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

Page 12 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

maddie

Diamond Member
Jul 18, 2010
4,723
4,628
136
Oke, but how is it with the total cost and reliability. I have been reading that silicon interposers are fragile for temperature changes and effects like warping must be fully controlled or the microbumps between the asics and interposer or substrate and interposer might come loose. Underfill seems to alleviate that but mounting for example a heatsink is a difficult task. And everything that needs to be done precise and with strict tolerance becomes more expensive.
I understand a silicon interposer itself is cheap, but the whole production process towards the end product may be more expensive. At least that is my understanding of it.

As a sidenote the silicon die bridges in EMIB may not be so bad when thinking of pass through when a designer uses standardized modular technology to connect it all together.
And that is what AMD has done with IF. And when looking at the picture , one only sees 2 chips, but i wonder if there is no limitation to connect 4 dies at once with one square silicon bridge.
Then it may not be that bad as you might think. I do have to note, i do not know the maximum size a silicon bridge in EMIB technology can be.

EDIT

T
he silicon bridge die could perhaps also be a hexagon or octagon if the cpu dies are small and rectangular enough. That would allow for 6 or 8 chips to connect to one silicon bridge die. All at once.
Would be like spokes on a wheel. How many cores does one want ?

wheel-with-spokes.jpg
Both the ICs and SI are silicon, meaning similar coefficients of thermal expansion.
Once you go beyond a certain # of interconnects you will have the need for microbumps. HBM2 stack =1024 bit at present. EMIB or full SI will not matter. Alignment will be to same tolerances.
IF won't allow changes to signal pathways without new masks. IF is scalable but not after fact.
Although there are ways to slice non-square die from a wafer, I doubt this will happen for most cases, reason is costs.
Remember, almost anything can be built, but does it make any sense.

I have never found the costs savings in absolute terms to be that critical to justify the reduced flexibility of EMIB designs. Percentage wise you might get reasonable savings, but if its only a couple $ on a $75-100 design, then not worth it IMO considering the negatives.

AMD's priority is to get interposer assembly volumes fully worked-up by making the tech cheaper thus allowing lower-end higher volume designs. HBM2 has to be more widely used for this to happen. Chicken & egg.
 
May 11, 2008
19,306
1,131
126
Both the ICs and SI are silicon, meaning similar coefficients of thermal expansion.
Once you go beyond a certain # of interconnects you will have the need for microbumps. HBM2 stack =1024 bit at present. EMIB or full SI will not matter. Alignment will be to same tolerances.
IF won't allow changes to signal pathways without new masks. IF is scalable but not after fact.
Although there are ways to slice non-square die from a wafer, I doubt this will happen for most cases, reason is costs.
Remember, almost anything can be built, but does it make any sense.

I have never found the costs savings in absolute terms to be that critical to justify the reduced flexibility of EMIB designs. Percentage wise you might get reasonable savings, but if its only a couple $ on a $75-100 design, then not worth it IMO considering the negatives.

AMD's priority is to get interposer assembly volumes fully worked-up by making the tech cheaper thus allowing lower-end higher volume designs. HBM2 has to be more widely used for this to happen. Chicken & egg.

Oke. Interesting. But how is it with mechanical strength with silicon interposer technology ?
 

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
I lol'd when I read that also. Analysts are the absolute worst.

Most of them are really bad. But Kevin Krewell rightly speculated that Apple possibly brokered this deal and will most likely use it in Macbooks and maybe even iMacs.
 

maddie

Diamond Member
Jul 18, 2010
4,723
4,628
136
Oke. Interesting. But how is it with mechanical strength with silicon interposer technology ?
Not sure I understand the question.

An interposer is made on older node fab lines. That means that the interposer is about the same thickness as a normal die.

You might be thinking about HBM memory modules where the individual memory ICs are thinned to prevent the total stack from being too tall. In the unassembled state, they are very thin, flexible and fragile.

That however, is not a result of the SI or EMIB but a function of stacked ICs being mixed with other ICs.
 

maddie

Diamond Member
Jul 18, 2010
4,723
4,628
136
Not sure I understand the question.

An interposer is made on older node fab lines. That means that the interposer is about the same thickness as a normal die.

You might be thinking about HBM memory modules where the individual memory ICs are thinned to prevent the total stack from being too tall. In the unassembled state, they are very thin, flexible and fragile.

