GeForce GTX 480 and GeForce GTX 470

[realitycheck]

Yes, but PS3 supporters love to just point at the FLOPS score of the system to claim its faster. The Cell's vector processors are SPEs, by the way



More or less. The Xenos also has unified shaders whereas the PS3 basically has a 7900GT with half the memory bus.



I haven't played Uncharted 2, but I have played KZ2 - it doesn't make extensive use of Cell's SPEs. Cell's SPEs generally don't do rendering, so a pretty game does not mean good use of the SPEs, which was my point. KZ2's CPU usage tops out at 60% under the heaviest loads: http://www.qj.net/qjnet/playstation-3/haynes-killzone-2-not-close-to-maxing-out-ps3.html

Which of course shouldn't be a surprise to anyone here - CPUs don't make pretty pictures, GPUs do.

That said, KZ2 really isn't anything graphically special. The particle effects are good, but the world is very straight edge and simplistic - which is graphically simple and easy to render.

I've played KZ2 and GOW2 back to back, both were equally good for consoles. GOW2 was better in some areas, and in others KZ2 was better. Both were played on a 60in plasma.

One of these days I might try Uncharted 2, but I kind of doubt it.

I dont think KZ2 is off the wall amazing in either game play or graphics but if you don't play uncharted 2, and you have a PS3, you are doing yourself the biggest injustice in the world. GOTY. In fact, I wouldn't be surprised one bit if it was Sony's new number one system seller. The Uncharted series should be.

 

[realitycheck]

Anyway, to get this thread back on track......I sure do hope we get a launch date and it's soon. My refund will be here in 14 days, and if I like what I see from team green, I'll buy and have it overnighted.

 

[Kowan]

I'm trying to finish Uncharted before starting Uncharted 2. Bought and still in the box.

 

[Jhalf]

Anyway, to get this thread back on track......I sure do hope we get a launch date and it's soon. My refund will be here in 14 days, and if I like what I see from team green, I'll buy and have it overnighted.

I'm in the exact same boat... my refund acts as my "tech savings account" every year!

 

[Cool Vibrations]

There are other options, you know...

Nvidia is allowing these new SLI modes on existing cards, so you could get two MUCH CHEAPER Nvidia cards and run Nvidia's triple-monitor solution that way. Getting two cheaper Fermi-based cards will be a possibility at some point in the future as well.

There's also the Matrox TripleHead2Go (and the recently released Mview), which split one DVI or DisplayPort connection into 3. This would allow you to run three monitors with a single GTX470 or 480. This also works on existing Nvidia hardware.

There's also SoftTH, which would let you use a single GTX470 or 480, with a secondary card like an 8400GS, to drive triple-monitor. This also works on existing Nvidia hardware.

Current Nvidia drivers can't do 5040x1050 on some DX10 games. For BF:BC2, I have to force the game to run at dx9.

 

[dannyai]

I cant believe the amount of people who still think , nvidia's fermi will beat ATIs hardware.

3 months before release, no cards seen. no benchmarks seen. no one even has one. no leaks.

this proves 1 thing. the card is either not ready, or performs like CRAP.
and its the second one. the card performs like CRAP, and they cant clock it high enough to beat the ATI cards. so they are scratching their heads at nvidia, thinking how to sell this thing.

The other thing is, this card is going to pump alot of heat into everyones case. I have no idea why you all think nvidia will just bring some magic three months from now. this card will be the FX 5800. the NV30 that was a piece of crap.

for those that are waiting for DX11 from nvidia forget it.
ATI is your company. at least for another 2 years from today.

 

[Hornet]

I cant believe the amount of people who still think , nvidia's fermi will beat ATIs hardware.

3 months before release, no cards seen. no benchmarks seen. no one even has one. no leaks.

this proves 1 thing. the card is either not ready, or performs like CRAP.
and its the second one. the card performs like CRAP, and they cant clock it high enough to beat the ATI cards. so they are scratching their heads at nvidia, thinking how to sell this thing.

The other thing is, this card is going to pump alot of heat into everyones case. I have no idea why you all think nvidia will just bring some magic three months from now. this card will be the FX 5800. the NV30 that was a piece of crap.

for those that are waiting for DX11 from nvidia forget it.
ATI is your company. at least for another 2 years from today.

and somehow you happens to know so much about it

 

[crackbone]

I cant believe the amount of people who still think , nvidia's fermi will beat ATIs hardware.

