ATI Talks Lithography

[Vince]

What does MS have to do with the speculation that the PC R500 part may be built using a 110nm process? Haven't we already been over the points of IP licensing? It seems, as well, that if R500 is still scheduled for 2004 then there is, unless my maths fails me, another year before 2005?

So, are you inplying that an entity such as Microsoft will take ATI's "IP" - which I think they said will be R500 based for XBox2 (correct if wrong) - and basically do the entire back-end of the design themselves? Which, to increase concurrency of design to absorb the influx in possible gates will require, at the least, some physical resuse scheme. Not to mention this is less effecient than designing an architecture native to a given process.

So, something doesn't add up here. Are you implying that the XGPU isn't R500 based and is a very custom extention?

I mean, just saying "Licensing their IP" or some such comment to that effect is useless IMHO. You need to do something to that intangible "IP" to create a tangible "IC" out of it. There are looming questions over how to take this mystical "IP" (what form?) and manifest it in a new processor - which you seem to be implying is more advanced than that which it's derived from.

And, my math is saying 6 months between 2H 2004 and when a console needs to be in production for an early 2H2005 launch. Too many voices...

 

[cthellis42]

Now it'll be even more interesting to see who Microsoft is choosing to fab the GPU, since this is the reason they wanted an IP deal anyway. I was frankly expecting the R420 generation to play out longer, so had assumed R500 would be lower than 110nm at least. But if even R500 will be out next year...

Since we haven't heard much of what Nintendo's designs are and how ATi is supplying the GPU for N5, we can only mull over Xbox2 right now. By all accounts the console is still slated for 2006, so that's a LOT of time between then and R500's launch. Since it's supposed to scale down to 90nm, I assume it will appear in Xb2 in at least that form, quite possibly also modified towards more advanced features depending on who Microsoft goes to for fabbing (one would assume Intel, but you know what happens when you assume...). They COULD perhaps even push to 65nm, but that seems a little impractical to start on at this stage in the development format they've chosen. (Maybe moving to 65nm somewhere in the lifespan is in the works, though.)

Kind of interesting, though... Is Microsoft trying to push early, taking STI at their word of getting out in 2005 and pushing to launch side by side? Are they wanting to use solid but older tech so as to not be the money-sink Xbox itself has been? Are they going to license the IP and really try to push the process themselves, since they'll control that end fully and won't have to depend on another company for it?

Hadn't expected R500 in 2004, though, that's for sure.

 

[Dave Baumann]

So, are you inplying that an entity such as Microsoft will take ATI's "IP" - which I think they said will be R500 based for XBox2 (correct if wrong) - and basically do the entire back-end of the design themselves? Which, to increase concurrency of design to absorb the influx in possible gates will require, at the least, some physical resuse scheme. Not to mention this is less effecient than designing an architecture native to a given process.

Why are we repeating the same old discussions? 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. This isn't exactly someting that hasn't occured before - PowerVR designed the "IP" for PowerVRDC and NEC were the ones that did the back end for SEGA, similar thing with the GC.

 

[Vince]

Basically we don't yet know exactly how it will pan out

Works for me. heh.

 

[jvd]

First off there is no way they will make a cost effective 256mb edram chip.

Second of all with 32megs or more of edram there is no need to have 256 megs of main ram . I don't see the point with such a huge bandwidth on the on die ram.

The issue isn't the bandwidth (well, that is an issue, but a different one), it's the space. Personally, I think even 256MB of main RAM is too little, unless they are planning on going with some sort of memory model similar to the NGC. Even if we will have enough power to move towards more procedural methods, space can still be taken up rather quickly.

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 .

 

[nonamer]

Why do I have the feeling that this will be another 10 page debate marathon? No reason, just seems like all the right people are in place.

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.

I keep on hearing the Intel as fab idea. I find this impossible. Intel isn't a foundry and they're very reluctant to let other access to their fabs.

 

[jvd]

Now it'll be even more interesting to see who Microsoft is choosing to fab the GPU, since this is the reason they wanted an IP deal anyway. I was frankly expecting the R420 generation to play out longer, so had assumed R500 would be lower than 110nm at least. But if even R500 will be out next year...

