Fusion is native GPU on CPU in 2011/32nm

[Idontcare]

We've learned that Fusion is a native chip that represents a synergy of CPU and GPU on a single chip, at least the one that comes in 2011 is. It won't simply be two dies in MCM on the same chip. AMD has announced two APU products, delayed all the way to 2011 and at least one of them will be a 32nm chip.

http://www.fudzilla.com/index....view&id=10721&Itemid=1

Assuming this is true for the moment, I wonder if this is the intended application for the HiK/MG 32nm bulk CMOS node on the roadmap for TFC. In which case these could be bulldozer cores in the APU?

Or would it make more sense for AMD to put the Stars core into Fusion on bulk CMOS at 32nm?

Also someone with better recollection of history help me out here, assuming the 2011 timeline comes out to be true then how many years will it have taken AMD to get this thing out from time of purchase of ATi? Will it be 6yrs at that point?

 

[soccerballtux]

Merging companies is ridiculously difficult. ATI was never a smart move, it was an acquisition born of hype and overexcited management.
Are you surprised it's taking them this long?

 

[nyker96]

Originally posted by: soccerballtux
Merging companies is ridiculously difficult. ATI was never a smart move, it was an acquisition born of hype and overexcited management.
Are you surprised it's taking them this long?

well they did do 4xxx series pretty well.

 

[Idontcare]

Originally posted by: soccerballtux
Merging companies is ridiculously difficult. ATI was never a smart move, it was an acquisition born of hype and overexcited management.
Are you surprised it's taking them this long?

At the time I thought the move was daring and bold and yes a bit risky. Exactly the kind of thing you need management to not be afraid of doing when they are an underdog with a 5:1 resource deficit to the competition.

In hindsight it does appear to be a dreadful resource drain owing to the near immediate ramifications of having their core cashflow product (Athlon X2) be nearly entirely negated overnight when C2D was launched. Something no one at AMD was clearly envisioning as a plausible outcome of 2006 or else they would have no doubt been doing things differently in the days of the ATI acquisition.

Hindsight is 20/20 so no doubt I agree with just about anyone in that the ATi deal was not in the best interest of AMD's shareholders. It was in the best interest of ATI shareholders, they made out like a bandit.

I am not surprised it will take 6yrs to roll Fusion if that is what the final timeline tally comes to (I still don't know, when was fusion first envisioned?). It takes a solid 4yrs to develop a full-fledged new architecture such as Bulldozer or C2D. So developing a new product such as fusion would take about 4yrs if resourced to do it in such a timeframe, but AMD has no doubt been forced to slim down on the resourcing across the board so a 2yr add-on for diminished resource in combination with the turmoil initially generated by the merger activities can explain a lot.

But still 6yrs is a hell of a long time in the PC world. Impressive that management made a multi-billion acquisition with an eye on the market 6yrs down the road.

 

[soccerballtux]

Originally posted by: nyker96

Originally posted by: soccerballtux
Merging companies is ridiculously difficult. ATI was never a smart move, it was an acquisition born of hype and overexcited management.
Are you surprised it's taking them this long?

well they did do 4xxx series pretty well.

Their drivers are still hit and miss a lot of the time. For instance the 8800GT is faster than the 4850 at Unreal Tournament 3, by a lot. That should not be.
I really have nothing against ATI, though, if you're thinking I'm a fanboy.

IDC-- I guess I'm not surprised at all it's taking them this long. At the time of the acquisition, ATI was still a relatively new player. They hadn't developed any trump-graphics-cards on their own. Their first real success was the Radeon 9x00 series in my opinion; and that wasn't even their own IP, they bought ArtX (the company that designed the GPU for the Gamecube) and then implemented a lot of that design as their 9x00 series. That's how I've heard it at least. So when AMD was talking about ATI, it immediately seemed silly to me, no matter how nice it looked on paper. Nvidia would have been smart (there was no chance of that though), they have proven time and again they can produce a good card (I forgive them for the GF 5900) if they need to.

I'm not concerned about them though, long term they look great. Desktops are becoming faster than anything anybody needs or wants. I think the desktop gaming crowd will become an ever smaller market to cater to. Laptops will be where the new performance race is, and AMD will likely have the best CPU+GPU combination. A real, dedicated GPU (no Larabee silliness) right next to a great [plenty fast for my and most everyone's laptop purposes] CPU.

