ATI FY 3Q 2006 Results and CC

Why is that? I don't think the change from 24 to 32 pipes will increase transistor count a lot. R580 is twice as larger as g71. How much difference can the additional 8 pipes bring in in terms of the transistor count?

 

for nV's chips the way they are set up it would make around a 80 million trany difference. The r580 is around 380 mill trany's., opps nV's chip will still be smaller but not by much

 

Hmm, refering to yourself in the third person..... Clear sign of green boxites (also known as demintus positis reputatium)



j/k
Thanks for the summary, much appretiated.
epic

 

$40M+ profit isn't good enough? How is $40M in profit failing?

 

Heh. Not that it matters of course, but NVIDIA's focus on margins definitively have given them a financial edge over ATI. At the same time, they pretend it barred their entry in some markets (such as 2G mobile phones) because the margins wouldn't be worth the trouble, and I question whether that was such a good idea personally.

Uttar

 

Given ATI's margins are 50% plus in that arena, it would tend to suggest that to attempt to compete with ATI they are having to undercut.

 

In regards to the RV570 it does look like this will be much needed in the mid range space, however I could not quite grasp why it would be delayed if yields were so good ? Maybe there is no delay ?

Regards

Andy

 

Maybe TSMC has had problems with 80nm - and this is why UMC is getting a look in. It's amazing that the analysts didn't press this point. Sorta: "Our mainstream parts are late, so our margins are hit by having to use enthusiast-level GPUs." "Oh, OK." Lame.

Jawed

 

Is it just me, or this produced no news whatsoever? Btw thx geo

 

I wouldn´t call it problems, it´s more to do with TSMC´s plans to offer every customer a substitute for their 90nm process targets, meaning they need time to fully qualify not only their low-end, low-power and high-output targets, but they also need to cope with high-speed targets since TSMC is known for their customers to be very demanding especially in this field.

UMC is getting a look in because UMC always was ATI´s low-end FAB with substancial volume and very good yields and quality services. Based on that, UMC always wanted to grow even more, so they forced themselves to be competitive with TSMC not only process technology wise, but also on pricing and naturally this also leads to price biding between those foundries, since they need every FAB to be fully utilized to make money. Nothing is worse then having bought millions worth of equipment and leave it unused. UMC actually was in-front of TSMC, WRT their 80nm process, meaning they already shipped final silicon, while TSMC still has the advantage (and that takes additional time) that they offer more process targets.

Yields may be not quite as good as they´re on 90nm, which isn´t a secret, but i don´t buy that exaggerated "yields"-thing, not even for a second.

ATi is working with TSMC very closely and have been for years, meaning they both depend on each other heavily. If you take a look at ATI´s very early roadmaps, dating back to the beginning of this year, they always forecasted/planned those parts to be in full production in late august, early september, so there may be some delay involved, but since Orton didn´t exactly specifiy what caused this delay (and i´m pretty sure it´s because they still have X1800/X1900 inventory, which needs to be sold first -> the inventory "hangover" he also spoke of), it´s understandable, but not quite "definitive" that this has anything to do with the quality of 80nm in general. It´s nothing that should come as an unexpected surprise.

 

I think there's a fair amount of delay:



If I could find the HKEPC article that featured this, then I would link it, but I can't

I imagine this roadmap is about a year old. RV560/570 is clearly indicated as April/May/June.

Jawed

 

Even if you could find that HKEPC article, it would be worthless, since RV560/RV570 parts planned for April are completely out of the question.

If you look at that roadmap closely, RV560/RV570 is indicated as July/August. If you add 1 month you get into early septemberish and i basically agreed with you here (read my post again), but i also outlined that this delay doesn´t have anything to do with process technology at all, which is one thing that is worth pointing out i think, because if you talk about "problems" they have to be unusual.

You´re looking at an "estimation roadmap" (which is an outlook, not definitive) from the beginning of this year, which should tell you that these estimations can only be met if everything works out as planned. A slip of about one month is nothing surprising at all and certainly not that big of a story.

EDIT: Typos fixed.

 

That certainly depends highly on their margin model and market demand. I guess at a certain point they have to, because either they´ll run out of X1800 ASICs, their yields improve to the point where >80% of R580 ASICs are 100% functional and/or demand for GTs is a lot higher than their inventory of non-functional ASICs.

 

EDIT: DAMNIT, i really need to check before i post XD I knew the connect3d x1800gto was no longer on sale, but now HIS has a $30 rebate on theirs, bringing it to $170. not bad, not bad at all. pretty much scrap my whole post, although i am worried about the margins (i'll leave my post intact in case anybody's responding to it now -- disregard it tho XD) .

honestly it really sucks that they're late with 560/570. if their target was July/Aug, AND they're late (Sep? Oct?), then... that bothers me. i love my Connect3d x1800gto (that I got on sale); since it was cheap, I find its performance to be mind-blowing, and more than sufficient for my 1280x1024 LCD in all current games. However, that sale price has not persisted, and the 7600GT has dropped lower and lower. They need a mainstream card fast -- didn't they learn that with the 9500p? You'd think that would have been a priority.

I realize shit happens, and I am happy with MY card, but I'm SURE the margin on it sucked -- and that's a problem.

 

http://www.theinquirer.net/default.aspx?article=32791

---

Can R600 be clocked significantly higher than R580? There's been thoughts about R600 being hundreds of MHz faster (admittedly none serious), but wouldn't that require a High Speed process? Or am I misinterpreting the concept of TSMC's HS

Presumably R600 on the 80nm standard process at TSMC is going to be 5%-10% faster than R580

So, pretty much in line with earlier rumours, apart from NVidia supposedly doing stuff on 65nm quite soon, presumably G8x value/mainstream.

Jawed

 

"HIgh speed" is a refence to transistor switching speed inside the chip or the overall clocks? I tend to think it's the former, would be the opposite of low-power.

 

That Joshua guy certainly writes as if he knows what he's talking about. A refreshing change coming from the Inq.

Doesn't one imply the other?