The LAST R600 Rumours & Speculation Thread

[R300King!]

And your point is nonsensical:

If they thought this way, R520 would have been delayed more (as I remember some pretty significant performance increases from drivers there). And that was released in a similar situation (G71 on the horizon).

No. ATI didn't discover any of those driver improvents until later. I also said what if AMD found out about these driver imprvements NOW but needed just a bit more time to make them work across all applications and games? So delaying this launch does makes some sense when you think about it that way.

Also, back in the R520 days it was ATI, this is AMD's decision now..two different companies with two different strategies.

I also never said this IS what they're are absolutely doing I said it could be a possible reason.

 

[Tim Murray]

No. ATI didn't discover any of those driver improvents until later. I also said what if AMD found out about these driver imprvements NOW but needed just a bit more time to make them work across all applications and games? So delaying this launch does makes some sense when you think about it that way.

Also, back in the R520 days it was ATI, this is AMD's decision now..two different companies with two different strategies.

I also never said this IS what they're are absolutely doing I said it could be a possible reason.

Nope. They've had the hardware (at least, early spins of it) for so long that they haven't somehow missed a 20% performance boost across the board until three weeks before the already-scheduled press day. Even if they did, they wouldn't cancel the press day for that--they'd just make a big hubbub about it so reviewers would tell their readers. "Hey guys, this is already 20% faster than G80, but they say it's going to be EVEN FASTER soon. Crazy, eh?"

Nothing good for R600 will come from this delay.

 

[R300King!]

Doesn't work, because NV can still delay the G80 refresh slightly to increase clocks with a larger cooler or something.

The only way that NV is screwed that I can think of is if R600 wins because of bandwidth alone. But, honestly, I don't see that happening except at high-res/high MSAA.

What do you think NV could possibly increase over their 8800GTX through their refresh, 25%?, 35%? That's a lot. So even if NV did this increase and AMD got their 30-40% increase over the 8800GTX they both would probably be about even in the end. I never said anything about NV being screwed. I think if AMD can achieve this gain then they are probably safe for the 8900GTX not beating them. Both being about equal in then end actually.

 

[Tim Murray]

What do you think NV could possibly increase over their 8800GTX through their refresh, 25%?, 35%? That's a lot. So even if NV did this increase and AMD got their 30-40% increase over the 8800GTX they both would probably be about even in the end. I never said anything about NV being screwed. I think if AMD can achieve this gain then they are probably safe for the 8900GTX not beating them. Both being about equal in then end actually.

NV increased core clocks by over 50% when going from G70 (well, a stock 256MB 7800 GTX) to G71. I've heard that it's extremely difficult to get a G80 over 630Mhz (a GTX is 575 stock). With a process shrink, is it really that unreasonable to think that they'd have a shot of hitting 725 or more? Then add in The Mystery of the Missing MUL, and you've got your R600-killer right there (unless it hits a bandwidth limitation in most apps).

But that makes some assumptions about R600 that are probably completely wrong to begin with, so whatever.

 

[SugarCoat]

NV increased core clocks by over 50% when going from G70 (well, a stock 256MB 7800 GTX) to G71. I've heard that it's extremely difficult to get a G80 over 630Mhz (a GTX is 575 stock). With a process shrink, is it really that unreasonable to think that they'd have a shot of hitting 725 or more?

well your comparison was with a full process shrink. I have to say i dont expect anything high end in 65nm until the end of 2007 is upon us and any earlier would be quite a surprise so i think realistically you'd actually be asking how much more can nVidia get with GDDR4, more voltage and 80nm and perhaps a core revision. At least thats me, but i just dont see these extremely complex behemoths coming out on 65nm before or at the beginning of July/August.

 

[Tim Murray]

well your comparison was with a full process shrink. I have to say i dont expect anything high end in 65nm until the end of 2007 is upon us and any earlier would be quite a surprise so i think realistically you'd actually be asking how much more can nVidia get with GDDR4, more voltage and 80nm and perhaps a core revision. At least thats me, but i just dont see these extremely complex behemoths coming out on 65nm before or at the beginning of summer.

I really don't either. Then again, NV did remarkably well with G80 on 90nm compared to G71 on 90nm. So, who knows what they'd get with 80nm.