That however, is not a result of the SI or EMIB but a function of stacked ICs being mixed with other ICs.


edit:
One thing that struck me early on with EMIB was that the bridge was a mini silicon interposer embedded in an organic PCB matrix. This IMO should have a lot more susceptibility for differing thermal expansion. The locked in placed bridge microbump connections being stressed by the base and silicon ICs expanding differently. Intel obviously have found a working solution.
 
May 11, 2008
19,306
1,131
126
Not sure I understand the question.

An interposer is made on older node fab lines. That means that the interposer is about the same thickness as a normal die.

You might be thinking about HBM memory modules where the individual memory ICs are thinned to prevent the total stack from being too tall. In the unassembled state, they are very thin, flexible and fragile.

That however, is not a result of the SI or EMIB but a function of stacked ICs being mixed with other ICs.

edit:
One thing that struck me early on with EMIB was that the bridge was a mini silicon interposer embedded in an organic PCB matrix. This IMO should have a lot more susceptibility for differing thermal expansion. The locked in placed bridge microbump connections being stressed by the base and silicon ICs expanding differently. Intel obviously have found a working solution.

Well, the thing is that the product, for example a gpu +HBM2 package needs to be careful handled. Because the silicon interposer can chip easily. That is what i was wondering about.
Like mounting heatsinks or even heatspreaders.
Another thing that got me wondering is the assembly process. Mounting the dies on the silicon interposer and then on the substrate or other way around may cause warping of the silicon interposer, reducing the reliability of the mechanical strength of the bumps between the asic dies and the silicon interposer and the bumps between the silicon interposer and the substrate. I am sure it is all solved with techniques like underfill but at what cost.
Also, i read that warping the pcb of the final customer product like a gpu card is also more of an issue but that is more of the sheer size of silicon interposer technology combined dies as package and emib should have the same issues because it is also meant to create huge packages. That would be a headache to assemble the pcb. If the multi die package is placed first, i do not think they will be using any through hole components with a solder wave bath to prevent pcb warping. Although techniques exist to create a tiny local solder wave to only let the solder wave hit the component pins that are intended to be soldered and nothing else(By moving the pcb around with a robot that moves the pcb carrier) and i think this would reduce pcb warping greatly even elliminate it . Then again, going all SMT reduces costs as well but that aside.

For the possible different thermal expansion, i think Intel with EMIB has found a similar solution as silicon interposer technology, both have substrates and i assume that a substrate (organic pcb) is chosen that has similar thermal expansion behavior as silicon does.
 

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,619
136
It's opportunity cost. AMD could have made this MCM themselves and controlled this niche where Intel couldn't touch them.
No, a pure MCM wouldn't be able to distance Intel at this time. AMD is still in progress of polishing all their newly introduced pieces, be it Zen or Vega, why would they want to invest in sidetracking themselves with a solution that they may well be able to do better and cheaper at a later point.

Also I think people undervalue the purpose of AMD's embedded and semi custom business unit, that business essentially allows AMD to do research and development in areas they likely wouldn't focus on on their own while being paid by their customer for doing so. This likely vastly broadens AMD's experience beyond what they are pushing with solely their own R&D.

In this case Intel touts EMIB and improved power control as parts of the package (features that required a semi custom solution, a reason why Nvidia is an unlikely partner). While the former won't be available to AMD (and more specifically its foundry partners) to use without licensing they now know the pros and cons of it. The latter are likely improvements right in the GPU design requested by Intel that AMD can reuse in their own later designs, similarly to how experience from previous semi custom R&D made its way into current products.
 
  • Like
Reactions: SpaceBeer

nickmania

Member
Aug 11, 2016
47
13
81
I do not understand the idea that this is for Apple. IMHO the only different of this product with a ryzen-vega APU is that the graphics memory will be faster, due the HBM2 near the chip, instead DDR4.

Well, virtually no productivity task, 3d, Photoshop, 3d creation, video playing etc... will benefit from this, the only task that will be better will be 3d gaming... but Apple has no gaming market so do not get the idea that they have requested that.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,726
3,560
136
I've given this a little more thought and this has implications beyond Apple, and those don't bode very well for AMD, unless they have something planned of their own.

If they can make the GPU work with OmniPath instead of PCI-E, then Intel is going to be much ahead of everybody else in enabling true HSA using EMIB.
 
  • Like
Reactions: Drazick

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,619
136
If they can make the GPU work with OmniPath instead of PCI-E, then Intel is going to be much ahead of everybody else in enabling true HSA using EMIB.
Do you expect Intel to integrate OmniPath in consumer chips?
 

tamz_msc

Diamond Member
Jan 5, 2017
3,726
3,560
136
Do you expect Intel to integrate OmniPath in consumer chips?
That entirely depends on how they've thought about the scope of this product beyond thin laptops, but the implications are there. Not for a consumer oriented chip, but integration of OmniPath will result in overcoming a big hurdle to integrating the GPU in HSA.
 
