Do Kepler and Maxwell have an AIB future?

[swerus]

How is Intel gonna be a high end graphics competitor? Oh I know Magic!

 

[busydude]

I was thinking high end Fusion chip progression as it will change the discrete GPU martket into the future.

The discrete GPU market will change for sure after APU's enter the market. Only the lower and lower mid-range cards will be cannibalized, the higher-mid and high end cards will have their own market.

A lower mid range card in an eyefinity~2560 monitor enviroment, but not in a 1080P single display enviroment.

I am sorry, a 5670 is a lower mid range card, period. A single 5970 is bought to its knees in 3 X 1080p eyefinity, let alone a 3 X 30" eyefinity setup.

Almost all computer games will be developed first for consoles with eyefinity added for the computer version. But even if you have an eyefinity set-up, you can just add a graphics card of your choice to your quad bulldozer core Fusion chipped computer. If you have 3 20" monitors, your Fusion gpu already has you 2/3 of the way home. Just add a low cost AIB and you're there.

You seem to ignore the complexity involved in that setup. AMD already implements Hybrid CrossfireX and look at the improvement it delivers. A crossfire between a Discrete card and a GPU which is fighting for bandwidth resources with a CPU.. yeah right.

Oh, Bulldozer does not have a fusion part that has been announced yet. You still will need a discrete card to play games, same goes for Intel 2011 socket.

 

[Tsavo]

i say that we're a long way from seeing cpu+gpu packages beating high performance gpus. Nvidia's safe IMO

Especially at the rate Intel releases GPUs. We'll get a GTX 460 level igpu in 2023.

 

[psoomah]

How is Intel gonna be a high end graphics competitor? Oh I know Magic!

The past does not predict the future, the present does.

In the past Intel had a history of atrocious GPUs. That does not predict they will continue to do so.

In the present Intel is dead serious about advancing their graphics performance on their APUs as fast as possible.

In the present Intel can no longer use their 600lb gorilla status to bully the market.

In the present AMD's Fusion chips are an immediate and serious threat to their market share.

In the present Sandy Bridge doubles the previous generation's graphic performance plus tightened the cpu/gpu integration.

The following generation might double it again. It might triple it. It might quadruple it. Intel has the deepest pockets in the business. When it gets REALLY serious about improving it's graphics things are going to start MOVING. They've already spent several years digging deep into graphics with their Larrabee project. That knowledge is being applied to their APUs.

Intel is not competing with AMD's top end graphic chip performance, it's competing with AMD's fusion graphics performance ~ 1/4 of their top end chip.

Intel also has an unmatchable process edge and can bring their APU's to market at a smaller node a year before AMD.

I never claimed Intel will be a high end graphics player, only an APU player comparable to AMD, which will, in three or four years, be sufficient to provide a satisfactory gaming experience for the vast majority of computer gamers with their high end APUs.

Where is the 'magic' in that?

 

[swerus]

Intel can and most likely will improve their GPUs yes. They won't have to compete with AMD because they can strong arm OEMs like they did when AMD had a better processor.

If any of that was a worry then Intel IGPs wouldn't have been shoved into all those boards over the years and it would have been an ATI or nvidia there instead. There was always competition, and the competitor had a better product.

Intel is light years behind, and them deep sixing larrabee proved that.

 

[psoomah]

The discrete GPU market will change for sure after APU's enter the market. Only the lower and lower mid-range cards will be cannibalized, the higher-mid and high end cards will have their own market.

High end Fusion graphics capabilities are scheduled to double on a yearly basis. It's a MOVING TARGET.

2011 Llano APU = 5670
2012 APU = 5770
2013 APU = 5870
2014 APU = 5970

2014 console graphics = 1080P single display + console based game graphics based computer game graphics. A 2014 APU will eat that for breakfast.

That is likely sufficient for 95% of the computer gamers ... and games.

 

[busydude]

console graphics = Crappy graphics.

An 8800 GTX will eat console games for breakfast.

It's a MOVING TARGET.

Exactly. Its a moving target, in two years time a 5870 will perform like a lower mid range GPU. Do you realize that Graphic quality is a moving target too?

 

[psoomah]

Intel can and most likely will improve their GPUs yes. They won't have to compete with AMD because they can strong arm OEMs like they did when AMD had a better processor.

If any of that was a worry then Intel IGPs wouldn't have been shoved into all those boards over the years and it would have been an ATI or nvidia there instead. There was always competition, and the competitor had a better product.

Intel is light years behind, and them deep sixing larrabee proved that.

Intel HAS improved their gpu. Substantially.

They can NOT strongarm OEMs like they did in the past. That was part of the settlement and it was legally binding. Intel has no slightest desire to test the terms of that settlement.

