ATI Talks Lithography

[zidane1strife]

I don't really see why they will need more ram than the current top of the line gpus. Esp since that ram is used for fsaa. I just feel it will be to expensive and sony will end up loosing alot of money off these consoles .

My guess is the ps3 will at the very least exceed(due to the fact that it has 2chips that heavily contribute to gphx calc, and it's top tech, I'd say significantly exceed.) the perf of 2005 gpus, having the same amount of ram as gpus from 2yrs prior is obviously inadequate.

 

[London Geezer]

i know u guys love making comparisons between GPU's and PS3.. but we have to remember how different these architecture they will be... really, when you think about it, they might even end up using different rendering techniques altogether so...
it started with PS2, which you can't really compare to anything on the PC market (and associates). it will be even more so with PS3

 

[Nexiss]

Anyways, for Nexiss I say that ATI will offer a 90nm because that all it seems like they're offering, and MS is accepting that kind of risks because they're biting.

My point is, you don't know what ATI is offering. All we know is that it is a licensing deal. Second, you seem to say that MS could, and imply that they should, have gone with something better, when in fact you have no idea what they are willing to do or what their targets are or what ATI is offering them in terms of tech.
I think Dave said it perfectly:

Basically we don't yet know exactly how it will pan out, however as this is a licensing deal for "IP" its essentially up to MS how they choose to implement that IP.

 

[nonamer]

Anyways, for Nexiss I say that ATI will offer a 90nm because that all it seems like they're offering, and MS is accepting that kind of risks because they're biting.

My point is, you don't know what ATI is offering. All we know is that it is a licensing deal. Second, you seem to say that MS could, and imply that they should, have gone with something better, when in fact you have no idea what they are willing to do or what their targets are or what ATI is offering them in terms of tech.
I think Dave said it perfectly:

Basically we don't yet know exactly how it will pan out, however as this is a licensing deal for "IP" its essentially up to MS how they choose to implement that IP.

You're making too big of a deal of your point. It's all speculation, so don't try to discredit the speculation you may not like.

Besides, it's not like we're blowing smoke, this is Informed Guessing.

David_South#1, won't dual GPU raise heat? Unless you substantially underclock the GPUs, which means that a dual GPU provides less than double the power. Remember this is a console, not some PC rig. You can't put a lot of really hot things in there. You may be pretty sure that the Xbox2 will be dual GPU, but I'm pretty sure it won't. At best it will be dual core CPU, maybe a similar configuration for the GPU. However, there will be only one chip for the CPU and one chip for the GPU. Anything more will skyrocket the cost well in excess of even the PS3. Plus I'd doubt it would do much to increase performance due to the heat issue.

Otherwise I find myself in agreement with you. The K8's are much cooler than what Intel is demonstrating at the 90nm level, and they have the dual core and 64bit ability(which I doubt Prescott/Tejas will have, despite recent rumors). The entire Netburst architecture from Intel is simply the wrong idea for a console: Too big and too much focus on clockspeed. Nehalem, the replacement for the P4, looks to be the same. Only the P-M from Intel could really make it into the Xbox2. And yeah, Cell/PS3 is way overblown, but then again there isn't much to talk about for GC2 or XB2.

By the way, if the R500 is coming out in 2004, won't it be DX9?

 

[Saem]

I think the MS and ATI licensing business is basically going to work out like the following...

ATI will build the part and then MS will license that design. This way ms can say when, where and how it's built in an effort to control costs while paying ATI less than they would in their current Nvidia situation.

 

[jvd]

I think the MS and ATI licensing business is basically going to work out like the following...

ATI will build the part and then MS will license that design. This way ms can say when, where and how it's built in an effort to control costs while paying ATI less than they would in their current Nvidia situation.

Yes that is obvious from the deal. What isn't is who will do the back end. Which fabs will fab the chips and which design will it be based off.

Now 256megs of rambus ram which will be top of the line in 2005-2006 . A huge complex chip like cell. A gpu . Sound chip. System ram. Blue ray and many other things will be a huge cost in 2005. I would wager much much more expensive than 500$.

With gpu's of the time they will be 500$ in price too. Why you feel that the ps3 will be faster than these beasts I have the foggiest. You will see better results off the bat for the ps3. But thats because pc game deisgn is behind the tech curve. Where jsut recently we have hit about the year mark for games to be coded for the newest hardware . I.e Half life 2 - 9700pro . THe 9700pro was released in august of 2002. Half life 2 will follow before 2004 (unless its delayed again) .

Now current gpus are meant to make money off the design and cost less than the 500$ they are charging. IF you feel sony who is in number 1 spot is willing to take huge losses on hardware to stay that way. You must figure ms will take huge losses on hardware for a second round to maintain that. So we could very well see an intel 64 bit cpu . 512megs of ram for the gpu and a 128 megs of system ram. Which would be more than enough for the system. Or perhaps if they go dual gpu . They will have 1024 megs of ram. Half for each chip .

I know we will see some very interesting tech coming in the xbox 2. It wont just be an off the shelf small form pc .