Since we haven't heard much of what Nintendo's designs are and how ATi is supplying the GPU for N5, we can only mull over Xbox2 right now. By all accounts the console is still slated for 2006, so that's a LOT of time between then and R500's launch. Since it's supposed to scale down to 90nm, I assume it will appear in Xb2 in at least that form, quite possibly also modified towards more advanced features depending on who Microsoft goes to for fabbing (one would assume Intel, but you know what happens when you assume...). They COULD perhaps even push to 65nm, but that seems a little impractical to start on at this stage in the development format they've chosen. (Maybe moving to 65nm somewhere in the lifespan is in the works, though.)

Kind of interesting, though... Is Microsoft trying to push early, taking STI at their word of getting out in 2005 and pushing to launch side by side? Are they wanting to use solid but older tech so as to not be the money-sink Xbox itself has been? Are they going to license the IP and really try to push the process themselves, since they'll control that end fully and won't have to depend on another company for it?

Hadn't expected R500 in 2004, though, that's for sure.

I believe the r400 was pushed off or cancelled and the r420 is its replacement with most of the r400 tech finding its way into the r500. So it is possible that ati is already working on the r600.

 

[Panajev2001a]

So, are you inplying that an entity such as Microsoft will take ATI's "IP" - which I think they said will be R500 based for XBox2 (correct if wrong) - and basically do the entire back-end of the design themselves? Which, to increase concurrency of design to absorb the influx in possible gates will require, at the least, some physical resuse scheme. Not to mention this is less effecient than designing an architecture native to a given process.

Why are we repeating the same old discussions? 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. This isn't exactly someting that hasn't occured before - PowerVR designed the "IP" for PowerVRDC and NEC were the ones that did the back end for SEGA, similar thing with the GC.

So, will they go to NEC for the back-end ? I know you were just making an example, but it would be funny as Nintendo is doing basically the same thing ( Nintendo has close ties with NEC, it seems )

 

[jvd]

So, are you inplying that an entity such as Microsoft will take ATI's "IP" - which I think they said will be R500 based for XBox2 (correct if wrong) - and basically do the entire back-end of the design themselves? Which, to increase concurrency of design to absorb the influx in possible gates will require, at the least, some physical resuse scheme. Not to mention this is less effecient than designing an architecture native to a given process.

Why are we repeating the same old discussions? 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. This isn't exactly someting that hasn't occured before - PowerVR designed the "IP" for PowerVRDC and NEC were the ones that did the back end for SEGA, similar thing with the GC.

So, will they go to NEC for the back-end ? I know you were just making an example, but it would be funny as Nintendo is doing basically the same thing ( Nintendo has close ties with NEC, it seems )

Or ms can pay ati to do the back end. Or another company. Its up in the air and i'm very interested in how that pans out.

 

[cthellis42]

I believe the r400 was pushed off or cancelled and the r420 is its replacement with most of the r400 tech finding its way into the r500. So it is possible that ati is already working on the r600.

I thought R400 was basically delayed and renamed R500, with a new chip arch to cover the gap which needed to be in the 400's somewhere so as to not confuse everyone utterly. Hence R420.

I'm sure they ARE working on R600 already. (In fact I certainly hope they are.) Point is I wasn't thinking a "new architecture," as R420 is to R3xx now, would have as short of a run as it's shaping up to be. Makes me wonder on the whole just what it's going to amount to. Especially since it's going to start at .13 and likely move into .11, as nVidia seems to be doing with the NV4x line. Not as much room to stretch and shakes things out before R500 shows.

What is 420 going to be to 3xx? What is 500 going to be to 420? Questions, questions...

 

[Josiah]

I believe the r400 was pushed off or cancelled and the r420 is its replacement with most of the r400 tech finding its way into the r500. So it is possible that ati is already working on the r600.

I thought R400 was basically delayed and renamed R500, with a new chip arch to cover the gap which needed to be in the 400's somewhere so as to not confuse everyone utterly. Hence R420.

I'm sure they ARE working on R600 already. (In fact I certainly hope they are.) Point is I wasn't thinking a "new architecture," as R420 is to R3xx now, would have as short of a run as it's shaping up to be. Makes me wonder on the whole just what it's going to amount to. Especially since it's going to start at .13 and likely move into .11, as nVidia seems to be doing with the NV4x line. Not as much room to stretch and shakes things out before R500 shows.

What is 420 going to be to 3xx? What is 500 going to be to 420? Questions, questions...

