Was the 32 piped engineering samples the R580 all along

[imported_dwalton]

We have been hearing rumors of the 32 pipeline card from ATI for a long time. What we got was a 16 pipe R520.

Maybe this is the result of ATI being proactive with their refresh cards instead of reactive.

Correct me if I am wrong:

The R580 is to be the first admitted refresh part that adds pipelines to boost performance (I stepping out on the edge by saying this, but i've seen no speculation/rumors/information of the R580 being a 16 piped card). Some would name the 7800 a refresh card since its the NV47. However since nvidia change the codename and boardname, I don't think they think or want us to think of it as a refresh part.

Why design the 520 first and then the 580. Why not treat it as you would any other generation but using a twist of codenames and brandnames.

Designing the 580 first makes sense. You would run into 99% of the design issues from the start. Where designing a 16/24 pipeline card first might hide any issues related to its 32 pipeline brethen. I doubt you would see as many issues by designing a 32 pipe card first and cutting it down to 16 once the intial problems were corrected. Designing the R580 first would give you extra time to work out any issue related to taking such a big step in complexity and performance. However, this might shorten the time you have to work on the R520. You could remedy this by using a second team to come in after the design phase and tapeout to accelerate the time line of the R520. This would also explain all the talk about 32 piped engineering cards.

I think the reason we are seeing 16 pipeline R520 is because ATI overestimated their ability to build a 32 pipe card to compete with anything Nvidia had to offer before DX10.
I don't believe ATI planned a 16 pipeline card running at 600 Mhz. I speculate that ATI planned to build the 580 first as a 32 pipeline card, working out any kinks, then cutting it down to the 24 pipe R520 to compete with the 7800 GTX/GT. The 16 pipe cards was a result of yield/leakage issues related to the 32/24 pipe version.

I think that ATI speculated that anything that Nvidia produced after the 7800 but before its DX10 part would be the same arch as the 7800 just a bump in the pipes and speed while maybe being DX 9.0L compliant. I believe ATI felt that if they could bump up their efficiency as well pipeline and speed, nvidia would have a hard time reacting with simple pipe and speed bumps.

By bumping up the efficiency of the arch of the R5xx, ATI wouldn't be forced to run at higher speeds to keep it competitive with Nvidia. Looking at the fact that the 6800 Ultra has the same pixel/texture fillrates as a X800Xl and it becomes clear that the R420 arch was as not as efficient as the NV40. By improving the efficiency of the R5xx and maintaining the ability to run at higher speeds, ATI would of forced nvidia to up its speed on its cards. This would of probably affected yield as well as power consumption. By bumping up the efficiency ATI could simply keep pace with Nvidia until the R600, while reducing the amount of investment in the R580/520.

This is all speculation and wouldn't bet a dollar on it being valid. Just something I'm doing to fill the time until we see if the R520 is the glorious return of the 9700, a 5800 in deguise or something in between.

 

[roguerower]

It makes sense. Hell of a first post. Welcome to Anandtech

 

[Munky]

You know, I was thinking the same thing - if the r520 was originally designed as a 16 pipe card (as the rumors say) then what are those 32 pipe samples? I cant imagine how bad the heat and yields have to be that what was originally a 32 pipe card has to be cut in half to 16 pipes.

I dont think that the r580 was developed first, but perhaps the 2 teams were working on it simultaneously, and whatever problems the r520 team faced would be taken into account by the r580 team. IMO, I think the r520 would offer roughly equal performance to the 7800gtx, so it would be pretty even this gen. But if Ati releases a 32 pipe r580 in early 2006, then Nv would need at least a 32 pipe card clocked as high as the r580 to compete, and with the 110nm process it's not gonna happen. I think Ati took a bigger initial step in designing the new card, and that's why they were more likely to run into problems and delays. But now, Nv is gonna have to take the same big step up in designing the g70 refresh or else they will fall behind.

 

[johnnqq]

which generation is supposed to support dx10?

 

[Pabster]

Let's hope the 32-pipe samples are not R580 or we're going to wait EVEN LONGER for that elusive part.