  • Like
Reactions: Drazick

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
I do not understand the idea that this is for Apple. IMHO the only different of this product with a ryzen-vega APU is that the graphics memory will be faster, due the HBM2 near the chip, instead DDR4.

Well, virtually no productivity task, 3d, Photoshop, 3d creation, video playing etc... will benefit from this, the only task that will be better will be 3d gaming... but Apple has no gaming market so do not get the idea that they have requested that.

Apple exclusively uses:
Intel CPUs.
&
AMD GPUs

This connects both, and Apple is also fanatical about shrinking products, so another plus in that regard.

So really this does seem to be an ideal product for high end Apple laptops that currently used dGPU.

So that line of speculation is perfectly understandable. But of course it is still speculation.
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
I see Apple as likely to be the major driving force behind this product, but I absolutely wouldn't count out Microsoft with an eye towards the Surface family of products. As much as Apple needs this, so does MS.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Bollocks. It's a chip going in Macbooks and such. You won't be able to get it anywhere else. It's an OS X product, not a Win10 product.

It will also be in a Intel NUC according to Intel roadmap. In fact that roadmap was the biggest hint these rumors were true to begin with because the package size of that kaby-lake G chip in the roadmap was ginormous.

I'm sorry that you can't understand why AMD giving away its biggest competitive advantage to their biggest competitor could possibly be a bad thing. Your basically non-sensical, insulting and completely uninformative reply says more about you than it does about me.

Do you know how much Intel pays AMD for this? No? Well how can you then say it is non-seniscal?
 
  • Like
Reactions: Phynaz

Timorous

Golden Member
Oct 27, 2008
1,538
2,537
136
I said they use less power and perform comparably to intel CPUs. Yes, they will be seen as the "lower cost" option because the laptops you will see with Raven Ridge will be much less expensive, much lower performance and probably lower volume than whatever they produce with intel for Apple.

If the only way AMD can get into Apple computers is by being tied to an intel CPU, then Ryzen will never be used by Apple and it will be viewed by Microsoft and others as a low-cost alternative to intel only to be used in cheap, low-margin products. This will hurt consumer perception of AMD Ryzen and will not help AMD in the long run in taking marketshare from intel.

You cannot take marketshare from intel by selling intel CPUs. I would think that is a simple concept to understand. Perhaps others here believe AMD's main competitor is nvidia and not intel, I disagree.

AMD would not be able to take marketshare from Intel in this segment regardless because majority of OEMs at this level pair Intel with Nvidia and AMD does not even get a look in. This product is a shot at Nvidia and will help AMD gain some marketshare in the dGPU laptop space.

As far as spin I see it as far easier to spin it to the OEMs as Intel needing to come to AMD to achieve their GPU performance targets and AMD were the only partner with the experience and technologies to realise this end goal. It also tells the OEMs that if Intel are willing to work with AMD on a high end product that AMD must have enough QA and QC to meet Intels standards and as such if AMD produce a competing product in the future they may be more willing to actually use it.

Realistically I don't see AMD producing something like this till Zen 2 and 7nm is available. That will be a much better platform to build on.
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
heh, this will become a nightmare for the research firms. How do you calculate market share for this product ?? GPU is AMD but the End product is Intel.
 
  • Like
Reactions: Drazick

LTC8K6

Lifer
Mar 10, 2004
28,520
1,575
126
How can it be Apple only when the whole rumor was supported by the idea and the roadmap regarding a NUC?
 

DrMrLordX

Lifer
Apr 27, 2000
21,583
10,785
136
It will also be in a Intel NUC according to Intel roadmap. In fact that roadmap was the biggest hint these rumors were true to begin with because the package size of that kaby-lake G chip in the roadmap was ginormous.

Fair point, though I have to say that AMD getting any of their products into an Intel NUC is a huge win for AMD. You darn well know that wasn't a market where AMD could have sold Raven Ridge or similar, at all.
 

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,619
136
heh, this will become a nightmare for the research firms. How do you calculate market share for this product ?? GPU is AMD but the End product is Intel.
How are laptop sales which include an iGPU and a dGPU calculated (so like every laptop with a dGPU unless we talk DTR niches with iGPU-less Ryzen CPUs)? Same deal here.