It wasn't a worry IN THE PAST.
There was not ubiquitous Broadband and streaming A/V IN THE PAST.
AMD Fusion was not challenging it IN THE PAST.

 

[blastingcap]

I laid out a timeline to 2H 2013. What specific parts of my reasoning concerning that timeline do you disagree with?

Kepler and Maxwell are specifically the GPU architectures mentioned by JHH that my reasoning led me to conclude would not be viable in the AIB graphics market. If you think it is premature, please address specific points I made with specific rebuttals to those points. Specious generalizations are not condusive to developing reasoned and differentiative discourse.

As I've previously stated, high-end GPUs take up a ton of space compared to CPUs. In the past we have seen things like math coprocessors getting absorbed into the CPU, but nothing as huge as a high-end GPU that many people don't need...

Also the memory subsystem for VRAM is much faster than regular RAM. I guess it's possible to have a sideport for VRAM or something though.

I suppose you could have a system where there are different combinations of CPU+GPU power so you could have high CPU and low GPU power, or medium of both, or low in both, or low CPU and high GPU, or stuff like that, and in fact I agree that this makes sense when latency becomes too large of an issue.

But I don't think the market will move that fast, certainly not in time for Kepler which is due only a year from now. Even Maxwell is, what, 2013? That's an awful lot of change for 3 years, to go all the way down to zero add-in boards. Not even for legacy mobos? Remember, there are tons of existing mobos that presumably will still be in service years from now.

Plus there is the pro-graphics market.

If we're talking 2015 or beyond, maybe... I mean, I am not even disagreeing with the meat of what you're saying, just the timeframe.

 

[psoomah]

Exactly. Its a moving target, in two years time a 5870 will perform like a lower mid range GPU. Do you realize that Graphic quality is a moving target too?

Specifically, like a 5670 does today.

Graphics performance need is a creeping target. Moving, yes, but moving very slowly. Far slower than computer graphics performance.

 

[tviceman]

APU's are a long time (or never) from replacing mainstream (and higher) discrete graphics. They may set a new standard for minimum requirements needed to run applications, but APU's coming out next year won't even be able to beat the G92 chip from Nvidia - which came out in 2007.

 

[psoomah]

APU's are a long time (or never) from replacing mainstream (and higher) discrete graphics. They may set a new standard for minimum requirements needed to run applications, but APU's coming out next year won't even be able to beat the G92 chip from Nvidia - which came out in 2007.

The top AMD APU coming out next year (2011) will have ~ 5670 graphics. That is scheduled to double every year.

2012 = 5770 APU graphics.
2013 = 5870 APU graphics.
2014 = 5970 APU graphics.

Are you contending 5970 performance APU graphics will not replace mainstream discrete graphics in four years?

 

[AnandThenMan]

Intel HAS improved their gpu. Substantially.

Not really. They have increased the speed, but feature wise Intel is far behind. They don't support DX11, and who knows when they actually will. Driver wise, they are hopeless, gaming on Intel is a dismal experience (I see people attempting it all the time on their laptops, many just give up).

The problem Intel has is they don't have the discreet graphics market to drive the technology forward. That is what pushes Nvidia and AMD/ATI to better hardware at a breakneck pace. I am not disagreeing with the overall substance of your arguments, I think they for the most part are bang on. I think Nvidia absolutely must reinvent themselves, or they will for certain end up a niche player, fighting for an ever shrinking market.

And remember, Nvidia has pretty much burned their bridges with AMD, especially when they outright said that no one wants AMD processors, so who cares about making chipsets for the AMD platform. AMD could very well in the future have the processor performance crown, and Nvidia will be shut out of any revenue from the platform demand. And for the doubters, 4 years ago almost no one thought ATI would regain the performance crown from Nvidia. In fact, many thought ATI was finished, and a prevailing attitude was AMD purchasing ATI was a terrible business move that could put AMD out of business, or at best, all that would be left of ATI was chipsets because that's all AMD cared about.

The point is never say never, and don't burn your bridges. And most importantly, have a backup plan. (recall that JHH stated that the best way to success is to have no backup plan, and to tolerate failure).

I also 100% agree that Fusion type products are going to become a larger and larger percentage of the market. It will get to the point where "graphics in everything" will be good enough where people will be able to play most games at 1080p without much fuss or compromise. There will always be discreet, and always be the ultra high end, but it will represent a much smaller % of the market, going by number of parts shipped. So where does that leave Nvidia? It's a scary future for them if they continue with the same business model and tactics they are employing now.

 

[AtenRa]

Lets start with Llano, it will be released in 2011 at 32nm and perhaps it will have the same performance as a HD5670 (I believe it will be near 5570). AMD and Intel change lithography every 2 years so in order to have a 5770 performance in an APU (Double the transistor count) and keep the same die size will have to wait until 2013. If they want to have a 5870 performance in the APU they will have to wait until 2015 for 15nm lithography.