 

[Soccerman06]

Originally posted by: dwalton

We have been hearing rumors of the 32 pipeline card from ATI for a long time. What we got was a 16 pipe R520.

Maybe this is the result of ATI being proactive with their refresh cards instead of reactive.

Correct me if I am wrong:

The R580 is to be the first admitted refresh part that adds pipelines to boost performance (I stepping out on the edge by saying this, but i've seen no speculation/rumors/information of the R580 being a 16 piped card). Some would name the 7800 a refresh card since its the NV47. However since nvidia change the codename and boardname, I don't think they think or want us to think of it as a refresh part.

Why design the 520 first and then the 580. Why not treat it as you would any other generation but using a twist of codenames and brandnames.

Designing the 580 first makes sense. You would run into 99% of the design issues from the start. Where designing a 16/24 pipeline card first might hide any issues related to its 32 pipeline brethen. I doubt you would see as many issues by designing a 32 pipe card first and cutting it down to 16 once the intial problems were corrected. Designing the R580 first would give you extra time to work out any issue related to taking such a big step in complexity and performance. However, this might shorten the time you have to work on the R520. You could remedy this by using a second team to come in after the design phase and tapeout to accelerate the time line of the R520. This would also explain all the talk about 32 piped engineering cards.

I think the reason we are seeing 16 pipeline R520 is because ATI overestimated their ability to build a 32 pipe card to compete with anything Nvidia had to offer before DX10.
I don't believe ATI planned a 16 pipeline card running at 600 Mhz. I speculate that ATI planned to build the 580 first as a 32 pipeline card, working out any kinks, then cutting it down to the 24 pipe R520 to compete with the 7800 GTX/GT. The 16 pipe cards was a result of yield/leakage issues related to the 32/24 pipe version.

I think that ATI speculated that anything that Nvidia produced after the 7800 but before its DX10 part would be the same arch as the 7800 just a bump in the pipes and speed while maybe being DX 9.0L compliant. I believe ATI felt that if they could bump up their efficiency as well pipeline and speed, nvidia would have a hard time reacting with simple pipe and speed bumps.

By bumping up the efficiency of the arch of the R5xx, ATI wouldn't be forced to run at higher speeds to keep it competitive with Nvidia. Looking at the fact that the 6800 Ultra has the same pixel/texture fillrates as a X800Xl and it becomes clear that the R420 arch was as not as efficient as the NV40. By improving the efficiency of the R5xx and maintaining the ability to run at higher speeds, ATI would of forced nvidia to up its speed on its cards. This would of probably affected yield as well as power consumption. By bumping up the efficiency ATI could simply keep pace with Nvidia until the R600, while reducing the amount of investment in the R580/520.

This is all speculation and wouldn't bet a dollar on it being valid. Just something I'm doing to fill the time until we see if the R520 is the glorious return of the 9700, a 5800 in deguise or something in between.

You better keep making all your posts with this quality or the citizens of ATOT will put your balls in a vice (or ban). This is what you get when you make a first post that good.

which generation is supposed to support dx10?

ATI has a simple naming system for this, 8x00 for DX 8.0, 9x00 for DX 9.0, When DX 10.0 comes out ATI will name their cards the 10x00. Nvidia's will be the 8x00 series I believe.

 

[exar333]

thats possible, but didnt just last week the "industry" was lauding the great fabrication and silicon yields of the R520? or was that just cover for this whole plan? i wonder...

to be honest, i just think ATI was caught off guard by Nvidia actually launching the G70 series in volume and not a paper launch...i think ATI got too used to "vaperware" launches that they were afraid to lose face by doing exactly that, and waited until they could fab more....

thats my opinion...i like both companies, dont get me wrong...but i think ATI is feeling the heat right now...I (and you all) are anxious to see how they can respond with R520/R580...

Edit: oh yeah...since this is your first post, i should really degenerate into a flame war...