3 months before release, no cards seen. no benchmarks seen. no one even has one. no leaks.

this proves 1 thing. the card is either not ready, or performs like CRAP.
and its the second one. the card performs like CRAP, and they cant clock it high enough to beat the ATI cards. so they are scratching their heads at nvidia, thinking how to sell this thing.

The other thing is, this card is going to pump alot of heat into everyones case. I have no idea why you all think nvidia will just bring some magic three months from now. this card will be the FX 5800. the NV30 that was a piece of crap.

for those that are waiting for DX11 from nvidia forget it.
ATI is your company. at least for another 2 years from today.

LOL.

Ok Troll. What you wrote above might be correct, however anyone in a situation that doesn't need a card right this moment, is doing themselves a complete injustice by not waiting for Fermi based cards to drop.

Scenario 1 : Fermi kicks the tail out of ATI 5800 series which forces ATI price drops. Either way, whatever you are going to buy, you win.

Scenario 2 : Fermi doesn't overly destroy ATI in performance numbers, but launches at a relatively decent price point.

Either way, you know what you are getting in to. You have all the information and then you can make an informed buying decision. I see no reason to have all the information.

Lastly, If you count NV out, then you are truly even more asinine than your post alludes to.

 

[fdiaz78]

I cant believe the amount of people who still think , nvidia's fermi will beat ATIs hardware.

3 months before release, no cards seen. no benchmarks seen. no one even has one. no leaks.

this proves 1 thing. the card is either not ready, or performs like CRAP.
and its the second one. the card performs like CRAP, and they cant clock it high enough to beat the ATI cards. so they are scratching their heads at nvidia, thinking how to sell this thing.

The other thing is, this card is going to pump alot of heat into everyones case. I have no idea why you all think nvidia will just bring some magic three months from now. this card will be the FX 5800. the NV30 that was a piece of crap.

for those that are waiting for DX11 from nvidia forget it.
ATI is your company. at least for another 2 years from today.

Lots of assumptions here buddy. I for one prefer Nvidia's hardware and drivers over ATI's problematic drivers. Things just work out of the box when I play.

Having said this you should root for both teams and stop being such a fanboy. More competion + price war = we win. Look a the big picture.

 

[Hamish Campbell]

The problem with waiting is that the 480 and 470 are probably a few months away from being in stores (online or otherwise). The time frame is seriously looking to be more like April or May of 2010 with nvidia paper launching in March.

If I were in the market for a video card today, buying a 5850 or 5870 is still easily the best option.

 

[Angus66]

I'm still in shock that there is still no hard data - pre-release evaluation cards, actual performance numbers from impartial review sites, etc - for the new Nvidia cards.

I'm not partial to either camp, but the silence from Nvidia speaks volumes IMHO.

 

[Hamish Campbell]

I'm still in shock that there is still no hard data - pre-release evaluation cards, actual performance numbers from impartial review sites, etc - for the new Nvidia cards.

I'm not partial to either camp, but the silence from Nvidia speaks volumes IMHO.

These are all signs of a much later release date that some people are willing to admit. We'll be lucky to see anything in march.

 

[Xzyrus]

http://www.fudzilla.com/content/view/17524/1/

Earliest performance info is CeBIT maybe? If AIBs are supposed to have cards to show off for that, there should be some performance info coming then if there aren't any leaks earlier. That may also indicate that we won't be seeing the cards until later in March or closer to April, but thats Fudzilla for you.

 

[cannondale06]

I cant believe the amount of people who still think , nvidia's fermi will beat ATIs hardware.

3 months before release, no cards seen. no benchmarks seen. no one even has one. no leaks.

this proves 1 thing. the card is either not ready, or performs like CRAP.
and its the second one. the card performs like CRAP, and they cant clock it high enough to beat the ATI cards. so they are scratching their heads at nvidia, thinking how to sell this thing.

The other thing is, this card is going to pump alot of heat into everyones case. I have no idea why you all think nvidia will just bring some magic three months from now. this card will be the FX 5800. the NV30 that was a piece of crap.

for those that are waiting for DX11 from nvidia forget it.
ATI is your company. at least for another 2 years from today.

 

[Rossi~]

I cant believe the amount of people who still think , nvidia's fermi will beat ATIs hardware.

3 months before release, no cards seen. no benchmarks seen. no one even has one. no leaks.

this proves 1 thing. the card is either not ready, or performs like CRAP.
and its the second one. the card performs like CRAP, and they cant clock it high enough to beat the ATI cards. so they are scratching their heads at nvidia, thinking how to sell this thing.

