary ecosystem not under their control.

 

[Yotsugi]

They ruined the trashcan Mac Pro with thermally throttled inefficient AMD GPUs

I wouldn't call Keklers "efficient".
The trashcan was bad as a design.