I don't see anything beyond a pretty decent core clock increase and GDDR4 for G81. Then again, I can imagine an "8800 GTX Extreme" on G80 with GDDR4 as soon as R600 comes out (think 7800 GTX 512) with the real G81 to come later. This wouldn't surprise me, actually; the lack of any news whatsoever about G81 suggests it's a June-July part at the earliest.

 

[R300King!]

NV increased core clocks by over 50% when going from G70 (well, a stock 256MB 7800 GTX) to G71. I've heard that it's extremely difficult to get a G80 over 630Mhz (a GTX is 575 stock). With a process shrink, is it really that unreasonable to think that they'd have a shot of hitting 725 or more? Then add in The Mystery of the Missing MUL, and you've got your R600-killer right there (unless it hits a bandwidth limitation in most apps).

But that makes some assumptions about R600 that are probably completely wrong to begin with, so whatever.

LOL Now you're stretching it bigtime. And then AMD delivers their refresh and kills NV with their 512bit bus and mego bandwidth. Knocking NV in the dirt, so there!

I think after all is said and done they both will most likely be even with the slight edge going to AMD at higher resolutions.

 

[nonamer]

Maybe not. Remember that wonderful little tidbit a while back about the R600 being faster than everyone even thought it would be using just a software tweak?

What if? ....just what if AMD was all ready to show off the R600 at that showing (the one they now cancelled) and it was going to be about 10%~15% faster than the 8800GTX. AMD obviously knowing that the 8900GTX refresh would MAYBE catch it and/or surpass it but decided to go ahead anyway with the launch because they know they needed to answer with something and had delayed so long already.

But what if that "software tweak" turned out to be good rather than bad? What if they said, "Hey we can increase the R600 by 30%~40% over the 8800GTX if we just adjusted or tweaked in this area a bit more? Because if AMD had come out with the card at 10-15% improvement it would be very hard to get any secondary news out later that it got bigger 30-40% improvements with the drivers. This would cover any 8900GTX NV would launch.

People tend to look at the initial benchmarks and then make their decision. So AMD didn't want anyone to make their decision until they could take full advantage and get the absolute maximum performance out of their card as they knew they could.

Just a possible theory. ;-)

This is nearly as bad as WaltC's theory. If it's <10% then it's a tie for most people, and the only thing that matters is getting the product out. Also, if this software tweak can really improve performance by that much, surely they can come up with some beta drivers that unleashes it in some games at launch.

So try not to delude yourself. The delay of the R600 can only be due to negative reasons.

 

[Anarchist4000]

Drivers take what, about 1 month to get WHQL'd? If I remember back when ATI first started playing with their memory controllers they were getting some crazy performance increases just by tweaking the controller. If they're planning on using all that bandwidth for AA/AF(multichip bridges are the only other thing I can think of) and they were getting boosts of 20-40% back then with just 4x and 6x AA in various games. The problem with that now is they've got a rather insane amount of bandwidth and those old controller tweaks made things more efficient with higher levels of AA/AF. If R600 was a comparable architecture to R520/580 they'd need a rather insane amount of processing power to suck up all that bandwidth.

I'm not sure the delay would be a driver related issue unless they're having serious problems which haven't really been hinted at so far. If the scheduler was somehow programmable maybe, but I don't see any massive performance increases to be had outside of AA/AF, where they would likely be crushing G80 because of all that bandwidth anyways, so why delay?

The problem is more likely something that could lead to a recall to begin with. So either something didn't get certified or some form of manufacturing defect. I can't see needing a respin because they should have seen that coming a while back and wouldn't have scheduled editor's day. If anything they'd demo some engineering samples and just push the launch date back.

 

[Razor1]

Maybe not. Remember that wonderful little tidbit a while back about the R600 being faster than everyone even thought it would be using just a software tweak?

What if? ....just what if AMD was all ready to show off the R600 at that showing (the one they now cancelled) and it was going to be about 10%~15% faster than the 8800GTX. AMD obviously knowing that the 8900GTX refresh would MAYBE catch it and/or surpass it but decided to go ahead anyway with the launch because they know they needed to answer with something and had delayed so long already.

