The real reasons Microsoft and Sony chose AMD for consoles [F]

[Erenhardt]

Guys, the reason the GT3e die is so large is that it's done in the name of power efficiency. You want more performance at less power? Go ultra parallel, clock lower.

If your power budget is bigger, then you can clock way higher to save on die space.

Then why GT3 is clocked at 1300mhz?

 

[krumme]

Then why GT3 is clocked at 1300mhz?

Rofl

 

[Exophase]

Guys, the reason the GT3e die is so large is that it's done in the name of power efficiency. You want more performance at less power? Go ultra parallel, clock lower.

If your power budget is bigger, then you can clock way higher to save on die space.

I very much doubt that GT3e uses a different physical implementation for its GPU than GT3. It probably looks exactly the same as the GT3 in the dual core chip which is spec-wise identical to GT3e except lacking the ability to interface with Crystalwell (and even that may not be true). And we know from pictures that it uses close to 100mm^2, not > 170mm^2.

 

[SiliconWars]

This is a quite bizarre thread. Are we actually at the stage where people are seriously suggesting that Intel can do anything "because it's Intel"? (thank you Keysplayr for that pearl of wisdom, that's a big help).

Seeing is believing. When Intel show they can manufacture a better GPU than AMD or Nvidia at a competitive price, we can start talking about it. The best they can offer so far we've got HD 6670 performance at $500.

When it comes to small cores Intel has nothing to match Jaguar currently. Silvermont might get them there, but these consoles are launching in a couple of months. Launching in a couple of months means the silicon has to be available for a lot longer. Silvermont was never going to be ready in time - are we supposed to believe that the current Atom was good enough?

Intel was never in the equation - they don't have the capability of anything AMD could offer. They probably didn't even put in an offer because they knew they were so far out of the race anyway. Nvidia might have embarrassed themselves by attempting a Tegra 4 offering instead.

AMD is on record as stating they are getting paid towards the top-end of client offerings for their console SoC's.

We are excited about our semi-custom pipeline and the large opportunity for AMD to deliver semi-custom silicon with ASPs at the higher range of our client offerings.

They didn't need to sell it cheap because they were the only game in town from the start.

 

[galego]

My position is neutral; your position makes a positive assertion that is unsupported by the evidence.

Your position is anything but neutral:

It is certainly not the case that Intel couldn't provide a custom SOC. It is simply that Intel didn't want to do so, because they didn't feel it was in their best business interests. AMD did.

The "real reason" AMD won the contract is that it needed to, much as who gets a job sometimes boils down to who is willing to accept the lowest salary.

Moreover, your non-neutral position is built entirely on air: you lack any bit of evidence supporting your personal beliefs.

 

[SiliconWars]

I am forced to agree with galego's stance on this - all the evidence points to AMD having the only game in town.

I am not saying that Intel is incapable of competing had they chosen to, but it is much more absurd to claim that the could compete but they just didn't want to. Seriously, listen to your argument and you'll see it is on a very weak foundation.

There is absolutely no evidence that Intel is capable of producing a SoC on this level of complexity. There is even less reason to believe that they can do it economically. They literally have nothing going for them in this space, but instead of just accepting that Intel is behind AMD in some areas, you choose to invent reasons for why Intel wasn't interested instead. There's no sense in it, it's like saying AMD could easily beat Intels cpu's overnight if they tried.

Please feel free to prove me wrong by giving me an example of what kind of SoC Intel could offer that is a viable alternative to what AMD won with.

 

[galego]

That being said, it has always seemed to me that intel has done exceptionally well in synthetic benchmarks but gaming performance seems awry in comparison to mobile discrete GPUs. For instance, the GT3 exceeds the 650M in most synthetics, while the 650M is actually faster in many games. So I don't know how to explain that.

That is easy to explain:

Intel says that GT3e will be as fast as a GeForce GT 650M. That sounds pretty competitive. Haas: Their comparison is misleading on a number of fronts. First, they use synthetic benchmarks that they optimize for to compare performance. We prefer to use games to measure a GPU’s gaming performance, and optimizing for games requires a lot more effort.

Haswell is being massively rejected in the market for which it was designed because it is slow, power hungry, and expensive:

Notebook buyers can get much better performance at a significantly lower cost by selecting a GeForce notebook. OEMs don’t seem all that impressed with GT3e, as it’s power hungry and expensive. We expect only a tiny number of notebooks will come with GT3e.

Which PC OEMs will be offering Haswell notebooks with discrete GPUs?
Haas: Every major PC OEM will be offering notebooks with Haswell and discrete NVIDIA GPUs.

And still some people here believes that Intel had some opportunity to win a console contract. LOL

Intel was never in the equation - they don't have the capability of anything AMD could offer. They probably didn't even put in an offer because they knew they were so far out of the race anyway. Nvidia might have embarrassed themselves by attempting a Tegra 4 offering instead.

AMD is on record as stating they are getting paid towards the top-end of client offerings for their console SoC's.

