Official: ATI in XBox Next

[Joe DeFuria]

IMGTEC did, apparently MS really liked what they saw from the technology but weren't so sure about IMGTEC's ability to get the tech out without any delays.

I heard something similar. Not sure about how much MS liked the tech or not, but I heard that IMGTEC was in the running, but was dropped because it was seen as too high a "risk".

Which is a little unfair considering IMGTEC's only console dealing so far had no delays AFAICR.

Actually, AFAICR, there was DC delays. Very low initial shipments due to low volumes of the PowerVR chip. But I suspect the major skepticism would come from IMGTEC not having actually produced a modern GPU for some time. No T&L even...let alone pixel shaders.

Which brings me back to saying "I told you so!" (Not to you Teasy...to IMGTEC.) Creating and shipping a high-end PC chip has more importance than just the sales that you get from it. In fact, the sales of the high-end chip is probably the least, financially speaking, important. The mind-share you get with it allows you to drive the lower end....and then it establishes you as a "real player."

Imagine if IMGTECH had taken it upon themselves to ship a high-end Series 4. We may very well have been looking at ImgTech in the next x-box.

This isn't official info BTW, just what I've heard through the grapevine.

Same here!

 

[Magnum PI]

I see that it's pointless to debate on this issue blablabla

that's just your arguments that are pointless.

 

[Joe DeFuria]

I see that it's pointless to debate on this issue against DeFuria and Baumann so I should stop now. Just to let you know that this is a total waste of time and I'm stopping right now. You can claim whatever you from this post but I don't care. Enough is enough.

All we ask is for speculation to at least have a basis in fact. Speculation based on....baseless assumptions and other speculation is, well, pointless indeed.

 

[Megadrive1988]

Actually, AFAICR, there was DC delays. Very low initial shipments due to low volumes of the PowerVR chip. But I suspect the major skepticism would come from IMGTEC not having actually produced a modern GPU for some time. No T&L even...let alone pixel shaders.

Which brings me back to saying "I told you so!" (Not to you Teasy...to IMGTEC.) Creating and shipping a high-end PC chip has mor importance than just the sales that you get from it. In fact, the sales of the high-end chip is probably the least, financially speaking, important. The mind-share you get with it allows you to drive the lower end....and then it establishes you as a "real player."

Imagine if IMGTECH had taken it upon themselves to ship a high-end Series 4. We may very well have been looking at ImgTech in the next x-box.

I agree 100% here.

If Imgtech had their 4 pipeline, on-chip T&L (off-chip ELAN was amazing)
highend Series 4 part out in 2001 or 2002, they could have been in Xbox2
-Since MS had seriously concidered GIGAPIXEL for the original Xbox, without them even having one product out, then MS would no doubt have
given PowerVR a chance, since they've had several PC, console and arcade chips in widespread use. just nothing modern. And PowerVR3 was too old (lacking T&L) to compete at the highend, that's why it came out as a value product.

 

[Joe DeFuria]

-Since MS had seriously concidered GIGAPIXEL for the original Xbox....

Incidentally, I always believed that GigaPixel was not chosen for the same reason: too risky on the basis of not having demonstrated a mass produced part based on it's tech. I certainly can't fault MS for choosing nVidia, but man, I would have loved to have seen them attempt it with GigaPixel tech...

 

[Megadrive1988]

Incidentally, I always believed that GigaPixel was not chosen for the same reason: too risky on the basis of not having demonstrated a mass produced part based on it's tech. I certainly can't fault MS for choosing nVidia, but man, I would have loved to have seen them attempt it with GigaPixel tech...

Yes, and now another chance for GigaPixel tech to be used in a console is lost, since 3Dfx had swallowed GigaPixel prior to Nvidia swallowing 3Dfx


Maybe GigaPixel+3Dfx+Nvidia in PS3's rasterizer?