Lets see what a discrete card will be in 2015,

2009 AMDs 40nm 58xx had 2.15 Billion transistors
2011 AMDs 28nm 78xx will have more than 4B Transistors and 2x the performance of 58xx
2013 AMDs 20nm 88xx will have more than 8B Transistors and 2x the performance of 78xx
2015 AMDs 14-15nm 98xx will have more than 16B Transistors 2x the performance of 88xx

So at 2015 we have an APU with 58xx performance and Discrete Graphics with 98xx performance, meaning discrete graphics will be more than 8x faster than 58xx APUs

At that rate APUs will always be Low End Entry Level graphics like Integrated Graphics are now.

 

[distinctively]

I consider every point I raised has a direct bearing on the question the thread post asked. If that is so, the entire post stayed on topic.

If you consider this is not so please address which point(s) you consider were sufficiently off topic to cause you to say that.

I am interested in reasoned, differentiated, logic based discussions on what is admittedly a slightly controversial subject matter, but a perfectly valid (and interesting) one when presented in a point by point manner as I did.

THAT kind of discussion is what stretches the mind and makes Forums fun and interesting.

At least it is so for me.

I said thread (referring to the S/A thread which ended up in a tangent to holodecks). I thought your post was well stated. I'm guessing that you misunderstood my post.

 

[Nemesis 1]

Lets start with Llano, it will be released in 2011 at 32nm and perhaps it will have the same performance as a HD5670 (I believe it will be near 5570). AMD and Intel change lithography every 2 years so in order to have a 5770 performance in an APU (Double the transistor count) and keep the same die size will have to wait until 2013. If they want to have a 5870 performance in the APU they will have to wait until 2015 for 15nm lithography.

Lets see what a discrete card will be in 2015,

2009 AMDs 40nm 58xx had 2.15 Billion transistors
2011 AMDs 28nm 78xx will have more than 4B Transistors and 2x the performance of 58xx
2013 AMDs 20nm 88xx will have more than 8B Transistors and 2x the performance of 78xx
2015 AMDs 14-15nm 98xx will have more than 16B Transistors 2x the performance of 88xx

So at 2015 we have an APU with 58xx performance and Discrete Graphics with 98xx performance, meaning discrete graphics will be more than 8x faster than 58xx APUs

At that rate APUs will always be Low End Entry Level graphics like Integrated Graphics are now.

Actually intel will have the ability to Add to the APU as they please. . I think that you underestamate intel . Even tho Intel announced SB to release In 2011 . 22nm is coming befor 2012. What chip I don't know. But SB 22nm should appear in 2012 as 22nm. Than in 2013 we should see Ivy bridge 16nm. Your saying that when intel shrinks SB as has been done on last 2 generation. There will be no APU improvement on the shrink that is a very foolish sratement. Intel will improve APU every tick tock. I may have the naming scheme screwed up , Haswell at 22nm shrink to rockwell at 16nm . SB and Ivy are same Ibelieve just a shrink and what ever intel adds or removes.

 

[n0x1ous]

Intel HAS improved their gpu. Substantially.

They can NOT strongarm OEMs like they did in the past. That was part of the settlement and it was legally binding. Intel has no slightest desire to test the terms of that settlement.

It wasn't a worry IN THE PAST.
There was not ubiquitous Broadband and streaming A/V IN THE PAST.
AMD Fusion was not challenging it IN THE PAST.

You know what else was part of the settlement? Intel continuing support for PCI Express (or something like it i believe) for six years. So saying AMD/intel will lock out discrete gpu only for their APU wouldn't be possible until at least 2016

 

[Genx87]

Why do people keep comparing Fusion to Nvidia high end products? They wont compete against each other as long as there is a computer gaming market. Fusion threatens sub 100 graphics. I dont expect fusion to ever have the capability as a platform to compete with high end discrete graphics. With that thought in mind Nvidia's mindshare and marketshare in high end discrete gaming graphics wont be affected by fusion. AMD will need to keep pumping out high end products to destroy Nvidia's market there. Fusions has to deal with Intel.

 

[Genx87]

The top AMD APU coming out next year (2011) will have ~ 5670 graphics. That is scheduled to double every year.

2012 = 5770 APU graphics.
2013 = 5870 APU graphics.
2014 = 5970 APU graphics.

Are you contending 5970 performance APU graphics will not replace mainstream discrete graphics in four years?

Where are these APU's magically getting the memory bandwidth to actually perform like a 5970?

 

[Juncar]

Nothing stops Intel from purchasing Nvidia's consumer graphics division nor luring away their engineers. Intel has the resources but not the knowledge and experience for discrete gpu. If they do acquire the needed talent, then you can see more improvements in their APUs and maybe even start seeing Intel discrete graphics card.