 

[Acanthus]

It will all depend on where performance and cost land.

If you can get midrange gpu performance, and midrange cpu performance, and be cheap enough to beat out individual parts that can top it...

It's a tall order in my opinion, however Intel is going to have a harder time with the transition if they want any performance for gaming/other gpu accellerated software.

I would think 32nm would be the minimum for cramming multicore CPU + basically a northbridge onto a single package. 45nm would make for a huge chip.

 

[Phynaz]

Originally posted by: nyker96

Originally posted by: soccerballtux
Merging companies is ridiculously difficult. ATI was never a smart move, it was an acquisition born of hype and overexcited management.
Are you surprised it's taking them this long?

well they did do 4xxx series pretty well.

Those parts would have already been in design before the merger.

 

[Idontcare]

Originally posted by: Phynaz

Originally posted by: nyker96

Originally posted by: soccerballtux
Merging companies is ridiculously difficult. ATI was never a smart move, it was an acquisition born of hype and overexcited management.
Are you surprised it's taking them this long?

well they did do 4xxx series pretty well.

Those parts would have already been in design before the merger.

I was assuming the same thing all this time but Anand has got a new story posted on the frontpage detailing the history of the R770 and the timeline suggests it all got started about the same time as the acquisition.

Meaning the R770 was no doubt thoroughly vetted by AMD management in its genesis and AMD management greenlighted the decision maker's plan to get it done. So it would appear that we really can lay the success of R770 at the feet of post ATI/AMD merger management.

 

[Janooo]

Originally posted by: Idontcare

Originally posted by: Phynaz

Originally posted by: nyker96

Originally posted by: soccerballtux
Merging companies is ridiculously difficult. ATI was never a smart move, it was an acquisition born of hype and overexcited management.
Are you surprised it's taking them this long?

well they did do 4xxx series pretty well.

Those parts would have already been in design before the merger.

I was assuming the same thing all this time but Anand has got a new story posted on the frontpage detailing the history of the R770 and the timeline suggests it all got started about the same time as the acquisition.

Meaning the R770 was no doubt thoroughly vetted by AMD management in its genesis and AMD management greenlighted the decision maker's plan to get it done. So it would appear that we really can lay the success of R770 at the feet of post ATI/AMD merger management.

Didn't the acquisition happen in 2006? RV770 design started in 2005. So 2011 is about 5 years for Fusion.

 

[heyheybooboo]

As I understand it the delay on Fusion has been out in the 'wild' for some time.

There was talk that the initial GPU on the die would be a half-node 40nm. I have no clue but I'll guess that AMD's delay could be based upon financials, engineering, heat issues, the market - even SSE5.

The adoption of SSE5 instructions is (or was) needed for Fusion to shine in it's primary purpose - improved parallel processing.

 

[piesquared]

Originally posted by: Idontcare

I was assuming the same thing all this time but Anand has got a new story posted on the frontpage detailing the history of the R770 and the timeline suggests it all got started about the same time as the acquisition.

Meaning the R770 was no doubt thoroughly vetted by AMD management in its genesis and AMD management greenlighted the decision maker's plan to get it done. So it would appear that we really can lay the success of R770 at the feet of post ATI/AMD merger management.

I think this is a very good indication of the direction AMD were and are heading. AMD simply must have known the design choices ATI were taking, prior to aquisition and obviously approved of these decisions. Then, if you remember AMD's presentations of how both companies were persuing very similar strategies, and how the aquisistion of ATI made alot of sense on that merit, I think a picture starts to emerge of what the future might look like.

Took alot of balls for AMD to take that leap of faith, and trust a companies engineers and department heads, whom were just recently aquired.

 

[Idontcare]

Originally posted by: piesquared
Took alot of balls for AMD to take that leap of faith, and trust a companies engineers and department heads, whom were just recently aquired.

I agree. Particularly during the timeframe that the R600 was released, during all those ongoing internal monthly and quarterly project reviews.

 

[Phynaz]

Originally posted by: heyheybooboo
As I understand it the delay on Fusion has been out in the 'wild' for some time.