Nvidia Pwns! ATI is the SuXorz!! *wink* j/k

 

[imported_dwalton]

Originally posted by: munky

I dont think that the r580 was developed first, but perhaps the 2 teams were working on it simultaneously, and whatever problems the r520 team faced would be taken into account by the r580 team. IMO, I think the r520 would offer roughly equal performance to the 7800gtx, so it would be pretty even this gen. But if Ati releases a 32 pipe r580 in early 2006, then Nv would need at least a 32 pipe card clocked as high as the r580 to compete, and with the 110nm process it's not gonna happen. I think Ati took a bigger initial step in designing the new card, and that's why they were more likely to run into problems and delays. But now, Nv is gonna have to take the same big step up in designing the g70 refresh or else they will fall behind.

While I might agree with you under normal circumstances ( normal circumstances being two group simultaneously working on two different GPU generation ), I find it more logical that one group or conglomerate of groups worked on the R580 during the design phase and the first tapeout and then splintered into seperate groups to work on the R520 and R580. Designing the R580 first gives you several advantages.

You have a longer time working on the R580 which will be the more complex card. Major issues in terms of yield/leakage that may be unique to the R580 will be seen earlier and allows for more time to compensate or fix. Designing the R520 first means additional design on the R580 which delays its tapeout.

The first tapeout will allow you to produce engineering samples of R520 and R580 at the same time. Simply disabling the the pipes and/or features on the R580 will give you the R520. Its easier and cheaper to disable pipes/features then it is to add as adding require more tapeouts. Producing the R520 first, forces the tapeout of two parts one R520 part and one R580. This increases cost, which would not be in the best interest of ATI

 

[Munky]

Well, I hope you theory is right, because I cant wait to get one of those 32-pipe moster cards.

 

[phr0m]

wow amazing first post and theory, but yeha i think your right about the r580 being made first

 

[Pete]

Interesting theory, dwalton (and welcome to AT ). I've seen some numbers posted at B3D for ATI's upcoming R5xx generation:

R520: 16-1-1-1
RV515: 4-1-1-1
R580: 16-1-3-1
RV530: 4-1-3-2

Given that we're going to see RV530--which is rumored to be closer to R580 than R520, and those mystery numbers seem to confirm that--at the same time as RV515 and slightly before R520, it's plausible that R580 was developed in tandem with R520 and cards are already in the wild. In reviewers' hands, though? Seems a little early for that, if ATI will push the R520 as a normal-life card, and not a FX 5800-like "mistake."

Now, the catch is, no one knows for sure what the numbers stand for, though they're apparently related to the pixel (not vertex) pipeline. We've already heard from AT that R520 is 16 pipes and RV530 is 12, so the first number may be ROPs. The rest aren't as easy to speculate about, though it does seem that R580 will be much better-equipped in some way than R520. That 3 multiple doesn't lend itself to 32 pipes (from R520's 16), though. Hopefully RV530 will give us a clue as to how R580 will improve on R520. Latest rumor is a 5 Oct. launch.

 

[ElFenix]

if the numbers posted earlier are any indicator, the x18 won't be a mistake on the level of the 5800U. the 5800U wasn't competitive in anything, sounded like a dustbuster, and was released more than 5 months later than the 9700 pro. and then, to add insult to injury, the radeon 9800 pro was released a month and a half later. the x18 seems like it will be competitive, released only 3 months later, and probably won't sound like a dustbuster, given ati's history with dual slot coolers.

 

[Topweasel]

Your right, those benches in the link you posted that had the 5800U winning were just miss marked.

But Again your right, Over all the 7800 only beats the X1800 in almost every benchmark but Farcry, Is a 1 Slot solution, and came out 4-5 months before the X1800 is projected to be released, and (Difference determined later) was available in mass (and I mean mass) quantities the day it was released.

 

[jasonja]

Originally posted by: Topweasel

But Again your right, Over all the 7800 only beats the X1800 in almost every benchmark but Farcry, ..

Which benchmarks?

 

[Pete]

Probably these unconfirmed benches.

 

[Topweasel]

Originally posted by: Pete
Probably these unconfirmed benches.

Yep its one of those grain of salt things, cept this is right inline with just about everything I heard. They might vary a little when the real ones come out, but anyone expecting Better then 7800GTX performance shouldn't hold their breath.

 

[Bartokomus]

Am i understanding the innunendo here, and R520 and R580 will have same quarter launches?

or am i misunderstanding?