Nvidia G71 - rumours, questions and whatnot

I can't see how you figure. If the 7900 GTX equals the X1900 XTX and is priced at $499 as DigiTimes has suggested, any fool in there right mind would be able to see that the 7900 is the better deal by far.

 

I wouldn't say "by far". There's only a 4% difference between the XT and the XTX.

 

If G71 is 24 pipe, it may match X1900XTX, but in newer games such as FEAR, ATI will retain the advantage.

Why would Nvidia price 7900GTX at $499 if it is comparable? With new top end card, Nvidia have increased price.

 

why we don't count nVIDIA puting 24 Pipe and 3 ALU in each pipe , that can explain the moderate clock speed ?!

 

Maybe for G80, but G71 is generally viewed as more like a refresh, and such a large architecture change is thus considered doubtful.

 

That would be a large architectureal change, something that is not expected for a late life update.

The R5xx was designed from the beginning to use multiple shader processors per texture unit (even thought the R520 maintained a 1:1 ratio).

 

Ah. My misunderstanding. Moving on!.............

 

The more important question is why not? Yes, NVIDIA has typically matched ATI on price or increased price, but in all of those instances NVIDIA held the performance crown the majority of the time. If they break even or slightly underperform, it makes complete sense that they'd try to kill ATI in a pricing war. Think about it, if they're even and NVIDIA prices the 7900 GTX at $649, then it's a toss-up as to which card enthusiasts buy and it pretty much comes down to preference. Price the 7900 GTX $100-150 under the X1900 XTX and the majority of those ppl buying based on preference are going to forego that to save money. Only the hardcore ******s or performance zealots are going to drop the extra cash in a situation like that. Not to mention this hasn't even factored in manufacturers like BFG and eVGA who are definitely going to push the card higher if they can.

 

... and if G71 is indeed just a die-shrink, it's going to be cheap compared to R580.

 

It's not a large change for a refresh; think of NV40 vs. G70 or R520 vs. R580. The question would rather be if it makes any sense.

 

Am I the only one that thinks from a historical perspective that it'd be funny as hell if ATI has a much bigger die than NV at the same process node for the flagship gpu? I can't be, can I?

 

Why exactly would it be funny as hell?

 

Because it's a reversal of most of what each has said over the last few years. ATI has plugged "smaller" chips with no extraneous unusable "checkbox" features, clocked high.

NV has been very loud that what they learned from NV30 was to rely on transistors, not clocks --they'd rather have a big chip with a lower clock. When Wavey asked Jen-Hsun about large chips, he basically said "that's my problem, customers don't care how big it is". If (and this is still a BIG IF) G71 is "just" a die-shrunk G70, then it is going to be quite a bit smaller than R580. G70 is already a touch smaller at 110nm.

 

Wow, it's really going to be that smaller? That's why I did not get your point earlier. Well, I can't wait to see if ATI's memory controller pays up. The G71 seems much more conservative.

One thing I've observed about ATI's latest architectures is that they perform well after they are about one year old or older. They seem to make design choices that are more forward looking. Like for example the number of shaders in the r580, they don't seem so usefull on current games, but I bet the x1900 is a better buy(compared to GTX 512) if you plan to keep it for at least 2 years.

 

But R5XX had a new scheduler designed to accommodate the scaling. You may be right, of course, but I still think it would take some work to do this going from G70 to G71.

 

Right.
http://pc.watch.impress.co.jp/docs/2006/0127/kaigai01l.gif
and I also think, thats useless to raise the raw power that much when there is no change to the memory interface.
Only three weeks till launch and we still dont know anything exactly. Imo not a good sign.

 

How sure are you about 32pipes? The InQ says 24 and 650.

 

Well, Anandtech claims to have a die shot of the 7900 GT:
http://www.dailytech.com/article.aspx?newsid=907

Anybody want to try to measure the die size? I would assume that this is the same as the higher-end G71, just with some pipelines disabled.

Edit: Visually, it looks a lot smaller than the 7800GTX, X1800, or X1900 chips. But the lack of a heat spreader could be deceiving me.

 

I suppose that you could say that in purely die space terms 3(MADD+ADD)=BADD

ATi have added say 20% more transistors for the massively increased fragment shading and taken the hit on the die size which has gone up by 20% as well. The question is whether this 20% increase in die size gives them a good increase in speed and lets them charge more to willing punters.

nvida seem to be risking just increasing the speed and having enough yield whilst getting the benefits of a smaller die. They might be betting on being say 10% down in speed per chip if it is 20%+ cheaper and then relying on SLi to keep the "buzz" going on web forums up and down the land as them being quickest.

I think the r580 has got more legs in it, but by the time it starts stretching it's legs everyone will be talking about G80/R600 and the r580 will be as exciting as granny.

 

It looks in the 180-220 range to me after some quick measurements but that seems a bit low. I'd have to put a massive disclaimer on my ex-school wooden 300mm rule and "counting Pcie teeth" system I just used.