I don't know if it has been out in the wild for some time, but I did state that Fusion was a 2011 product 7 months ago.

Somebody give me a cookie

 

[Idontcare]

Originally posted by: Phynaz

Originally posted by: heyheybooboo
As I understand it the delay on Fusion has been out in the 'wild' for some time.

I don't know if it has been out in the wild for some time, but I did state that Fusion was a 2011 product 7 months ago.

Somebody give me a cookie

Up to you to decide if those are chocolate chips or ex-lax...

 

[heyheybooboo]

Originally posted by: Phynaz

Originally posted by: heyheybooboo
As I understand it the delay on Fusion has been out in the 'wild' for some time.

I don't know if it has been out in the wild for some time, but I did state that Fusion was a 2011 product 7 months ago.

Somebody give me a cookie

Yer wild, ain't yah?

Ars Technica on SSE5 implementation ...

This three-operand format is one of a number of enhancements, some of which seem to be geared toward turning x86 into an ISA that is worthy of use in a GPU.

 

[hans007]

Originally posted by: Idontcare

Originally posted by: soccerballtux
Merging companies is ridiculously difficult. ATI was never a smart move, it was an acquisition born of hype and overexcited management.
Are you surprised it's taking them this long?

At the time I thought the move was daring and bold and yes a bit risky. Exactly the kind of thing you need management to not be afraid of doing when they are an underdog with a 5:1 resource deficit to the competition.

In hindsight it does appear to be a dreadful resource drain owing to the near immediate ramifications of having their core cashflow product (Athlon X2) be nearly entirely negated overnight when C2D was launched. Something no one at AMD was clearly envisioning as a plausible outcome of 2006 or else they would have no doubt been doing things differently in the days of the ATI acquisition.

Hindsight is 20/20 so no doubt I agree with just about anyone in that the ATi deal was not in the best interest of AMD's shareholders. It was in the best interest of ATI shareholders, they made out like a bandit.

I am not surprised it will take 6yrs to roll Fusion if that is what the final timeline tally comes to (I still don't know, when was fusion first envisioned?). It takes a solid 4yrs to develop a full-fledged new architecture such as Bulldozer or C2D. So developing a new product such as fusion would take about 4yrs if resourced to do it in such a timeframe, but AMD has no doubt been forced to slim down on the resourcing across the board so a 2yr add-on for diminished resource in combination with the turmoil initially generated by the merger activities can explain a lot.

But still 6yrs is a hell of a long time in the PC world. Impressive that management made a multi-billion acquisition with an eye on the market 6yrs down the road.

I think the problem was timing.

All the products Amd has released from the former ATI, ati probably wouldh ave released anyway.


THe timing problem was that ATYT was overvalued at the time. If they had waited until say right now, it would have been a great move (seeing as the stock market is dead).

I think strategically it was a good move. Just AMD needs more money. At the time of the buyout, AMD was profitable and it seemed like something that could work.

 

[Idontcare]

Originally posted by: hans007
I think the problem was timing.

All the products Amd has released from the former ATI, ati probably wouldh ave released anyway.

THe timing problem was that ATYT was overvalued at the time. If they had waited until say right now, it would have been a great move (seeing as the stock market is dead).

I think strategically it was a good move. Just AMD needs more money. At the time of the buyout, AMD was profitable and it seemed like something that could work.

What I find to be interesting is what was revealed today in Anand's article. Basically the ATI folks conceded internally that the R600 was going to lose the halo and thus ATI was about to see some dark and dreary revenue days in the near-term.

That this info was kept from the shareholders is what caused the stockprice to be inflated as it was based on expectations of an earnings projection given by management and analysts who are fed info from management.

So AMD paid a premium for ATI because ATI was sitting on the info that they had a strategy knowing they were about to release a dud of a product that was going to kill their EPS and thus their forward looking P/E.

It's about as absurd of a "strategic manipulation" or "error of omission" as it gets, but all is fair in M&A. Look at what happened to the financial businesses after all their M&A activity. Wamu really did their shareholders a lot of good buying that company with all the CDS on their books.