 

[whql]

Barrons article on XBox 2:

ATI's Moment of Glory

TEN MINUTES AFTER THE MARKET OPENED THURSDAY, ATI Technologies became the world's largest computer graphics company...the stock market's measure, at any rate. Investors thrilled at news that the Canadian chipmaker would supply the graphics technology for Microsoft's next generation Xbox gaming console. ATI shares briefly popped up 12%, to 13.81, valuing the Markham, Ontario, firm at $3 billion. That exceeded the $2.8 billion valuation of Nvidia, the current Xbox graphics supplier and ATI's bitter rival. By week's end, ATI shares were 13.17 and Nvidia's were 16.15.

Rumors of the Xbox win have helped ATI shares more than double in recent months, while shares of Nvidia have dropped by about half. That's narrowed the hefty premium investors had been paying for Nvidia's predicted earnings, in comparison with ATI's (see "A Graphics Shoot-Em-Up," July 7).

While ATI will recognize some engineering payments from Microsoft in its current September quarter, neither firm's financials will show much immediate impact. Sales of Xbox chips are a material part of Nvidia revenue -- 19% in the recent July quarter. But the Xbox's successor isn't due for a couple of years and Nvidia spokesman Derek Perez says that his company isn't changing any financial guidance.

ATI and its fans say the Microsoft deal shows that ATI has ousted Nvidia as the performance leader of the graphics business. "This agreement cements ATI's position as the prime graphics supplier for the future of the games industry," crowed chief executive K.Y. Ho in the press release.


Power Supply: The blackout across wide stretches of the U.S. and Canada didn't stop the Nasdaq Composite from gaining 3.5% for the week, to close at 1,702. Tech stocks have surged so powerfully in recent months that Credit Suisse First Boston downgraded the entire group last week, from "overweight" to "neutral."

That boast overlooks the commanding lead that Sony and its PlayStation2 enjoy in the electronic gaming business. Sony continues to develop its own graphics technology, even if ATI will be supplying graphics wizardry for future Xboxes from Microsoft and for GameCubes from Nintendo. The $8 billion worth of revenue that Sony pulled in from the PlayStation last year amounted to 12% of the company's total revenue. But the gaming products kicked in 57% of Sony's operating profit.

The next generation PlayStation, due in 2005 or 2006, will step up Sony's offensive to dominate home entertainment, and perhaps, home computing. A patent granted to Sony in February (U.S. patent number 6,526,491) suggests that future PlayStations could be multiprocessing monsters. The patent humbly aims to redesign the Internet and all the computers attached to it, replacing them with a grid of identical processor modules. Handheld devices might have just one of these processing modules, while a network server computer would have a bunch. When needed, these modules could band together -- even across a network -- to crunch through particularly heavy computing jobs. Important stuff, like a 3-D war game.

An overwhelming installed base of computers stands in Sony's way. But maybe a world change starts at the doorstep. Networked Sony devices around a home could implement Sony's multiprocessing vision, with souped-up PlayStations participating in the computing power grid.

Can ATI help Microsoft battle Sony's plan for an army of graphics processing power? Yes, says ATI financial chief Terry Nickerson. ATI's graphics savvy could make the "Xbox Next" into a platform that attracts

 

[zidane1strife]

From the speculation I made above and assuming that Nvidia was thinking of totally custom designing a 65nm GPU right from the start and ready by 2005/2006. This would definitely be powerful enough to match PS3 and it would also be very expensive and draining, and this is what I'm imagining is Nvidia's plan.

Well, that would be the case if there was something seriously wrong with the design of the ps3(suppah difficult, massive bottlenecks, etc.). It might not be otherwise, you have to remember that the ps3 will likely have(if ps2 is anything to go by.) massive chips for that process, chips that would only be viable/profitable in the long term with 45nm, in other words it is nigh 45nm design on a 65nm process at a loss. In addition from what’s been going around it’s not one, but two chips that will be massively contributing to the gphx calcs.

