ATI FY 3Q 2006 Results and CC

[Geo]

That Joshua guy certainly writes as if he knows what he's talking about

You get that's JoshMST here? The penstarsys.com fella? Which presumably means they are paying him at least in cheese sandwiches, so good-oh on him.

 

[Jawed]

You get that's JoshMST here? The penstarsys.com fella?

I couldn't help wondering the same thing myself...

Jawed

 

[_xxx_]

Doesn't one imply the other?

No, high switching speed = higher current = more heat. Which is a limiting factor for "big" chips. I think for the gfx-chips they'd use the "medium" process as a compromise, but I'm definitely not sure.

 

[trinibwoy]

You get that's JoshMST here? The penstarsys.com fella?

Oh, didn't know that. Thanks.

 

[trinibwoy]

No, high switching speed = higher current = more heat. Which is a limiting factor for "big" chips. I think for the gfx-chips they'd use the "medium" process as a compromise, but I'm definitely not sure.

Yes it implies all those things as well of course but that's not what I was saying. I would think that higher overall clocks require higher transistor switching speed (or at least it's one way to get there).

 

[Jawed]

I think the problem is we don't really understand the mechanism that determines clock rates for a given process, e.g.:

• physical properties of the process, voltages, leakage erm...
• maturity of the process
• nature of the libraries
• yield curve versus die size

etc.

Jawed

 

[Sunrise]

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

According to some, TSMC’s 80nm standard is three to four months late, as ATI was expecting to start production on 80nm parts in early spring.

In that context, only relying on your common sense, with a little bit of insight:

Why should ATi draw an AIB roadmap in January that outlines RV560/RV570 as being ready for P in ~July/August (which seems to be only about 1 month off from current plans), when some of TSMC´s 80nm process targets are/were having a "delay" of 3-4 months? "Early spring" is starkly far-fetched.

TSMC´s 80nm process and it´s process targets need time to be fully qualified, ATi knows that, people that work in the industry know that, it´s nothing unusual, Josh should know that, too.

However, if we strictly talk about only the sampling date in general, ATi´s roadmaps were adjusted from S: in March to S: in May and there could be about a dozen reasons for that.

 

[_xxx_]

Yes it implies all those things as well of course but that's not what I was saying. I would think that higher overall clocks require higher transistor switching speed (or at least it's one way to get there).

Hard to explain, but not really. You need the right balance between sw. speed and the current dissipation to reach highest possible clocks.

EDIT: the diameter of the wires/routes limits the current draw as well, so there's only so much "stable" power you'll be able to pump into the chip on a given process. Unless you nitro-cool it, of course

 

[Jawed]

TSMC´s 80nm process and it´s process targets need time to be fully qualified, ATi knows that, people that work in the industry know that, it´s nothing unusual, Josh should know that, too.

I don't for a minute think anyone's forgetting that stuff.

When TSMC apparently started having problems with 80nm, they presumably kept their customers informed. ATI would have revised its roadmap, and this is the roadmap that reflects the status as of late 05/early 06. Though it's turned out to be over-optimistic.

However, if we strictly talk about only the sampling date in general, ATi´s roadmaps were adjusted from S: in March to S: in May and there could be about a dozen reasons for that.

I presume you're referring to some other roadmap than the one I've shown. I can't find anything that's got S:May on it.

Oh yeah, why has RV535 dropped off radar?

Jawed

 

[trinibwoy]

Hard to explain, but not really. You need the right balance between sw. speed and the current dissipation to reach highest possible clocks.

Ok, I think I understand what you're saying now.

 

[Sunrise]

....and this is the roadmap that reflects the status as of late 05/early 06. Though it's turned out to be over-optimistic.

There is no roadmap that talks about late 05/early 06. Either you simply mix up different things here, or you have access to information not known to the public.

I presume you're referring to some other roadmap than the one I've shown. I can't find anything that's got S:May on it.

These are all still accessible through this board / HKEPC.

WRT RV535: Please keep this discussion with RV560/RV570, otherwise one side doesn´t know what the other side is talking about.

 

[Jawed]

There is no roadmap that talks about late 05/early 06. Either you simply mix up different things here, or you have access to information not known to the public.

I was referring to the roadmap I posted earlier. I don't know the precise date it was created (creation date of that picture is 21 February). I guess it reflects the status as of the beginning of the year. Stuff that gets leaked doesn't always leak the day it gets created.

These are all still accessible through this board / HKEPC.

I don't have the pix and I've become tired of searching. I was hoping you would actually show what you were referring to.

WRT RV535: Please keep this discussion with RV560/RV570, otherwise one side doesn´t know what the other side is talking about.

Fair enough. Until there's some facts about the progress/delays in 80nm it seems to me there's nothing more to say.

Jawed

 

[Sunrise]

I don't have the pix and I've become tired of searching. I was hoping you would actually show what you were referring to.

Hm. First post of B3D´s "RV560/570 Gemini roadmap" thread. Thought you still remembered that one.

 

[Jawed]

I've got that, but scanning over thumbnails its appearance as anything but a "roadmap" made me miss it, sigh

Jawed

 

[Maintank]

X1900XTX king of the hill. "Others forced to double up gpus per card" to compete. <Zing!>

Heh, well your competition can do something like that when they have a product at nearly half your die size and consumes less power.

He shouldnt really be emphasizing this imo as it is actually a blemish for ATI. Yes nvidia is using two cores to do this, but their two cores dont cost much more than his single core, eat less power, and delivers higher performance while driving his high end down.

 

[Blacklash]

Thanks, Geo.

I always love how Orton defends his highend parts...lol



Anywho, seems like a pretty good quarter for the most part, mostly in the consumer sector.

He ought to look at how X1900 XT Crossfire performs vs a single 7950 GX2 @ 1600x. It is also half the required space, more quiet, and has much reduced power consumption. It sometimes trails by 10% and other times leads by the same. It's a better all around deal if you play @ 1600x.

http://www.xbitlabs.com/articles/video/display/nvidia-gf7950gx2.html

You may also currently own it for around 290usd less than X1900 Crossfire.

I didn't mind my Crossfire setup til I changed to a super quiet Antec P180-B case. I have the below config currently and am just waiting on my E6600.

Intel Pentium D 930 @ 3.9 GHz|Mushkin XP2-6400 (3-3-3-10)
Asus P5W DH Deluxe (i975x)|EVGA 7950 GX2|Samsung SyncMaster 214T
WD Raptor 74GB HD(x2)|Pioneer-711B5PK|Auzentech HDA X-Plosion|Klipsch
ProMedia Ultra 2.0|FSP FX700-GLN|Antec P180 B|Windows XP Professional x64 Edition