They didn't need to sell it cheap because they were the only game in town from the start.

Interesting link. This part is very important:

Our low-power single-chip embedded SoC delivers more than two times the compute and four times the graphics performance of our competitors’ offerings.

 

[blackened23]

Just because intel hasn't created a high performance high TDP graphics part, doesn't mean they can't - the graphics core used in the GT3 is scalable just as the Jaguar graphics core is. It can work in extremely low power devices with less performance, or go all the way up to high performance with added stream processors - performance is generally linear with the number of stream processors added. That is exactly what AMD did with the SOC in the PS4, and intel could design a similar product.

Intel is obviously focused on the mobile market with their integrated graphics and they haven't yet made a high TDP graphical processor. Again, that doesn't mean they can't do it - they have the technology to do it.

I can see an argument for AMD being a better choice simply by the fact that intel tends to do extremely well in synthetics, while they are VERY hit and miss in actual games performance. For whatever reason, AMD and nvidia graphics for the most part tend to be very much more consistent in actual games - whereas, as an example I stated earlier, you can compare the GT3 to the 650M where the GT3 trounces it in synthetics. Yet the 650M is faster in actual games. Maybe this played part as to why AMD is chosen, but I don't know what the actual reason is. Nor does anyone here, but I don't think it's a technology issue. Both AMD and intel had the technology to create an appropriate x86 SOC for a console - But, again, I can see an argument for AMD being a more appropriate choice for gaming. Thus far all of intel's graphical efforts have been hit and miss in actual games, and they have been slightly behind in terms of features.

 

[blackened23]

Haswell is being massively rejected in the market for which it was designed because it is slow, power hungry, and expensive:

Are you kidding me? I understand you dislike intel for whatever reason but you're really reaching here. Come on.

 

[Charles Kozierok]

I am not saying that Intel is incapable of competing had they chosen to, but it is much more absurd to claim that the could compete but they just didn't want to.

It's not absurd at all. Many companies, in fact most companies, decide what sorts of markets they will target based on their relative expertise and their business requirements.

Anyway, I will admit that I overstated a couple of my comments earlier. I really don't know the details on why AMD won this contract. I'm just saying that Intel not winning it doesn't automatically prove a lack of "expertise", any more than a high-end steakhouse turning away a busful of tourists looking for $6 cheeseburger lunches.

And what do you expect the company to say, "we're doing this for peanuts because we're desperate"? Please. All you are going to get from them is positive spin, because that's part of their job. They are going to spin it positively no matter what the reality is.

 

[Charles Kozierok]

Are you kidding me? I understand you dislike intel for whatever reason but you're really reaching here. Come on.

I'd love to know what is supposed to be in that article to support the claim that "Haswell is being massively rejected in the market for which it was designed". All I see is "Haswell is a great CPU". Maybe he's using an alternative definition for "great" than everyone else...

 

[mrmt]

There is absolutely no evidence that Intel is capable of producing a SoC on this level of complexity. There is even less reason to believe that they can do it economically.

And here's the catch. Economically to who? I think you can agree that AMD works with far smaller brackets for ROI and cash flows than Intel. Right now, they are scrapping at the bottom of the barrel in terms of margin and cash flows, so we have a company eager to do business. This is AMD's greatest "advantage" for this kind of contract, because they will go as low as they can, while Intel - and Nvidia - have no incentive to do so. Let's look at an hypothetical scenario:

AMD has $40 in costs, but needs only 30% gross margins with the given volumes.

Intel has $38 in costs, but they need 50% gross margins on the given volumes.

Even if with smaller cost Intel would lose the contract because the ROI levels are bigger at Intel.

AMD could indeed had the smaller cost to develop both processors and Intel could have balked at the extra costs to develop these processors, but this is a secondary issue. The smaller ROI plays a much bigger role here than you think. In other words, you are mistaking "not selling as low as the customer want" with "not being able to do if the price is right".

 

[NostaSeronx]

The real reason is accessibility, Intel has to be fabbed at Intel, Nvidia has to be fabbed at TSMC. AMD can be fabbed at TSMC, GlobalFoundries, etc. Sony and Microsoft both demanded to have parts be able to be fabbed at three separate foundries.

Initial Foundry - GlobalFoundries
Second Foundry - TSMC
Third Foundry - UMC or Samsung

 

[galego]

Are you kidding me? I understand you dislike intel for whatever reason but you're really reaching here. Come on.

Just read the quotes given in my message (or go to the source). It not only states that OEMs are massively rejecting Haswell integrated graphics, but explains why. I will repeat part of the quotes given before:

Notebook buyers can get much better performance at a significantly lower cost by selecting a GeForce notebook. OEMs don’t seem all that impressed with GT3e, as it’s power hungry and expensive. We expect only a tiny number of notebooks will come with GT3e.

Every major PC OEM will be offering notebooks with Haswell and discrete NVIDIA GPUs.