IOW it’s not as clearly cut, if the design of the ps3 is an excellent one, the likeliest scenario is that it would outperform a single 65nm gpu, since it would be two chips against one, and they’d be nigh 45nm.

Assuming there is no gigantic prob. In the design(massive bottlenecks, etc), I’d say you’d need another 2 chip solution(3 chips if it’s pc derivative, two for the gpu and one for the cpu), or 30nm or below to significantly exceed performance… unless it is on-paper theoretical, in which case you could sacrifice features and edram to push raw specs to the limit… but at a serious cost in terms of real-world performance…

that's why I think Xbox 2 will ultimately be out in 2006. there are many reasons for 2006 instead of 2005.

If that happens… it’ll be a year after both sony and Nintendo have both been on the market… and assuming Nintendo doesn’t launch another purple box with a handle(WTF?!?), they’ll likely acquire a lot of their lost market share back.

I'm not entirely sure what you're saying, but expecting _any_ 65nm product to start shipping in H1/05 is rather optimistic, if not illusory.

The likeliest launch date for the ps3 is march 2005… It is likely to feature nigh 45nm parts(read above paragraphs.).

CELL is going to have a lot of eDRAM so yes it would benefit greatly from 0.065u. as a large chunk of die space will be needed for all that eDRAM. However assuming the next XGPU does not contain any eDRAM, it will have comparable logic area using only 0.09u because there's no eDRAM to used up die space.

Cell’s eDRAM occupies less space than common eDRAM, and it’s a chip that will likely be beyond what’d be viable in the long term through the use of 65nm.

From what I’ve heard, not containing eDRAM further separates the theoretical performance from the REAL-world performance.

 

[bbot]

I used to believe until today that the cell chip would be 1tflops. Of course there was this Japanese article of a speech given by Ken Kutaragi at TGS 2002, that indicated that a single cell-chip will be only 32gflops.

http://www.watch.impress.co.jp/game/docs/20020921/tgsf.htm

The problem was that the patent for cell had a date on it of Sept 26, 2002. I thought that was when it was filed. Today, I looked at the patent and found out that it was actually filed March of 2001. So the infomation in Kutaragi's speech is newer than the information in the patent by more than one year.

The bottom line is I'm no longer sure that a single cell chip will be the 1tflops super chip that everybody seems to think that it is. A Pentium 4 with a high clock speed could easily match the computational power of a 32gflops cell chip.

 

[zurich]

 

[incurable]

I'm not entirely sure what you're saying, but expecting _any_ 65nm product to start shipping in H1/05 is rather optimistic, if not illusory.

The likeliest launch date for the ps3 is march 2005… It is likely to feature nigh 45nm parts(read above paragraphs.).

BS. Plain and simple BS. Let me explain:

To have any kind of volumes at a launch in March/05, chip production should have started late Q3/04 at the latest. Before production can commence, you need time for installation & qualification of the equipment as well as pilot production runs to gain experience and samples to seed the developers. So, if you want to start the production in late Q3/04, you better get the equipment delivered in Q1/04 (very optimistic).

Now just tell me, which company is planning to ship production-worthy lithography equipment for the 65-nm node in very early 2004? (that is 5 months from now, max.)

cu

incurable, still wondering about the term "nigh 45nm parts" as the 45 nm node is 4 (!) years away

 

[PC-Engine]

Cell’s eDRAM occupies less space than common eDRAM, and it’s a chip that will likely be beyond what’d be viable in the long term through the use of 65nm.

No duh, however the advantage over 1T-SRAM-Q is only ~ 25% and since CELL will likely have a lot of eDRAM something like 64MB, that amout of eDRAM per chip will require a significant amount of die space. So at 0.065u you end up with the same amount of logic area as a chip which has no eDRAM made on a 0.09u process. In other words the GPU going into Xbox2 can still be made on a 0.09u process and still allow roughly the same amount of logic area because there's no 64MB of eDRAM to use up precious die space

If you have a lot of extra baggage then you will need more room.