The other thing is, this card is going to pump alot of heat into everyones case. I have no idea why you all think nvidia will just bring some magic three months from now. this card will be the FX 5800. the NV30 that was a piece of crap.

for those that are waiting for DX11 from nvidia forget it.
ATI is your company. at least for another 2 years from today.

Ahahahahahahahahahaaaah hahahahaha!!!

cock.

 

[BinarySynapse]

I cant believe the amount of people who still think , nvidia's fermi will beat ATIs hardware.

3 months before release, no cards seen. no benchmarks seen. no one even has one. no leaks.

this proves 1 thing. the card is either not ready, or performs like CRAP.
and its the second one. the card performs like CRAP, and they cant clock it high enough to beat the ATI cards. so they are scratching their heads at nvidia, thinking how to sell this thing.

The other thing is, this card is going to pump alot of heat into everyones case. I have no idea why you all think nvidia will just bring some magic three months from now. this card will be the FX 5800. the NV30 that was a piece of crap.

for those that are waiting for DX11 from nvidia forget it.
ATI is your company. at least for another 2 years from today.

Yeah, it always turns out this way. I mean they did the same thing with the 8800GTX back in 2006... wait...

 

[vengence]

I cant believe the amount of people who still think , nvidia's fermi will beat ATIs hardware.

3 months before release, no cards seen. no benchmarks seen. no one even has one. no leaks.

this proves 1 thing. the card is either not ready, or performs like CRAP.
and its the second one. the card performs like CRAP, and they cant clock it high enough to beat the ATI cards. so they are scratching their heads at nvidia, thinking how to sell this thing.

The other thing is, this card is going to pump alot of heat into everyones case. I have no idea why you all think nvidia will just bring some magic three months from now. this card will be the FX 5800. the NV30 that was a piece of crap.

for those that are waiting for DX11 from nvidia forget it.
ATI is your company. at least for another 2 years from today.

Obvious Troll is Obvious.

btw, as far as no one has benchmarks even leaked ones.... hardware-canucks claims to have some. I'm not saying they are right, but wow you need to stop drinking the kool-aid

 

[Digital Viper-X-]

Yeah, it always turns out this way. I mean they did the same thing with the 8800GTX back in 2006... wait...

Really? the 8800GTX came out 6 months after ATI's DX10 card? Oh wait, it was the opposite, and the 2900 sucked, he could be onto something, it's very possible the reason Fermi isn't out yet, is because it can't compete with the ATI offers at a price / performance ratio, I mean if the high end fermi, barely beats out the 5870 and costs $200 more who's going to buy it other then the raving fanbois?

 

[Mr. K6]

Really? the 8800GTX came out 6 months after ATI's DX10 card? Oh wait, it was the opposite, and the 2900 sucked, he could be onto something, it's very possible the reason Fermi isn't out yet, is because it can't compete with the ATI offers at a price / performance ratio, I mean if the high end fermi, barely beats out the 5870 and costs $200 more who's going to buy it other then the raving fanbois?

Tough to say. NVIDIA sure as hell isn't delaying Fermi for shits and giggles. I'd guess it's probably faster than the 5870 - it'd have to be at this rate - but at what monetary/power/pride cost?

 

[vengence]

Really? the 8800GTX came out 6 months after ATI's DX10 card? Oh wait, it was the opposite, and the 2900 sucked, he could be onto something, it's very possible the reason Fermi isn't out yet, is because it can't compete with the ATI offers at a price / performance ratio, I mean if the high end fermi, barely beats out the 5870 and costs $200 more who's going to buy it other then the raving fanbois?

That doesn't make any sense. They have a worse preforming part so instead of launching it, they are going to wait till the hype is bigger so it can fail more?

 

[Spewn]

Obvious Troll is Obvious.

btw, as far as no one has benchmarks even leaked ones.... hardware-canucks claims to have some. I'm not saying they are right, but wow you need to stop drinking the kool-aid

What's with the "simulated" 5870 benchmark results? How do you do that?

 

[cannondale06]

That doesn't make any sense. They have a worse preforming part so instead of launching it, they are going to wait till the hype is bigger so it can fail more?

yeah his comment make no sense. if they have product ready then of course they would release it because low sales is still better than no sales.

 

[Gorankar]

What's with the "simulated" 5870 benchmark results? How do you do that?

They have a note about it at the beginning of the article. They are showing us Nv provided benches from an alleged fermi rig they can't 100% duplicate for the 5870 comparo. They only got close with the one they put the 5870 in. The numbers are prolly within 10% of the truth. Take them with salt or with out at your own discretion.