AFAIK R400 did not become R500, R500 is its own beast. R400 is gone forever, it was not the right solution for its planned timeframe and neither will it be the right solution for R500 timeframe. R600 is at best only in the paper planning/R&D phase. R420 is based heavily off of R3xx tech, with some parts duplicated to increase performance, by now the design must be wrapping up. R400 was originally planned to be here by now, R420 is a stop-gap measure to fill in the void until R500 arrives.

 

[jvd]

I believe the r400 was pushed off or cancelled and the r420 is its replacement with most of the r400 tech finding its way into the r500. So it is possible that ati is already working on the r600.

I thought R400 was basically delayed and renamed R500, with a new chip arch to cover the gap which needed to be in the 400's somewhere so as to not confuse everyone utterly. Hence R420.

I'm sure they ARE working on R600 already. (In fact I certainly hope they are.) Point is I wasn't thinking a "new architecture," as R420 is to R3xx now, would have as short of a run as it's shaping up to be. Makes me wonder on the whole just what it's going to amount to. Especially since it's going to start at .13 and likely move into .11, as nVidia seems to be doing with the NV4x line. Not as much room to stretch and shakes things out before R500 shows.

What is 420 going to be to 3xx? What is 500 going to be to 420? Questions, questions...

AFAIK R400 did not become R500, R500 is its own beast. R400 is gone forever, it was not the right solution for its planned timeframe and neither will it be the right solution for R500 timeframe. R600 is at best only in the paper planning/R&D phase. R420 is based heavily off of R3xx tech, with some parts duplicated to increase performance, by now the design must be wrapping up. R400 was originally planned to be here by now, R420 is a stop-gap measure to fill in the void until R500 arrives.

with the r420 all but done (it will be done in a few short months). I would be surprised if they didn't move the r600 from paper to the more advanced design phase. 1 1/2 years to 2 years is about right for a complete design of a gpu is it not ?

 

[cthellis42]

I've seen a lot of stuff like this, though.

I wonder if Uttar will stop by and clarify anything further, as they pulled the information from NFI:

Just a small reminder because some people seem not to know:
The R500 "is" the R400, which was cancelled early this year. There are many reasons to this, among which the fact that ATI didn't feel they could deliver the R400 in the required timeframe, which is Q1 2004 ( it seems the August 2003 timeframe was either BS, or that it had already been delayed a bit before getting cancelled. )

The R500, thus made by the same team which worked on the R400, might of course have more features than the original R400 design had, because, well, it'll only be launched in Q4 2004, best case scenario.

The R420, which is based on the R3xx design with, of course, many tweaks, is to be launched in H1 2004, probably Q1 in fact. The estimated tape-out date and the transistor count are unknown.

That's probably, in fact, where I remember getting the information from--checking it out when it was first put up. Of course if I'd remembered correctly, R500 in Q4 2004 was mentioned here too, so I shouldn't have been surprised after all.

Be interesting to see what products R420 delivers, if it's having a run notably under a year.

 

[jvd]

I've seen a lot of stuff like this, though.

I wonder if Uttar will stop by and clarify anything further, as they pulled the information from NFI:

Just a small reminder because some people seem not to know:
The R500 "is" the R400, which was cancelled early this year. There are many reasons to this, among which the fact that ATI didn't feel they could deliver the R400 in the required timeframe, which is Q1 2004 ( it seems the August 2003 timeframe was either BS, or that it had already been delayed a bit before getting cancelled. )

The R500, thus made by the same team which worked on the R400, might of course have more features than the original R400 design had, because, well, it'll only be launched in Q4 2004, best case scenario.

The R420, which is based on the R3xx design with, of course, many tweaks, is to be launched in H1 2004, probably Q1 in fact. The estimated tape-out date and the transistor count are unknown.

That's probably, in fact, where I remember getting the information from--checking it out when it was first put up. Of course if I'd remembered correctly, R500 in Q4 2004 was mentioned here too, so I shouldn't have been surprised after all.

Be interesting to see what products R420 delivers, if it's having a run notably under a year.

I dunno it make sense that many of the things being done with the r400 would be moved into the r500. It seems like a waste of money and tech not to mention time just scraping the whole r400.

