Southern Islands ... why?

[dzoner]

It's about the money Lebowski.

AMD currently has a 40nm 5xxx cash cow on their hands at all market segments/price points. Economically, why change that up anytime soon except to ...

a) move to a smaller, more cost effective node.
b) respond to competitive pressures from Nvidia.

If S.I. is on the 40nm node, that would nullify a).

That leaves what Nvidia will be bringing to the table ... after all, if AMD had reason to believe Nvidia wasn't going to bring much heat for the rest of 2010, they would have little reason to do much at all ... maybe do a lower cost 'hybrid' 5790 part to replace the 5830 (or shove it up to a higher price point) and a 5890~5950(5790x2?) to hit the $500 price point and take back the single gpu crown.

It would be very time/resource economical to tape out only one 'hybrid' chip if that is all that was really needed. It would also give them working knowledge and experience of most of N.I.'s architecture at 40nm, paving the road and freeing up resources to concentrate on implementing the full N.I. on GF's 28nm.

This would also keep product line confusion to a minimum ... bringing out hybrid S.I. cards that compete with existing 5xxx cards ... why? what would they be called? Better to wait and roll out an entire 6xxx product line over a few months like they did with the 5xxx line.

At best, Nvidia is unlikely to BETTER AMD's cost/performance in the low end and mainstream graphics segments, and more likely will be worse. Timeline wise all I've seen so far for Nvidia's GF104/108 is 'june at the earliest' ... meaning what? ... sometime in august or september?

Only reason I can see for AMD to do a complete (6xxx) S.I. refresh at 40nm is if the progress of GF's 28nm bulk process is such that N.I. will be delayed into 2H of 2011.

 

[TemjinGold]

Probably the same reason why they rolled out a 4890 when they were in this very position last time.

 

[Daedalus685]

They can't delay a part for no good reason. They would have planned SI as soon as 32nm was canned. Likely they were counting on Fermi being released in bulk and having lots of competition. Just because that didn't happen is no reason to slide your feet. Should they instead throw all the money they spent on the R&D away instead of releasing what could pull them farther ahead and give valuable info that can be used on NI?

They work to a schedule, if the competition falls behind that is great.. but they don't fundamentally care, at least they shouldn't. All they can control is how awesome their own is, not how bad the other guy is.

 

[railven]

Why does the story of the tortoise and the hare come to mind.

Sure, you got a big lead, might as well sleep on it - what bad can happen right?

I already bought into the 5x series, I hope the 6x series isn't a total monster...for once I hope its just a step increase and not leap.

 

[Daedalus685]

Why does the story of the tortoise and the hare come to mind.

Sure, you got a big lead, might as well sleep on it - what bad can happen right?

I already bought into the 5x series, I hope the 6x series isn't a total monster...for once I hope its just a step increase and not leap.

Well for good or bad I can't imagine them "leaping" without a die shrink. I wouldn't hold my breath for anything better than 20% mm for mm. Not to say they won't get more than that is they make something huge.. but I doubt they'd drop their strategy that quickly... Expect new features, better tessellation, and a touch more performance with NI giving us a whole new everything.

Does anyone know of historical examples of in between architectures like this?

 

[Piotrsama]

SI is ATI's Plan B, and it originated because of TSMC's problems.
Plan A was to go straight to NI on 32nm (or 28).

 

[cbn]

If ATI is going to be using Globalfoundries for 28nm, how much would building a smaller version of NI on TSMC 40nm help them?

Doesn't keeping the Evergreen core allow them to save process technology costs?

 

[Daedalus685]

If ATI is going to be using Globalfoundries for 28nm, how much would building a smaller version of NI on TSMC 40nm help them?

Doesn't keeping the Evergreen core allow them to save process technology costs?

We don't know for sure what will actually be on GF 28nm.. it may yet still be only low end. I doubt they will not use TSMC at all.

Fundamentally they gain nothing manufacturally from SI at TSMC as it relates to GF... but it could tell then the direction they might want to go with certain things.

I really think we will see SI at 28nm BEFORE we see NI to work out the bugs in the tech.. but who knows.

 

[OCGuy]

Yes SI is a test run on the same node. Changing up Arch and moving to a smaller node simultaneously doesnt always work out.

 

[tviceman]

Yes SI is a test run on the same node. Changing up Arch and moving to a smaller node simultaneously doesnt always work out.

Changing the architecture on the current node will provide an interesting challenge for ATI. I think some fair assumptions can be made regarding their next gen architecture the new high end single GPU part will perform better than their current best single GPU, and the new architecture will be more complex than their current.

Regarding the second assumption I made, and this correlates directly with what many people feel Nvidia is currently battling, I wonder how ATI will manage to keep their dies "small" and TDP's under reasonable control given that they aren't moving to a smaller process.

It will be interesting to see if they make any "sacrifices" compared to certain parts of their current architecture to maintain their small die strategy.