 

[vengence]

What's with the "simulated" 5870 benchmark results? How do you do that?

They have a note about it at the beginning of the article. They are showing us Nv provided benches from an alleged fermi rig they can't 100% duplicate for the 5870 comparo. They only got close with the one they put the 5870 in. The numbers are prolly within 10% of the truth. Take them with salt or with out at your own discretion.

Yeah pretty much. While I'm sure Nvidia has tilted the benchies in their favor, I doubt they are stupid enough to distort it so much it has no basis in reality.

 

[Mr. K6]

That doesn't make any sense. They have a worse preforming part so instead of launching it, they are going to wait till the hype is bigger so it can fail more?

What hype? No hype is going to build until some new info is released or a launch date is set. If they wait to launch it, they can work longer on improving the part.

 

[BinarySynapse]

Really? the 8800GTX came out 6 months after ATI's DX10 card? Oh wait, it was the opposite, and the 2900 sucked, he could be onto something, it's very possible the reason Fermi isn't out yet, is because it can't compete with the ATI offers at a price / performance ratio, I mean if the high end fermi, barely beats out the 5870 and costs $200 more who's going to buy it other then the raving fanbois?

You might well be right. My point was that his assertion that Nvidia's silence PROVES that Fermi is going to suck is fallacious.

Before the 8800 came out, ATI was expected to beat out Nvidia because of rumors that the 8 series wouldn't be fully DX10 compliant and that Nvidia couldn't get the performance from teh chips.. blah blah blah... Look how it turned out then... The 8800 kicked ass and the 2900 came late to the party and forgot the punch.


I'm NOT saying that Fermi is going to kick ass. I'm saying that it being late to the show and not hearing anything official about it proves absolutely dick about how the card is going to perform.

 

[Digital Viper-X-]

That doesn't make any sense. They have a worse preforming part so instead of launching it, they are going to wait till the hype is bigger so it can fail more?

man you're so good at seeing the bigger picture, they delay it, to revise it, and tweak it as much as possible to make up the gap, even if it's 10% faster then the 5870, but costs $200 more, who the hell is going to buy that when a 5970 is at the same price point? the whole point is in simple math for you

card not fast enough to compete with 5870/5970 on price/performance = card not ready
card not fast enough to compete period(it's slower then 5870) = card not ready
card not ready for mass production = card not ready

if Fermi was ready, it would be launched, since Tesla cards are "launched" and ready, then GTX 470/480 should be too, since the hardware is nearly identical, the only other explanation is that Fermi doesn't compete well with AMD cypress / hemlock cards.(as in either slower, or marginally faster to considerably more $)

 

[vengence]

What hype? No hype is going to build until some new info is released or a launch date is set. If they wait to launch it, they can work longer on improving the part.

Right, they magically can take an 18 month design cycle and suddenly make architectural changes that turn it from uber suck to epic win in 6 months.

 

[vengence]

I'm NOT saying that Fermi is going to kick ass. I'm saying that it being late to the show and not hearing anything official about it proves absolutely dick about how the card is going to perform.

man you're so good at seeing the bigger picture, they delay it, to revise it, and tweak it as much as possible to make up the gap, even if it's 10% faster then the 5870, but costs $200 more, who the hell is going to buy that when a 5970 is at the same price point? the whole point is in simple math for you

card not fast enough to compete with 5870/5970 on price/performance = card not ready
card not fast enough to compete period(it's slower then 5870) = card not ready
card not ready for mass production = card not ready

if Fermi was ready, it would be launched, since Tesla cards are "launched" and ready, then GTX 470/480 should be too, since the hardware is nearly identical, the only other explanation is that Fermi doesn't compete well with AMD cypress / hemlock cards.(as in either slower, or marginally faster to considerably more $)

Volume. It all comes down to volume. If you have a part that is hard to manufactor, you push what you do make into the highest margin cards while you fix the manufactoring problems. It doesn't matter if the card is 10 times faster than a 5970 if you can sell it for more than 10 times the margin as a tesla, they will sell tesla till production can saturate tesla demand. It has nothing to do with the preformance of the card and has everything to do with the insane margins on Tesla parts.

 

[Digital Viper-X-]

Volume. It all comes down to volume. If you have a part that is hard to manufactor, you push what you do make into the highest margin cards while you fix the manufactoring problems. It doesn't matter if the card is 10 times faster than a 5970 if you can sell it for more than 10 times the margin as a tesla, they will sell tesla till production can saturate tesla demand. It has nothing to do with the preformance of the card and has everything to do with the insane margins on Tesla parts.