 

[Nexiss]

bbot,

It's likely just a confusion of terminology. What is probably being referred to is a single APU (which is described as being capable of 32GFlops at 4GHz). Put 8 APUs and some other stuff together and you get a "processing element" at 256 GFlops, and 4 PEs for a "Broadband Engine" capable of 1 TFlop.

 

[arhra]

bbot,

It's likely just a confusion of terminology. What is probably being referred to is a single APU (which is described as being capable of 32GFlops at 4GHz). Put 8 APUs and some other stuff together and you get a "processing element" at 256 GFlops, and 4 PEs for a "Broadband Engine" capable of 1 TFlop.

If they could produce a single-chip which runs at 1tflop, and which is going to be in the ps3, why do they need a whole cabinet (something like a 40U server rack, i'd assume) to reach 16tflops? Is the PS3 going to be the size of a 2U server? Would make the xbox seem positively svelte...

 

[arhra]

aargh. double post.

 

[Entropy]

To have any kind of volumes at a launch in March/05, chip production should have started late Q3/04 at the latest. Before production can commence, you need time for installation & qualification of the equipment as well as pilot production runs to gain experience and samples to seed the developers. So, if you want to start the production in late Q3/04, you better get the equipment delivered in Q1/04 (very optimistic).

Now just tell me, which company is planning to ship production-worthy lithography equipment for the 65-nm node in very early 2004? (that is 5 months from now, max.)

You aren't all that optimistic.
193-nm immersion can be used down to 45nm, and leveraging 193-nm would seem to be an attractive path to take for something like this (armchair perspective warning!). Do I believe in a volume launch in March 2005? Absolutely not. I can believe that they may be aiming for it though, but that's something else altogether.

Entropy

 

[zidane1strife]

still wondering about the term "nigh 45nm parts" as the 45 nm node is 4 (!) years away

Nigh 45nm means that the design is very expensive and hard to implement at 65nm, and will likely cause massive losses... until it is moved into 45nm were it will be profitable or at least stop further losses.

As for the launch of ps3 likely announcement is early 2004, thus we have only to wait for a few months to find out if they're on schedule(although from what I've heard it appears they're on schedule.).

No duh, however the advantage over 1T-SRAM-Q is only ~ 25% and since CELL will likely have a lot of eDRAM something like 64MB, that amout of eDRAM per chip will require a significant amount of die space. So at 0.065u you end up with the same amount of logic area as a chip which has no eDRAM made on a 0.09u process. In other words the GPU going into Xbox2 can still be made on a 0.09u process and still allow roughly the same amount of logic area because there's no 64MB of eDRAM to use up precious die space

That might be the case if we were talking about one chip... but we're talking about two chips... logic components in each are likely to be at least .5B transistors. We have two chips...

 

[PC-Engine]

That might be the case if we were talking about one chip... but we're talking about two chips... logic components in each are likely to be at least .5B transistors. We have two chips...

So Xbox2 won't have a cpu?

 

[zidane1strife]

So Xbox2 won't have a cpu?

Well, I mean one that contributes massively to the gphx calcs. For all we know it might be a custom one, thus my answer of 2 chips if they're both doing massive contributions, or 3 as in two gpus if the cpu was not to contribute as much do to being pc derivative.

From what I've heard the BE is very capable in terms of gphx related calcs...

 

[nonamer]

still wondering about the term "nigh 45nm parts" as the 45 nm node is 4 (!) years away

Nigh 45nm means that the design is very expensive and hard to implement at 65nm, and will likely cause massive losses... until it is moved into 45nm were it will be profitable or at least stop further losses.

Very massive indeed... The only "nigh next generation fab process" chip I'm aware of is ths GScube (it was something like 450+ mm^2 or some huge number like that). If it was something like that you can kiss goodbye to a PS3 in 2005 since it would lose way too much money.