 

[Panajev2001a]

The present IBM STI chip or the only working model I know of is only 550Mhz.
And is not near the scale most people are ranting about.

Let me make two guesses:

1) it is not a 65 nm chip

2) it is a prototype and not the final silicon of the 2005 CELL chip which will finish in PlayStation 3 ( shipping in late 2005 ).

What you are hearing about it is probably the CELL implementation that Kutaragi and the rest of STI are going to demosntrate March next year ( you will see some improovements as well ).

 

[zidane1strife]

Now 256megs of rambus ram which will be top of the line in 2005-2006 . A huge complex chip like cell. A gpu . Sound chip. System ram. Blue ray and many other things will be a huge cost in 2005. I would wager much much more expensive than 500$.

With gpu's of the time they will be 500$ in price too. Why you feel that the ps3 will be faster than these beasts I have the foggiest.

Well, IIRC the ps2 launched with a price equiv to about US 380$ in japan, and it was sold at an approxx... 200$loss... IOW it costs approxx 600$...

The ps3 unlike d'ps2 will apparently benefit from far more R&D(more engineers, more cash, more exp, etc...), it is a two chip solution, and two chips that will likely be VERY LARGE.

While on the pc we'll likely have sub30Gflops cpus and tons of bottlenecks and far lower b/w compared to the ps3. It is my understanding that pc gpus cannot be sold at a $100-200+loss, and thus their price is not real cost to produce, but the price at which it is sold with the intention of generating a profit...

 

[Vince]

The present IBM STI chip or the only working model I know of is only 550Mhz.
And is not near the scale most people are ranting about.

Whoa, you sure its [as defined as any manifestation of it's pipelines] that fast this early?

 

[David_South#1]

zidane1strife,
“My guess is the ps3 will at the very least exceed due to the fact that it has 2chips that heavily contribute to gphx calcâ€
2Chips? I’m guessing some at this forum think Patent Figure 6 is for a PS3?
That or you suspect a different LSI STI then cellular design will happen.

london-boy,
“I know u guys love making comparisons between GPU's and PS3..
but we have to remember how different these architecture they will be... really, when you think about itâ€
Really when you think about it, what we have to remember is that there isn’t a PS3 to compare with.

Nonamer,
“David_South#1, won't dual GPU raise heat?â€
Not the way you might be thinking. The heat is spread across a larger area for it cols easier.
Look at an XGI offering. Compare that to the cooling solutions ATI & NVDA are offering.
I don’t agree with you’re suspicions about cost or clock either.
Smaller less complex dies are easier to produce and become cheaper with higher yield rates.
Take some time and watch XGI for a while. We’ll both learn something from that.

â€By the way, if the R500 is coming out in 2004, won't it be DX9?â€
I don’t think R500 will be out in 2004. Early 2005 seems more likely.
DX9 as opposed to what? This is just a joke, but Xbox2 will probably target DX9.5.

Jvd,
Sorry I missed that. Did you say something about the future being very expensive right now?

Panajev2001a,
1. Pretty safe guess. Then again 90nm is probably what a PS3 will be based or start on.
2. Things will indeed improve. And what type or form of LSI, PS3 will be is unknown.
Yep and yep. Ken’s already shown this to his friends (the Board of investors).
I’m still disturbed by what Shin’s disappearance means.

zidane1strife,
> Pricing and loss is always a guess of the costs over time.
>“The ps3...is a two chip solution, and two chips that will likely be VERY LARGE.â€
I didn’t want to take sides with Deadmeat. But if you think Fig6 is a PS3...take a time out.
>The Console and PC both have a very nice outlook this next time. DDR2 in graphics card may reach 50GB/sec.
To be honest bottlenecks are more of a concern for a console since it has a fixed design.
(And because new PC standard like PCI Express, larger FSBs, new RAM are in the works now.)

Vince,
At this time I believe is has only 4-APU. (I bet you were thinking of 32. )

 

[Vince]

At this time I believe is has only 4-APU. (I bet you were thinking of 32. )

I was figuring less, couldn't you tell by my comment?!. It's quite irrevelent at this early stage. Just that the design is functional is all that matters at this point, they have 2 years to take this most-likely buggy "IP" and make a 65nm IC out of it for PS3.

Oh, and a dual-chip solution isn't what I'd consider an economically intelligent long-term decision - heat be damned.

 

[Panajev2001a]

At this time I believe is has only 4-APU. (I bet you were thinking of 32. )

I was figuring less, couldn't you tell by my comment?!. It's quite irrevelent at this early stage. Just that the design is functional is all that matters at this point, they have 2 years to take this most-likely buggy "IP" and make a 65nm IC out of it for PS3.

Oh, and a dual-chip solution isn't what I'd consider an economically intelligent long-term decision - heat be damned.

As far as Figure 6 is concerned, for the BE ( maybe the VS might use less PEs ) that should be a probable configuration: I do not see them launching before Q4 2005 and their 65 nm process should have had good yelds for a while and that would also mean that selling a basically not really economically viable chip would not be too bad for SCE as the loss would have only be taken until the 45 nm die shrink is ready which should not take more than 1-1.25 years from the Q4 2005 launch.