AMD definitely got suckered into paying more for ATI than the fine folks at ATI knew they were worth, just give a read on Anand's article today. It was engineered to be as it became.

 

[RocksteadyDotNet]

I still don't get how this is going to work? RAM is waaay too slow for mid range graphics performance.

Or are they gonna have some GRam on the board aswell?

 

[Phynaz]

Who said anything about mid-range?

 

[Idontcare]

Originally posted by: RocksteadyDotNet
I still don't get how this is going to work? RAM is waaay too slow for mid range graphics performance.

Or are they gonna have some GRam on the board aswell?

You are thinking Fusion replaces CPU on mobo.

Think fusion replaces GPU on discrete card.

Not saying you buy your fusion card and plug-into PCIe slot. Just saying they have options and that doesn't include the thinking outside the box options like MCM'ed GDDR5.

 

[SickBeast]

I find that AMD always tries to do things in such an elegant way when it comes to engineering.

I'm sure that if they really wanted to, they could "stitch" a CPU to a GPU and put them in the same package.

They probably have their reasons to take their time with Fusion. It will make computers cheaper, which in a sense will cause AMD to make less money.

I'm wondering if Larabee will actually be extremely simple to implement into a CPU. It is just a bunch of x86 cores, after all. I'm wondering if Larabee may in fact be able to function as the core CPU by default. :Q

 

[VirtualLarry]

What segment is this product intended? Desktops? Laptops? All of the above?

 

[Idontcare]

Originally posted by: SickBeast
I find that AMD always tries to do things in such an elegant way when it comes to engineering.

I'm sure that if they really wanted to, they could "stitch" a CPU to a GPU and put them in the same package.

They probably have their reasons to take their time with Fusion. It will make computers cheaper, which in a sense will cause AMD to make less money.

I'm wondering if Larabee will actually be extremely simple to implement into a CPU. It is just a bunch of x86 cores, after all. I'm wondering if Larabee may in fact be able to function as the core CPU by default. :Q

It may seem elegant to the consumer and perhaps to the engineer outside the field but to the product manager and engineer on the project it is a simple consequence of limited resources.

Elegance in engineering is the art of "skipping" interim steps on a product evolution path and timeline.

If the journey is to go from dual-core to quad-core then Step A is to have a dual-core and Step B is to MCM two dual-core's to make a quad-core product and Step C is to integrate the cores and release a monolithic quad-core.

The elegance here would be to skip Step B. It is also the thing you must do if you have 1/2 the resources and are competing on a timeline with someone who has >2x your resources. You have no choice but to attempt to leapfrog the steps and milestones in the evolution pathway.

Doing so of course comes with an increase in risk. By leapfrogging Step B you save some money but you also lose out on the opportunity to increase the knowledge base of your employees and business practices. In addition you risk getting to Step C and finding out it was just too much of change from Step A for your company to execute properly. (kinda like Phenom and 65nm process tech IMO).

This is why one of the benefits of having resources, baby-steps in product development increases costs but decreases risk. Ergo the tick-tock model. I see elegance there too, elegance in optimizing project management to the resources and risk management practices that maximize shareholder value.

(same elegance in motion behind Itanium skipping 45nm node altogether and jumping from 65nm to 32nm directly)

So we see elegance in AMD's methodology where the reality is it is simply a necessity, a natural consequence of good project management in a resource scarce environment.

Originally posted by: VirtualLarry
What segment is this product intended? Desktops? Laptops? All of the above?

My assumption has been it is geared for low-budget applications in all segments. Your $300 emachines and your $400 laptops. Potentially aimed at the netbook market as well depending on how well the thermal envelope can be scaled on TFC's 32nm bulk hik/mg process space.

Maybe Viditor can help us answer your question?

 

[heyheybooboo]

Originally posted by: VirtualLarry
What segment is this product intended? Desktops? Laptops? All of the above?

I think the intention was (or is) to utilize the GPU as a 'coprocessor' - utilizing its performance advantage over the CPU in certain arithmetic functions and parallel processing.

 

[Phynaz]

Originally posted by: VirtualLarry
What segment is this product intended? Desktops? Laptops? All of the above?

Initially laptops..then we will see.

Never expect it on the high end. You would be talking a 300w TDP.