Intel has failed to convince gaming notebook OEMs with its best integrated graphics. How could it convince Sony and Microsoft for consoles?

Even if Intel was to develop tomorrow a SOC with 2x the graphics performance of GT3e: it would be much slower, power hungry, and expensive than the winner SOC made by AMD. End Point.

 

[rgallant]

just wondering why no ps4 or xbox chips have not fallen into the wrong hands by now.
-also my noob side is wondering if intel was missing some ip that took them out of the bidding , were as amd had it on file. re Larrabee 's effort trying to enter a market with out using any gpu ip.

 

[ThePeasant]

Just because intel hasn't created a high performance high TDP graphics part, doesn't mean they can't - the graphics core used in the GT3 is scalable just as the Jaguar graphics core is. It can work in extremely low power devices with less performance, or go all the way up to high performance with added stream processors

No one sees any problem with using this kind of reasoning... /:

 

[Silverforce11]

Too many people give Intel too much credit, in fact in their recent generations, IPC gains have been laughable and their iGPU perf/mm2 gains have also been abysmal. The only reason Haswell improved so much on graphcis compared to SB/IvB is because a huge proportion of the die is devoted to graphics (and Iris Pro's massive on chip e-vram). That's nothing special, and it wont help them compete with small discrete dies with their own GDDR5, especially not when they try to pawn their Iris Pro at ridiculous prices.

Compare Intel's die space usage to NV or AMD on discrete chips, and their perf/mm2 is horrendously bad, even with a node advantage.

 

[mrmt]

Compare Intel's die space usage to NV or AMD on discrete chips, and their perf/mm2 is horrendously bad, even with a node advantage.

How do AMD APU fare within the same constraints you put here?

 

[Charles Kozierok]

Just read the quotes given in my message (or go to the source). It not only states that OEMs are massively rejecting Haswell integrated graphics, but explains why. I will repeat part of the quotes given before:

Wow, Nvidia thinks that people are better off buying notebooks with Nvidia GPUs in them rather than using Intel IGPs.

This is truly earth-shattering news and proves beyond all doubt that Haswell is doomed to failure!

 

[Vesku]

http://www.anandtech.com/show/7085/the-2013-macbook-air-review-13inch/4

Anyone saying Intel could have offered a similar console solution needs to use the HD 5000 vs HD 4400 information to extrapolate how many EUs would be needed to hit roughly Radeon 7790 levels of performance.

 

[Exophase]

Number of EUs is just one part of the story. HD 5000 in 15W TDP solutions doesn't maximize the performance possible by the die investment. Iris 5200 Pro gives a better showing, although I don't know if even that couldn't go higher if given a larger thermal budget.

 

[Schmide]

I think AMD is getting their business now because in some ways they didn't take the steps that other competitors have. When I first heard of AMD back in the 80s me and my buddy coined a term that they were great at picking up crumbs. They were willing to fill gaps and put together alternatives that were cost effective and to some degree competitive.

The only chip I can think of that was unique to the point where it was somewhat uniquely compatible with the software industry was the Xenos. Even then it was only because it was halfway between standards. (dx9 leaning dx10: Unified shader model with longer instruction lists kind of a Shader model 3.0+)

So where am I going with this? AMD has done very little (if anything) to hinder software developed on their systems from running on competitors systems.

Spoiler

I.e. They don't have their PhysX or SSE2 (genuine) moment. I put this in spoiler because I don't want this to be about that.

The good: This should make games developed on their consoles easier to port.

The bad: This should make games developed on their consoles easier to port.

 

[Silverforce11]

How do AMD APU fare within the same constraints you put here?

One would have to wait and see how the console custom APUs with GCN cores perform and judge. If its 7790 performance levels on a SoC, then its pretty damn good.

Trinity and Richland are ancient VLWI cores and they are still pwning Haswell, minus the super big and expensive Iris Pro.

As to the scalability and "if Intel wanted to do it, they could" mentality.. just no. Intel wanted GPU grunt and HPC dominance.. still have not got it. Intel wanted a consumer GPU (Larrabee, Knights etc) that's competitive with Fermi.. sure, a few generations after, completely obsoleted by newer GPUs from NV/AMD. Intel has great expertise in CPUs, but their GPU is still far behind in perf/transistor/mm2... when it comes to consoles and the desirability of a SoC design, they simply cannot match.

 

[Vesku]

Number of EUs is just one part of the story. HD 5000 in 15W TDP solutions doesn't maximize the performance possible by the die investment. Iris 5200 Pro gives a better showing, although I don't know if even that couldn't go higher if given a larger thermal budget.

Just calculate with the 4400 -> 5000 data and get back to us with the EU count.

 

[Exophase]

Just calculate with the 4400 -> 5000 data and get back to us with the EU count.

If you're going to disregard my posts then leave it at that. Don't repeat yourself to me like I'm an idiot. I already explained why I don't think HD 4400 to HD 5000 is a fair representation of Intel's IGP theoretical scaling capability, so I really have nothing to offer in response to what you want.