16 pipes for R420 too?

If the XT version does indeed have more pipes on the same core, why would ATi clock it so much higher? Seems to me they'd want to maximize performance at all levels. If the differing pipe specs are correct, I wouldn't expect XT to come clocked much higher.

I do remember that ad on Anandtech touting 600-800MHz GDDR3, so I really expect the new cards to hit the bottom end of that range (as Dave previously said 800MHz was in too short supply, though I can't imagine ATi or nV selling too many 140W 800MHz top-end cards).

Are we expecting reviews at launch, or after? We already know Derek at AT is looking into a monitor that can go higher than 2048x1536, though he won't totally fess up as to its use.

 

Yields

The XT will probably be clocked higher because of better yields. Not everything can be maxed out and you might as well clock your best as high as you can and charge a premium for it.

 

Actually the XT would be clocked higher regardless of yields. Bucause Frankly The XT version is only about one thing. Proving leadership. ATi have stated more than once that they are willing to sacrifice margins for performance this time around.

The bottom line is the XT will be whatever it has to be to be number one. Its that simple.

The Pro and SE versions are a different matter.

 

Does that mean even selling the XT at a loss?

 

Sorry, but with the Black Diamond process still maturing and ramping on the 13 micron process and the extreme shortage of high clocked memory (above 600mhz) 16 pipe versions of R420 are going to be extremely rare at first. The situation may improve with time, but I'll stand by my earlier assertion that if there is a 16 pipe R420 at launch it will be available in 5800 Ultra like quantities.

 

No i dont think that at all.. I am talking about sacrificing margins for speed. Not profits on individual boards for speed...

 

That is just simply total Nonsense....

The low-K Process they are using on R420 is going just fine and has been for some time. Perhaps, just perhaps you should look for some info outside Nvidia public statements on why *they* are not using it.

Your Memory information may need to use a refit as well.

 

My Low-K info comes from a digitmes article (published elsewhere also) some time back, that basically stated TSMC had started ramping the process and expected an adequate yield level out of the process late this year.

 

Hi,

Just a theory about the 170Milllion and 200+Million transistors from ATI.

It could be that the 12 Pipeline part has about 170 million active transistors. and the 16 pipeline part has 200+ million transistors. so basically the 30+mil transistors will be deactivated in the 12pipe part.

Hrishi.

 

You mean an article like this one from late 2002?

Perhaps more recent comments would be in order:

From here last fall.

Ati has been shipping 500+ mhz parts for a while based on low-k, I don't see why they can't have improved that process in the last 6+ months enough to continue to do so on a larger chip.

 

Where does he say that you praise the Grantsdale for it´s performance ?

Anyway, i don´t think that anyone doubts that they won´t get a big portion of the DX9 marketshare when it´s released. What i´m questioning is that it´ll be able to run the same codepaths as mid/high end codepaths with all the settings turned down. I believe that it´ll use the DX8 codepaths that all other low end DX9 (FX5200 f.e) and DX8 hardware have to run and thus won´t be of that much help for the developers.

 

* Natoma looks at calendar, looks at 38 pages worth of 'speculation,' looks at calendar again, then for good measure

We are going to find out all the details in what, 3 days? Like kids the night before christmas.

 

If the only way to get some nuggets of info out from Dave is by having 30+ page threads then by all means post. 8)

 

That´s praise for an underperforming hardware.

 

I didn't say you were praising their performance, I said you were treating theoretically equivalently performing parts differently. Your NV34 review seems to vindicate you, but on the actual bbs threads, you seemed to have come to a different conclusion.

e.g.

You agreed that the 5200 might perform about the same as a Grantsdale, yet you assert that developers will treat Grantsdale as a DX9 part (they'll just have to dial all the details way down), yet no one is going to treat the 5200 as a DX9 part with details dialed all the way. To me, there's a disparity here. Developers don't think of 5200 as DX9 because of performance, but they will of Grantsdale? Can't you see why I'm confused.

If Grantsdale performs really badly at DX9, what will developers do? You intimate that they refused to look at DX9 because of Nvidia performance.

 

I think what he's trying to say is, because the 5200 doesn't and won't have the marketshare Grantsdale will get, developers aren't considering DX9 as the base target yet. When Grantsdale comes out, and Intel sells millions of them, as they always do, developers will start going after DX9 as the baseline because the installed base of DX9 hw will then ecompass everything from bottom of the line integrated, all the way upto high end enthusiast.

 

But, just like the 5200, Grantsdale-G will likely be so slow that developers won't treat it as a DX9 part.

 

http://www.viewsonic.com/products/lcd_vp2290b.htm