I don't think the amount of $ they can on Tesla cards will be greater then the amount they will make on full range of DX11 cards, every day that they don't have an offering, gives AMD more time to saturate the market and take away sales, and I'm not talking about $300 GPus I'm talking about $60-$150 gpus that most people will end up buying.

so the only logical reason to continue to delay these cards is that they simply are not performing or the cost to get them performing is too high, pretty sure low-mid range cards are not as hard to manufacture

 

[Gorankar]

This is just me passing gas really. It's something I have been thinking since Nv essentially killed off the high end 2xx cards with out having a replacement.
I think when they first layed the groundwork for Fermi, that Nv expected to be able to charge the same $550-600+ for single gpu high end cards and over $700-800+ for dual gpu cards. Like they were able to with the 7xxx and 8xxx series. I think Nv may have engineered the 2xx and Fermi around those price points.

Then AMD's 4xxx screwed up those price points by releasing cheaper cards that performed almost as good. Even worse, AMD later released the, 5xxx series tha Nv has still not answered with a shipping product. This has changed the price point for a high end single gpu card to $400 or so, and $650 for dual gpu cards imho.

If Fermi was engineered for those higher price points, that it now may not be possible to sell at, then Nv's lateness to market may partly just be them having issues scaling Fermi back to be able to compete at the lower price points AMD has established, while still keeping Fermi profitable. Not saying Nv can't sell single gpu Fermi for more than $400, but they will not be able to get a 8800Ultra level price for it except out of the die hard fans. Unless it royally blows away the 5xxx and AMD does not answer with a refresh that catches part way back up.

Just my opinion on a possible reason for the delays. It could just be that Nv ran into unforseen problems and the chip got delayed. Shit happens.

 

[vengence]

I don't think the amount of $ they can on Tesla cards will be greater then the amount they will make on full range of DX11 cards, every day that they don't have an offering, gives AMD more time to saturate the market and take away sales, and I'm not talking about $300 GPus I'm talking about $60-$150 gpus that most people will end up buying.

so the only logical reason to continue to delay these cards is that they simply are not performing or the cost to get them performing is too high, pretty sure low-mid range cards are not as hard to manufacture

Actually, the GF100 Tesla hasn't launched yet. Oh, and some of the manufactoring problems have been with the 40nm processes so they effect everything, beyond that it is based on die size.

 

[vengence]

This is just me passing gas really. It's something I have been thinking since Nv essentially killed off the high end 2xx cards with out having a replacement.
I think when they first layed the groundwork for Fermi, that Nv expected to be able to charge the same $550-600+ for single gpu high end cards and over $700-800+ for dual gpu cards. Like they were able to with the 7xxx and 8xxx series. I think Nv may have engineered the 2xx and Fermi around those price points.

Nv did engineer the 2XX around those price points. I think the GF100 was made closer to the revised price point though.

Then AMD's 4xxx screwed up those price points by releasing cheaper cards that performed almost as good. Even worse, AMD later released the, 5xxx series tha Nv has still not answered with a shipping product. This has changed the price point for a high end single gpu card to $400 or so, and $650 for dual gpu cards imho.

I STILL don't understand ATI's logic of undercutting Nvidia so hard on the 4XXX series. Yes, they gained market share, but man did they loose out on profits. They could have easily made 3 or 4 times thier profit margin on those cards. I'd happly trade 1/2 my sales for 3 times the margin.

If Fermi was engineered for those higher price points, that it now may not be possible to sell at, then Nv's lateness to market may partly just be them having issues scaling Fermi back to be able to compete at the lower price points AMD has established, while still keeping Fermi profitable. Not saying Nv can't sell single gpu Fermi for more than $400, but they will not be able to get a 8800Ultra level price for it except out of the die hard fans. Unless it royally blows away the 5xxx and AMD does not answer with a refresh that catches part way back up.

Refreshes are never more than about 10-15% faster than the original. I don't think there is any real chance of ATI coming through with a "killer" refresh. I ran the numbers in another thread, but even at 100$ per GPU die (current rumor mill averages around 50$ per die) they could sell it at 400$. I don't think they are going to as the current rumor mill puts it around 25% faster than a 5870. Given the current price of the 5870 and an assumed price drop to 350$ at the launch, we'll probably see a 400$ GTX 470 and a 500$ GTX 480. The 470 being linear price/preformance and the 480 being nonlinear price/preformance.

