Nvidia G71 - rumours, questions and whatnot

So nVidia is taking the midrange to new level (extra power connector, 256bit membus, quite hefty power-stuff) then?

 

Seeing the projected difference in performance/price between the GT and GTX versions, I could even imagine GT being a different (smaller) chip than the GTX, not the disabled-pipes-version.

 

If r520 was under endowed in the shader department and slightly unbalanced then increasing the theoretical shading power by 300% just to give a 50% increase in current extreme siutations surely has tipped it massively the other way and this currently unblanced design is where Ati have made their big gamble.

As mentioned above looking into the future is speculative and so whether this gamble will pay off or not is the big unknown. By the time r580 is showing it's muscle in games it might be up against two even cheaper smaller 80nm nvidia chips ???

Interesting though, nice to have a dichotomy on how to skin a cat quickest.

 

Possibly. But they wouldn't both be labeled "G71" if this were the case.

 

Geo, at least you´re having some brains, while i´m beginning to doubt Kristopher´s knowledgable "sources". You can safely go to sleep now, since from what i could measure, you´re spot-on.

Just for comparison (7600 vs. 7900GT):

 

That does say that the rumored "7600" and "7900" chips are one and the same. That doesn't absolutely state that the rumored "7900" chip is too small to be a 24-pipeline G7x part. If you could do the same comparison between one of those shots and and older shot of, say, a 6600GT die, then that would be a better basis for comparison.

Edit: Die shot of the 6600 here:
http://www.beyond3d.com/misc/chipco...id=71&orderby=release_date&order=Order&cname=

 

You got it.

 

Certainly not disagreeing with you here, but it's a little odd to see a 7800-style PCB when the early snaps from Tweakers.net (now pulled) showed something more like a 6600.

I believe the 7800 has memory chips on both sides of the PCB, so conceivably the odds-are-128-bit 7600 could use the same PCB as the 7800 or 7900, but with chips on the front only.

Edit: on further comparison, that PCB is actually significantly smaller than the 7800's, so I guess this is just the 7600 GT while Tweakers had the LE/GS.

 

Even less than that actually more like 16 pipes would fit in there!
Its so small from the look of it that im thinking the size of a 6600GT die!
Something is not right here..

Edit seems that other have come to reasoning too, wtf are Anandtech doing!?
Edit2 Actually ALL! should start read through threads not just replys, my bad..

 

7800's used different board designs, and the 7800 GT is smaller. That board is more inline with the 7800 GT board size, the layout is also consistent with a 256-bit bus chip design.

IMO, that is a 7900 GT board, the question is whether that is a "staged" leak and thats not the correct chip / package on it (for example, why are there free pin positions around the package?) On the flipside, is it not an assumption that VR-Zone's G73 chip is actually correct?

When trying to evaluate die sizes, relative to others, you might have a better frame of reference from naked board shots from reviews.

 

Thanks Dave. By my highly-scientific finger-to-monitor measuring method it's still a bit smaller than the 7800GT, but not nearly so drastic (and improbable) a reduction as I initially believed. These are the chips used, so there's 256-bits worth of memory on the front of the card. My red-faced retraction of the previous post goes here.

 

Btw, notice the 2 sizes of square markings on the socket?

 

And the "white dots" that are completely visible number 128.

So this is a common board for 7600GT (128-bit bus) and 7900GT (256-bit bus).

Jawed

 

However in the DailyTech pictures we have a 128-bit chip coupled with 256-bit worth of memory.

 

Le oops. http://www.beyond3d.com/misc/chipcomp/?view=chipdetails&id=71&orderby=release_date&order=Order&cname=

Edited upstream.

 

To be honest I'm still scratching my head over this...

Jawed

 

Or someone dropped a 7600 package in a 7900 pcb? "Why?", you ask? "Elifino", I answer.

 

Yeah, that's the one I meant to link last night. Apparently I never actually hit the copy tho. Doh. You'll note the pattern around the chip is the same.

At any rate, I'm still not buying that's a 7900 in both rather than a 7600 in both. To believe that is a 7900, I'd have to believe that they've accomplished a 256-bit bus on a 150mm2-ish chip. . .which is about 45% smaller than the first measured consumer 256-bit chip, R300. (The caveat is apparently no one outside Matrox ever measured Parhelia under its heat-spreader).

Which, come to think of it, is another way of saying in my mind there's a lower min of around 215mm2 for G71, even at 24ps.

Edit: And, if I reverse engineer the problem, assuming 50% increased density at 90nm vs 110 nm (which is probably best case), I come up with 222mm2 for G70 at 90nm. So I'd guess the real deal is going to be between 222mm2 and 250mm2, if it's 6 quads, that is.

 

Btw that dailytech shots shows 256M or 512M of memory?

 

http://www.samsung.com/Products/Sem.../GDDR3SDRAM/256Mbit/K4J55323QG/K4J55323QG.htm