 

[cbn]

We don't know for sure what will actually be on GF 28nm.. it may yet still be only low end. I doubt they will not use TSMC at all.

I really think we will see SI at 28nm BEFORE we see NI to work out the bugs in the tech.. but who knows.

Yep, I hope ATI pulls another "HD4770 style project" on Global foundries 28nm tech.

 

[dzoner]

They can't delay a part for no good reason. They would have planned SI as soon as 32nm was canned. Likely they were counting on Fermi being released in bulk and having lots of competition. Just because that didn't happen is no reason to slide your feet. Should they instead throw all the money they spent on the R&D away instead of releasing what could pull them farther ahead and give valuable info that can be used on NI?

They work to a schedule, if the competition falls behind that is great.. but they don't fundamentally care, at least they shouldn't. All they can control is how awesome their own is, not how bad the other guy is.

They work to a schedule, but that schedule was severly disrupted AND Nvidia faceplanted with Fermi. S.I. is a plan 'b' (or 'c') interim product, but by the time they finished up the design phase and were ready to do S.I. tapeouts, they surely knew the scoop on Fermi and that it posed little threat for the rest of the year.


N.I. at 28nm is where AMD really has a chance to really clamp down on Nvidia's GPU throat. Getting a six months (or more) jump on Nvidia at 28nm using GF to Nvidia's TSMC would put a world of hurt on Nvidia.

T

 

[digitaldurandal]

Probably the same reason why they rolled out a 4890 when they were in this very position last time.

They were and not in this position last time.

Nvidia 285 > 4870;

Hell

275 > 4870

They were in this position when they released the 9700 but Nvidia struck back with the 6800 series. Currently the 58xx > 275 and 285 and Nvidia's release was not very phasing at all. It barely bests the 5xxx series in most titles, has serious disadvantages and pricing issues.

They have really NEVER been in this comfortable of a position.

I believe the reason to release the refresh is to stay in this position. The more they beat Nvidia pricing and performance the worse off Nvidia's position will become, if Nvidia loses its giant fund advantage over AMD (has it already?) then AMD will likely stay on top for good.

 

[Daedalus685]

They work to a schedule, but that schedule was severly disrupted AND Nvidia faceplanted with Fermi. S.I. is a plan 'b' (or 'c') interim product, but by the time they finished up the design phase and were ready to do S.I. tapeouts, they surely knew the scoop on Fermi and that it posed little threat for the rest of the year.


N.I. at 28nm is where AMD really has a chance to really clamp down on Nvidia's GPU throat. Getting a six months (or more) jump on Nvidia at 28nm using GF to Nvidia's TSMC would put a world of hurt on Nvidia.

T

Si would have been planned before was 32nm was canned. This was well before fermi was in the works... Also, it is quite possible the refresh of the 5000 was always planned to be a hybrid of NI.

Si should not hurt Ni at all, and there is no reason to not keep the gas down. Perhaps they figure it will give them a leg up on the transition to 28nm as well.

They made the call a while ago, they won't change there minds because of a stumble. I think 28nm is likely a touch later than many thought as well. But SI was not something they threw together last month... It would have been started last summer at the latest to tape out now.

 

[SHAQ]

So is it NI that will coincide with TSMC's launch of high performance 28nm at the end of September? Or will they go with GF? And will Nvidia be the one with TSMC at the end of September and not ATI since they are snubbing TSMC by going with GF? I don't suppose any of this is known yet. ATI really doesn't have much of a lead since Fermi only needs to be tweaked to come out at 28nm.

 

[Schmide]

ATI really doesn't have much of a lead since Fermi only needs to be tweaked to come out at 28nm.

Dude Fermi still needs tweaking on 40nm. Forget the onlys.

 

[OCGuy]

They were and not in this position last time.

Nvidia 285 > 4870;

Hell

275 > 4870

They were in this position when they released the 9700 but Nvidia struck back with the 6800 series. Currently the 58xx > 275 and 285 and Nvidia's release was not very phasing at all. It barely bests the 5xxx series in most titles, has serious disadvantages and pricing issues.

They have really NEVER been in this comfortable of a position.

I believe the reason to release the refresh is to stay in this position. The more they beat Nvidia pricing and performance the worse off Nvidia's position will become, if Nvidia loses its giant fund advantage over AMD (has it already?) then AMD will likely stay on top for good.

Huh?

480 > 5870..so your scenario matches.

 

[edplayer]

AMD currently has a 40nm 5xxx cash cow on their hands at all market segments/price points. Economically, why change that up anytime soon except to ...

a) move to a smaller, more cost effective node.
b) respond to competitive pressures from Nvidia.

You could only come up with two reasons to change anything soon?


its a good thing that you don't work for ATi

 

[Patrick Wolf]

It's more of question of; Why not release new product that's better than current product?

Which makes this thread moot, since they will be releasing new product. Reguardless if it's "new" or a "refresh", it's still better and that's why they're doing it.

Unless they just want make beta testers out of consumers in prep for N.I.