 

[Nemesis 1]

Everyone here knows intels IGP history, But what so many fail to realize this is a differant game now. This is for all the marbles. That Intel knows . Everyone assumes AMD because of ATI to win here. I don't, I think Sandy as a starting platiform is an outstanding accomplishment. I look at the performance of the IGP as outstanding and I know we well see at the least 2x better @22nn some think 4x. I think outside chance 6x. What intels little EU performance holds right now is outstanding . I can't wait for intel to grow this tech with the use of large fast Cache, Intel has the right formula now they have to Improve . If intel manages 4x improvement @ 22nm all will take notice. Considering Intel has improved performance 25x since 2007 I think its doable.

With SB Intel got all the hardware on the chip . They needed the 32nm shrink to do that , @ 22 all those parts can be vastly improved . Its interesting for sure,

 

[railven]

console graphics = Crappy graphics.

An 8800 GTX will eat console games for breakfast.



Exactly. Its a moving target, in two years time a 5870 will perform like a lower mid range GPU. Do you realize that Graphic quality is a moving target too?

Lets start with Llano, it will be released in 2011 at 32nm and perhaps it will have the same performance as a HD5670 (I believe it will be near 5570). AMD and Intel change lithography every 2 years so in order to have a 5770 performance in an APU (Double the transistor count) and keep the same die size will have to wait until 2013. If they want to have a 5870 performance in the APU they will have to wait until 2015 for 15nm lithography.

Lets see what a discrete card will be in 2015,

2009 AMDs 40nm 58xx had 2.15 Billion transistors
2011 AMDs 28nm 78xx will have more than 4B Transistors and 2x the performance of 58xx
2013 AMDs 20nm 88xx will have more than 8B Transistors and 2x the performance of 78xx
2015 AMDs 14-15nm 98xx will have more than 16B Transistors 2x the performance of 88xx

So at 2015 we have an APU with 58xx performance and Discrete Graphics with 98xx performance, meaning discrete graphics will be more than 8x faster than 58xx APUs

At that rate APUs will always be Low End Entry Level graphics like Integrated Graphics are now.

I feel these two posts need to be read one after each other and then you'd get a better gist of what the OP is trying to say.

I don't agree that mid-range/high end will disappear, but I'd say they won't be as demanded.

When you considering where one poster said an 8800 GTX could eat console gaming - if the current rend of console gaming dictating developer trends, when a laptop/desktop ships with a console gaming equivalent hardware what motivation does the end user have to upgrade?

Only those that care about IQ will, and from recent conversation with friends (who now do console gaming) it doesn't seem to matter.

Add to it what initiative will developers have to make games granduer if a console port already runs fine on current PC hardware without the need of an expensive discrete card?

I think this is when nVidia/AMD(ATI) will have to step up and give the PC versions polish because if current trends continue - developers won't and publishers won't cover the cost to add extra features.

 

[StrangerGuy]

APUs are not going to even remotely touch the high-end but I do agree Nvidia is in a precarious situation as their low-end GPU market will be wiped out especially in mobile the segment and yet still facing the same stiff competition from AMD's midrange/high-end GPUs.

 

[Dribble]

These threads always over-estimate what gpu on the cpu can do. I expect performance to essentially be identical to integrated solutions that we've had on motherboards for many years. It's just that the gpu has moved from a separate chip on the motherboard, to being integrated into the cpu.

This is done to save money, not increase performance.

While fusion does look good, in reality it's no better then say the nvidia 9400 equipped integrated core 2 motherboards were when they came out. That didn't kill of discrete gpu's despite being sold by the bucket load, and so neither will fusion or sandy bridge.

 

[dguy6789]

Where are these APU's magically getting the memory bandwidth to actually perform like a 5970?

History has shown that as time goes on GPUs become more efficient and just don't need all of that bandwidth; next gen GPUs have matched or outperformed previous gen GPUs with far less memory bandwidth.

Geforce 6600GT was faster than the Radeon 9800 Pro but had a lot less memory bandwidth. Radeon HD 5770 was virtually as fast as a Radeon HD 4870 and had far less memory bandwidth. 5870 is just about twice as fast as a 4870 but it has less than a 50% increase in memory bandwidth compared to it.

All of that said, the first new APU from AMD is where the jump is going to be. The later performance jumps won't be as important immediately, it's the first jump that is. People don't seem to realize or appreciate how much faster a 5670 level IGP is compared to what is available now. We're talking about greater than Geforce 8800 GTS level graphics performance in the cheapest laptops and cheapest desktops from that point on. It will open a new world of gaming to the low end market that was previously unfathomable.