As for the 420. I believe it will push the dx 9 performance foward . Mabye 50-60 % faster than the r350s with ps2.0 and vs 2.0 . With perhaps pixel shader 3.0 and vertex shader 3.0 also included in the chip. With legacy apps I don't expect a huge performance in crease. I can already play every dx 7 and under game with 4-6fsaa and 16tap aniso. Don't really need more than that.

 

[cthellis42]

I think the thing was that the work they were doing with R400 wasn't going to be fully realized and beat out software by too much, so they'd refine it more and time it later, and enhance it to better scale. It's just if they still called it the R400 people would scratch their heads.

 

[Megadrive1988]

Josiah wrote:

AFAIK R400 did not become R500, R500 is its own beast. R400 is gone forever, it was not the right solution for its planned timeframe and neither will it be the right solution for R500 timeframe. R600 is at best only in the paper planning/R&D phase. R420 is based heavily off of R3xx tech, with some parts duplicated to increase performance, by now the design must be wrapping up. R400 was originally planned to be here by now, R420 is a stop-gap measure to fill in the void until R500 arrives.

Of everything I have heard over the last few months, this makes the most sense to me.

For months, people on various boards including Beyond3D have been saying that the original R400 was moved up in timeframe and re-named R500. Then the R420 being based on heavily beefed up R3xx tech. But that left a horrible question. The question of what would be happening with the *original* R500 design. I simply speculated that the original R500 would now be the R600. and not be out until sometime in 2006. That idea was disappointing to me, because I was guessing it ment that the XBox2 would *not* get to use the original R500 tech, and instead be based on the original R400 which had been moved up and re-named R500. Hopefully that is not the case at all.

So I am hoping that the R500 is in fact its own beast. That R400 is gone forever and has not been re-badged "R500". That the R500 *is* R500. And the R3xx-based R420 is the stop gap until R500 arrives. The way I see it now, thanks in part to your post Josiah, is the R420 design is done or close to done and waiting to go into production. The R500 is somewhat deep into development, but has a long ways to go. The XBox2 is being based on beefed up R5xx tech, with some custom stuff that MS wants (to help combat PS3 no doubt) And the R600 is a paper spec at the moment.

 

[David_South#1]

STI really is "driving the market"s focus on semiconductor design.
The fact that this thread has spent more time with erroneos STI stuff instead ATI is the proof of that.

Anyhoo, ATI R500 roadmap is in parrell design. Dual GPU.
XGI is doing this right now. And ATI stated they were going to follow this thinking also.

Deferred rendering also looks interesting. Kind of surprising that an idea that old hasn't caught on.
(PowerVR designs are the only examples I've seen.)

I'm not a computer major or have a degree. But I agree parellel is better.
The same way a digital movie is rendered in layers. Some tasks should be pushed over for seperate processing.

BTW-Looks like I'm going to win my bet with http://www.theinquirer.net ATI reporter.
I disagreed and said to him the R420 would be 130nm & GDDR2. He was claiming 150nm & DDR2.

 

[Megadrive1988]

I am thinking once again that if process technology for ATI-MS ever got to be a big problem, with Sony-IBM-Toshiba having a huge advantage, MS-ATI could simply counter with a duel-GPU configuration with 2 smaller GPUs on 90 nm - that would cut down on cooling issues would it not, even though duel GPUs brings up the cost, cost can be saved by having smaller less complex GPUs. It's always a trade-off anyway with these things. I know there are many that would poo-poo the more-than-one GPU approach. At least it's a possibility and I don't think anyone in their right mind could say that there is no possibility of it.

 

[David_South#1]

Yes, it would lower heat and energy numbers.
I'm pretty sure already that Xbox2 will be dual GPU.
I've also had the hunch for about a year now that a 64bit AMD processor would be the CPU.
(I know people don't see this since ATI-Intel seem to be a matched set. But with reason AMD just makes more sense. Still MS doesn't have to choose a processor until fall next year. Intel is releasing 64bit single core Xeons and other things in the mean time.)

I honestly don't think people are being conservative enough about STI and Consoles.
We should really view "Cell" discussions and PlayStation as two different things.
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.

 

[London Geezer]

maybe we should all stop and think an ATI card would cost MORE than the ENTIRE PS3. at least, at launch.

just a reality check for everyone...


EDIT: ooops sorry guy i only read the first page and posted... never saw the other 2 pages....

*runs away before being shot*