 

[SlowSpyder]

I like to see AMD with their foot on the gas. It's much better for us enthusiasts than say, oh I don't know, something like changing your 8 to a 9 on the sticker when you have the technology lead.

SI will probably be a feeler part like the 4770 was. They'll figure out what they need to do to make NI a success based on their work with SI. Just my guess.

 

[ronnn]

They need something to compete with fermi when it gets out the door.

 

[railven]

Well for good or bad I can't imagine them "leaping" without a die shrink. I wouldn't hold my breath for anything better than 20% mm for mm. Not to say they won't get more than that is they make something huge.. but I doubt they'd drop their strategy that quickly... Expect new features, better tessellation, and a touch more performance with NI giving us a whole new everything.

Does anyone know of historical examples of in between architectures like this?

They were and not in this position last time.

Nvidia 285 > 4870;

Hell

275 > 4870

They were in this position when they released the 9700 but Nvidia struck back with the 6800 series. Currently the 58xx > 275 and 285 and Nvidia's release was not very phasing at all. It barely bests the 5xxx series in most titles, has serious disadvantages and pricing issues.

They have really NEVER been in this comfortable of a position.

I believe the reason to release the refresh is to stay in this position. The more they beat Nvidia pricing and performance the worse off Nvidia's position will become, if Nvidia loses its giant fund advantage over AMD (has it already?) then AMD will likely stay on top for good.

Since they sort of tie together, the only other "in-between" design I can really think of was the 9800 series.

It was essentially the 9700 (RV300) with some tweaks and I believe a die shrink, I'd have to double check, that ATI released to fight off the oncoming 5800 launch.

It is a very similar situation to now:
ATI 9700 turning heads
Nvidia 5800 is launched late, hot, loud, and power hungry but beats the 9700 in most bench
ATI drops the 9800 continuing its victory
Nvidia drops the 6800 and continues to reign supreme until the 8800.

Funny how history repeats itself. I guess the GF200 (Fermi II) will be a monster!!!

 

[swilli89]

Since they sort of tie together, the only other "in-between" design I can really think of was the 9800 series.

It was essentially the 9700 (RV300) with some tweaks and I believe a die shrink, I'd have to double check, that ATI released to fight off the oncoming 5800 launch.

It is a very similar situation to now:
ATI 9700 turning heads
Nvidia 5800 is launched late, hot, loud, and power hungry but beats the 9700 in most bench
ATI drops the 9800 continuing its victory
Nvidia drops the 6800 and continues to reign supreme until the 8800.

Funny how history repeats itself. I guess the GF200 (Fermi II) will be a monster!!!

Haha WHAT? Almost as bad as some germans claiming the holocaust didn't happen. Sigh.. Alright. The 5800 did NOT beat anything in most benches. Maybe the 9500 pro? I bet that was faster as well though.

http://www.anandtech.com/show/1314/11

The 9700 Pro beats the 5950 Ultra here. A faster card than the 5900 and a MUCH faster card than the 5800. The 5800 launched and was an immediate failure. There was no benefit to owning them. Louder, hotter, and slower.

The 5800 was barely faster than a Ti4600.

Also note there, the X850 series were generally faster than the 6800 series. Although rare, the X800XT PE stomped all of the 6800's.

nVidia did not in fact, lead from the 6800 series until the G80. Although ATI stumbled with the x1800 and x2900, and NV was on top with the 7800 GTX and 8800 GTX respectively, they had a win with the X1900 series which traded blows with their competition.

I think the 9800 was a simple shrink to .11 micron with higher clocks.

 

[Kuzi]

Only reason I can see for AMD to do a complete (6xxx) S.I. refresh at 40nm is if the progress of GF's 28nm bulk process is such that N.I. will be delayed into 2H of 2011.

Yes, neither GF nor TSMC can to have 28nm process technology volume production-ready for this year, so ATI going with the same mature 40nm process can be a good move, as long as SI's die size doesn't end up being too large. It's likely SI will be a tweaked Cypress, with performance slightly faster than the GTX480 and a die size of about ~410mm^2.

Well for good or bad I can't imagine them "leaping" without a die shrink. I wouldn't hold my breath for anything better than 20% mm for mm. Not to say they won't get more than that is they make something huge.. but I doubt they'd drop their strategy that quickly... Expect new features, better tessellation, and a touch more performance with NI giving us a whole new everything.

I agree with you, we can't expect a large leap unless it's on a smaller process. And a 20-25% improvement might be all SI would provide, at least that is enough to be comfortably ahead of the GTX480, although by that time Nvidia might have a faster (hotter) card (GTX490?).

I think for the 28nm process, ATI will release a small test card based on the 5xxx series or SI, at the end of this year or early next year. This will allow them to adapt to GF's process, and produce high end NI cards in Q3/Q4 of next year.

 

[at80eighty]

Wonder if they'll make an SI to compete with the 5970. i'd get one then - atleast by the holidays if that's when they release it