 

[Panajev2001a]

I was figuring less, couldn't you tell by my comment?!. It's quite irrevelent at this early stage. Just that the design is functional is all that matters at this point, they have 2 years to take this most-likely buggy "IP" and make a 65nm IC out of it for PS3.

You do have a good point there, if the launch is thought to be Q4 2005 ( approximately ) we would have to have good volume mass-production start in Q2 2005 at the latest.

We are in Q3 2003 right now and that leaves just about ~2 years to debug the architecture and make the final 65 nm CELL chips for PlayStation 3.

Going by the very modular nature of CELL once the debugging phase is pretty much complete ( which involves the 1,024 bits bus inside each PE, the APU's design, the APU's Register File and SRAM based Local Storage, the design of the PU, the design of the DMAC and the cross-bar memory switch, etc... ), adding more APUs and PEs should not be impossibly hard: after you can build a single PE with 8 APUs pretty much bug-free, the step to 4 PEs can be done.

Work on the APU, work on the PU, work on a PE with 1 PU, 1 APU and 1 DMAC... then add 1 APU ( each time you add components you need to do more testing/debugging ), 2 APUs, etc... until you have 8 of them.

Then work on 2 PEs per chip, then 3 PEs per chip and then 4 PEs per chip.

 

[jvd]

Jvd,
Sorry I missed that. Did you say something about the future being very expensive right now?

Sorry what are you talking about ? I'm talking about costs of products in the future. Or are you going to arguee that the top of the line rambus ram will be cheap as hell. Or that a brand new cell chip that hasn't been used in any other tech will be cheap too. All this stuff in both systems will cost a bloody lot . I'm just wondering how much everyone expects each player to lose . We already know ms is willing to loose alot of money and to continue doing it. We know sony doesn't mind loosing a little bit of money if the price quickly drops down so they break even. See examples of ps2 and xbox for proof of this concept . What did you find wrong about my comments ?

 

[Josiah]

I don't really see why they will need more ram than the current top of the line gpus. Esp since that ram is used for fsaa. I just feel it will be to expensive and sony will end up loosing alot of money off these consoles .

My guess is the ps3 will at the very least exceed(due to the fact that it has 2chips that heavily contribute to gphx calc, and it's top tech, I'd say significantly exceed.) the perf of 2005 gpus, having the same amount of ram as gpus from 2yrs prior is obviously inadequate.

It should, most consoles are 1-2 years ahead of the curve when they are released. I'm betting though that an R500/R550 could handle a first gen PS3 game. If PS3 is really a software renderer you could get away with less transistors to accomplish the same tasks in hardware (the downside is you lose flexibility). In the same way that you can use a $2,000 PC to emulate an N64 in software, or you can just buy an N64 for $50 (extreme example I know, but you get my point).

 

[Vince]

If PS3 is really a software renderer you could get away with less transistors to accomplish the same tasks in hardware (the downside is you lose flexibility)

Whoa... who ever said that?

 

[Panajev2001a]

If PS3 is really a software renderer you could get away with less transistors to accomplish the same tasks in hardware (the downside is you lose flexibility)

Whoa... who ever said that?

I think he meant "... with HW texture filtering and AA sampling"

 

[Josiah]

I thought that was the concensus. Some common rendering functions would be supported in hardware, with the majority being completely software based.

 

[jvd]

I don't really see why they will need more ram than the current top of the line gpus. Esp since that ram is used for fsaa. I just feel it will be to expensive and sony will end up loosing alot of money off these consoles .

My guess is the ps3 will at the very least exceed(due to the fact that it has 2chips that heavily contribute to gphx calc, and it's top tech, I'd say significantly exceed.) the perf of 2005 gpus, having the same amount of ram as gpus from 2yrs prior is obviously inadequate.

It should, most consoles are 1-2 years ahead of the curve when they are released. I'm betting though that an R500/R550 could handle a first gen PS3 game. If PS3 is really a software renderer you could get away with less transistors to accomplish the same tasks in hardware (the downside is you lose flexibility). In the same way that you can use a $2,000 PC to emulate an N64 in software, or you can just buy an N64 for $50 (extreme example I know, but you get my point).

Sorry thats just not true. When the ps2 launched we had geforce 2 cards and 1ghz cpus. Now we have games like doom3 that will run on those cards at 30fps 640x480 which is what the ps2 runs games at . The doom3 quality even on those cards are on par with the visuals put out by the ps2 .

 

[Vince]

I thought that was the concensus. Some common rendering functions would be supported in hardware, with the majority being completely software based.

Well, I understand what you're thinking, but I must sincerly disagree. Although, I must confess this is probably more of a linguistic espression problem in articulating our thoughts.

So, I'll just posit a question, Will the NV4x and R4x0 be software based? The NV5x and R5x0? Because, already the "software" rendering we're "envisioning" for Cell's APUs is happening to an extent in microprograms/shaders. I think the hardware-software duality argument is increasingly like trying to draw a line in a massive pool of grey when it comes to shaders.