But what if that "software tweak" turned out to be good rather than bad? What if they said, "Hey we can increase the R600 by 30%~40% over the 8800GTX if we just adjusted or tweaked in this area a bit more? Because if AMD had come out with the card at 10-15% improvement it would be very hard to get any secondary news out later that it got bigger 30-40% improvements with the drivers. This would cover any 8900GTX NV would launch.

People tend to look at the initial benchmarks and then make their decision. So AMD didn't want anyone to make their decision until they could take full advantage and get the absolute maximum performance out of their card as they knew they could.

Just a possible theory. ;-)

Maybe the INQ misunderstood what was going on and that "software tweak" was actually what caused the delay, maybe it wasn't even a software tweak at all?

See basing theories on unsubstantiated rumors is a bad thing

The problem either has to do with performance or some fault in the card itself (and a fault that would have rendered the card practically useless in any paticular case would do), if the performance was faster then the g80, AMD would have launched, unless it wasn't cost productive (10-15% faster and faster later because of drivers tweaks IMHO would have been fast enough to launch). Last thing I can think of is AMD had been expecting something and it didn't happen.

Edit cost productive, AMD would have know the cost well before they set up Editor's day, unless they were confident about yeilds and yeilds turned out to be poor or less then thier target.

 

[Jawed]

Problem, what problem?

Jawed

 

[trinibwoy]

Did you not read everything I wrote?

I said AMD want the 30-40% improvement to beat just that, the 8900GTX.

Uh yeah, and exactly how does releasing R600 now avoid that situation? They can launch it now , beat up on G80 and still get the 30-40% "magic driver" out before Nvidia launches 8900. I don't mean to draw a parallel between the products themselves but I'm assuming there was a similar amount of sugar-coating and denial when the "monster" NV30 was late?

 

[nicolasb]

Problem, what problem?

Jawed

What is that supposed to mean? You think there really isn't a problem? Or you're satirising AMD's transparent attempt to cover up a very large problem?

 

[trinibwoy]

Edit cost productive, AMD would have know the cost well before they set up Editor's day, unless they were confident about yeilds and yeilds turned out to be poor or less then thier target.

You know that's an interesting question to ask. Does anyone know at which point in the process companies feel confident enough to schedule media briefings? I mean, it's not like they scheduled Editor's day and then hoped R600 was fast enough with decent yields. So how is it possible that they turned the ship around based on these factors - they would have a good idea of where they stood for a while now right?

 

[trinibwoy]

You think there really isn't a problem?

I agree with Jawed, there is no problem. You see, letting Nvidia ride high for 6 months is all part of the master plan. Cause Nvidia will get complacent, release a slightly souped up G80 and then ATi will unleash R600 on the world totally destroying anything Nvidia can produce and recoup their lost sales in record time! Brilliant strategy IMO.

 

[Geo]

You know that's an interesting question to ask. Does anyone know at which point in the process companies feel confident enough to schedule media briefings? I mean, it's not like they scheduled Editor's day and then hoped R600 was fast enough with decent yields. So how is it possible that they turned the ship around based on these factors - they would have a good idea of where they stood for a while now right?

Not necesarily. It can be a complicated problem. Driver status gets in there too. Why did they do the last spin can impact at what point they really have a good idea. Business intelligence (i.e. "we just found out dem other fellers are going to do x") etc. But, sure, cancelling an editors day is not a small matter; presumably there was a milestone reached to schedule it in the first place, and then "something happened".

 

[Bouncing Zabaglione Bros.]

I agree with Jawed, there is no problem. You see, letting Nvidia ride high for 6 months is all part of the master plan. Cause Nvidia will get complacent, release a slightly souped up G80 and then ATi will unleash R600 on the world totally destroying anything Nvidia can produce and recoup their lost sales in record time! Brilliant strategy IMO.

<slaps forehead> Yes of course! If WaltC was running AMD....

Minion: "Nvidia has been selling their next-gen DX10 cards for the last six months with no competition, they've made $50 million dollars, and are about to launch their refresh and midrange cards using the halo effect for their marketing..."

WaltC: "The fools! Now we have them just where we want them!"

 

[Geo]

You fool! You'll summon him! /runs away

 

[John Reynolds]

Reading the last few pages of this thread has been like a time warp back to 2002 and early 2003.

 

[nAo]

Reading the last few pages of this thread has been like a time warp back to 2002 and early 2003.

just different actors..