Nvidia G71 - rumours, questions and whatnot

nvidia has the time and money to do it , why not ? can't nv put another ALU in the G70 in 9 months or so , i guess G70 has been completed since april 2005 .

 

Possible try too use the memory modules as reference to get a pretty accurate diesize imo.
From a "first" look it seems very small as you say.

 

Aye. It seems that at least naiively, this would discount the possibility of a 32-pipeline part.

 

G70 doesn't have one either. FC only. It's certainly smaller. Power use may be reflected by the simplified power supply. Same memory arrangement, too. Interestingly, there are a couple of BGA rows around the chip. I wonder if the board is (or will be) common with another larger ASIC, or that this is an ES.

 

Hmm, the white markings would almost suggest that as well. Interesting...

 

At first glance that tiny die looks like bad news for Nvidia.

However a 24 card at 650 would probably eqaul or exceed the X1900XTX most of the time. It's 18% faster clockspeed. Going to anand's X1900XTX review and doing the math looks pretty good for 7900. However the X1900XTX will still have more brute strength.

Then figure it will be dirt cheap for Nvidia. They might have a winner in a different way. Highest end card in the $300's? See the possibilities?

However I dont. Seems Nvidia just aimed too low. It's like why the Unreal engine 3.0 was a success. Epic just shot for the better graphics than say, Doom engine. They just shot for a higher mark.

ATI if all this plays out, just aimed higher. It's usually the correct choice in graphics.

 

Hmm, allow me to clarify. What I meant about Unreal 3.0 eclipsing Doom engine is simply that Epic shot for a higher graphics target when they started development. So their engine outshined others when things started turning concrete months and years later.

You just need to shoot for pretty much the highest bar possible in graphics. Nvidia should have aimed for a 32 pipe part when they made that decision months ago. They probably thought they wouldn't need it...

 

UE3 is (at least) a generation ahead of the Doom3 engine, both in when it started development, and in the first release games. So I don't think the analogy applies.

Now, as for this product, if it's reasonably high-clocked, it could be quite an excellent performer for the money, making it a nice cash cow for nVidia. Recall, for instance, that at the NV40 launch, the most successful of the higher-end cards was the 6800GT, because it combined good performance with a decent price. It seems clear that they're aiming for the same thing with this product.

Now, a GTX version of the 7900 on a die size like that should be fairly competitive with the X1900XT. But how competitive really depends a whole lot on the clock speeds. We'll have to see how it all pans out.

The small die size also leaves room for a significant die size increase for their next-gen product, allowing nVidia to not have to rely entirely upon efficiency improvements to keep ramping up performance.

 

Okay, something's really wrong here. Maybe the fact that's it the middle of the night and I'm only typing this because a cat walked across my head and woke me up.

I decided to use NV43 as a baseline comparison, using the shot here: http://www.beyond3d.com/misc/chipcomp/?view=chipdetails&id=71&orderby=release_date&order=Order&cname=

And I ended up at almost exactly the same size as NV43, 150mm2. "That can't be right!", I muttered. So then I started looking at the resistor (or whatever they are) pattern around the two dies. Look for yourself, it's exactly the same.

Compared to G70 here:

http://www.beyond3d.com/misc/chipcomp/?view=chipdetails&id=106&orderby=release_date&order=Order&cname=

So then I pulled up 7600 die shot from VR-Zone:

http://www.vr-zone.com/index.php?i=3219

Guess what? It's a dead ringer for the dailytech shot.

That ain't no 7900. That's a 7600.

And if I'm right, rather than merely sleep-deprived, that's *two in a row* to cuff dailytech for.

Either that or someone's got a serious mis-information campaign going.

 

What shot there?

Yeah, 150mm^2 is way too small for a G71. I don't have anything to measure it with myself, though, so I can't comment on that. On the off chance that you are just sleep deprived, and dizietsma's measurement of ~200mm^2 is correct, then it might yet be a G71 (90nm should be ~66% the die size of the same part on 110nm, so that would put it at just above 200mm^2 for a die shrunk 7800GTX).

 

Only if it yields well at 600MHz+

Jawed

 

A die-shrunk G70 is the safe choice, though. Particularly if it was originally planned for Nov or Dec 05.

Also, it is a refresh (though G70 over NV40 is a bit of a whopper of a refresh).

Jawed

 

I would like to know if anybody has seen an official shader core die size on Xenos?

Surely with millions in the wild somebody has taken a ruler to one?

 

In reality, R520 seems to have slightly too little ALU resources to keep up with its texturing and memory bandwidth (it was a little unbalanced, if you like - not a huge amount though). So any increase in R580 was welcome.

It's not too difficult to find R580 50%+ faster than R520 in extreme (perhaps marginally playable) scenarios.

Jawed

 

It's also a mental game.

R520 I knew lacked brute strength. Call me dumb but that reduced it's appeal to me.

R580 I know has a lot of math power, more than any other card. I'd be much more likely to buy something based of it. Psychology.

This is where R580 will have a big advantage if 7900 is only 24 pipes indeed, for the next six months..

 

I think people still care more about benchmark results than technical specs.

 

QFT

Jawed

 

Not me though.

Besides, raw power will show out in the future, so it's not really silly.

 

Perhaps. The future is always speculative.

 

Sometimes people forget that these cards are made to play games - somehow i can't picture many expert people from here as game players