Just my opinion on a possible reason for the delays. It could just be that Nv ran into unforseen problems and the chip got delayed. Shit happens.

Rumor mills point toward them being on a third spin to get acceptable manufactor rates.

 

[Arklight]

I STILL don't understand ATI's logic of undercutting Nvidia so hard on the 4XXX series. Yes, they gained market share, but man did they loose out on profits. They could have easily made 3 or 4 times thier profit margin on those cards. I'd happly trade 1/2 my sales for 3 times the margin.

It is a fair assumption that the people who work for ATI know more than you do and probably came to the conclusion that they would not be increasing their margins by 3 to 4x. lol

 

[vengence]

It is a fair assumption that the people who work for ATI know more than you do and probably came to the conclusion that they would not be increasing their margins by 3 to 4x. lol

I think they have very capable engineers and economists. However I don't put much faith in management and marketers, they can screw anything up.

 

[Naieve]

I'm sure you can find it, but Nvidia definitely built the GF100 with lower price points in mind. One guy higher up the ladder at NV said they made sure they didn't make the mistake they did with the GTX 200 series.

Personally I think the real problem is that they are making a massive chip on a glitchy process. The 40nm has been nothing but problems for both ATI and NV. So since you figure that the extra size of the NV chips translates to any problem on the die making a bigger impact than it would on the smaller ATI chips. Instead of one problem causing 1 unit of ATI die to be bad, for Nvidia it would be like having 1.5 units of the die go bad. Add in the fact they took longer to get them into TSMC's hands, that means its going to take longer to fix those flaws.

 

[Sunin]

to me the price point is not the problem as long as the performance is there... That is truly the key part... if ATI is 400 for a card but NV is 600 but 1.5x faster then its an even swap and on a per card basis I have to go for the 600 card.

We will see. They may have learned, but I wonder if that means they will bring out the economy tiers as well in order to still keep the Performance card at the 400-600 range and the other cards instantly available?

 

[DualOwn]

$499 for the 448 shaders, im sure of it 99%.

 

[spine]

Personally I think the real problem is that they are making a massive chip on a glitchy process. The 40nm has been nothing but problems for both ATI and NV...

Exactly.

Nvidia's Fermi chip is insanely large for the process it's on. It's probably a world record.

And here's thing, while you might to be able to get 40nm to acceptable yields with a relatively small cpu die, the same may not be true for the huge Fermi die.

This is Nvidia's biggest problem, die wastage.

However, I think that they'll have optimised their process to mean that defective areas of any given wafer tend to result in just a bunch of shader cores being shut off. This means that defective dies aren't total wastes and can be sold off in budget cards.


Even so, I bet the delay is squarely in getting good yields for the high end parts.

 

[vengence]

Exactly.

Nvidia's Fermi chip is insanely large for the process it's on. It's probably a world record.

There are only 2 major discrete graphics card makers, so anything either of them does is a world record.

And here's thing, while you might to be able to get 40nm to acceptable yields with a relatively small cpu die, the same may not be true for the huge Fermi die.

This is Nvidia's biggest problem, die wastage.

It isn't intel's fabs. It's TSMC. Which is also different from global foundries. Each one's process presents its own unique pros and cons. Unfortunantly when they start designing these chips neither ATI nor Nvidia knows how good or bad the process will be. While it's not a shot in the dark, it can be a shot at dusk. ATI has gone "more conservative" with thier guesses at how bad the processes will be. The good part of that is cheaper chips, the bad part is slower chips.

However, I think that they'll have optimised their process to mean that defective areas of any given wafer tend to result in just a bunch of shader cores being shut off. This means that defective dies aren't total wastes and can be sold off in budget cards.

Yep. As long as they can turn off a few units and get something they can put into a GTX 470 instead of a GTX 480 your yields improve significantly. Infact (IIRC) that is exactly what they were doing with the 260s until the yields came up and they released the 260 216 core and eventually the 275 240 core.

Even so, I bet the delay is squarely in getting good yields for the high end parts.

The delay is because they had to respin to an A3 revision. That is a little bit of semantics though as they were simply reworking the chip to get better yields. Of course you could say they made changes to the design to make be faster than A2 at a set yield rate, or you could say you kept the same target speeds but increased the yield rate. In the end between the leaky processes that is TSMC's 40nm and the large die size you have a hard to manufactor product.

The end of march being 7 weeks away we should start hearing more soon.