The GTX 780, 770, 760 ti Thread *First review leaked $700+?*

[Xarick]

I am confused by that slide because they are showing the 760ti slower than the 670?

 

[blackened23]

I am confused by that slide because they are showing the 760ti slower than the 670?

Why would that be surprising

 

[sontin]

The difference between a GTX670 and GTX680 is in average less than 10%. I think it makes sense to make the cap larger so that people have a real reason to buy the GTX770.

 

[Smartazz]

The GTX 780 would have no competition from AMD. What incentive does Nvidia have not to charge a ton of money for it?

 

[Xarick]

Why would that be surprising

Because I was under the impression the 760ti is the same as a 670 but with higher clocks.. hence making it faster.

 

[toyota]

Hmmm.... This could mean $499-$549 is realistic...??

EDIT: Here is the full article. Still saying $599-$699 Range...

http://wccftech.com/nvidia-geforce-...ked-gtx-780-feature-gk110300-gpu-3-gb-memory/

According to the leaked roadmap, the GeForce 700 series would the flagship GeForce GTX 780 featuring a new GK110-300 A2 GPU and would be equipped with a 3 GB GDDR5 memory running across a 384-bit wide interface. It would feature 2496 Cores, 208 TMUs and 40 ROPs and would launch on 23rd May 2013

40 ROPs is not possible for a 384 bit bus for the way nvidia does memory

 

[blackened23]

Another thing that caught my eye, is the 770 being labelled as the "GK104-425". Maybe there are some additional improvements that make it faster than the existing 680.

 

[UaVaj]

The GTX 780 would have no competition from AMD. What incentive does Nvidia have not to charge a ton of money for it?

excellent question.

680 is msrp $500. competition is 7970ghz. $500 for this level of performance has been tamed and confirmed.

the last launch - titan (with zero competition). titan yielded 150% of 680's performance. that 50% of extra performance commanded a $500 premium. hence msrp $1000 ($500+$500).

using this same logic. 780 is predicted to be 30% faster than 680. that 30% of extra performance should command a $300 premium. that makes a 780 msrp $799 ($500+$300).

 

[toyota]

excellent question.

680 is msrp $500. competition is 7970ghz. $500 for this level of performance has been tamed and confirmed.

the last launch - titan (with zero competition). titan yielded 150% of 680's performance. that 50% of extra performance commanded a $500 premium. hence msrp $1000 ($500+$500).

using this same logic. 780 is predicted to be 30% faster than 680. that 30% of extra performance should command a $300 premium. that makes a 780 msrp $799 ($500+$300).

looking through reviews, overall I am only seeing the Titan being 40% faster than the 680. thats even more sad.

 

[motsm]

using this same logic. 780 is predicted to be 30% faster than 680. that 30% of extra performance should command a $300 premium. that makes a 780 msrp $799 ($500+$300).

You can't really make that comparison however, since the Titan is a new product line. The 780 however is part of the standard Nvidia family of cards, along with the 680, 580, and 480 before it. That doesn't mean NV won't price the 780 at some ridiculous premium, but I don't think using the Titan as evidence is very relevant here. I think the more relevant evidence would be the GTX 280's $650 launch.

 

[Will Robinson]

And so it continues....Titan convinces everyone that $800 is the new norm for a single GPU card.:\

 

[CitanUzuki]

Doesn't the compute performance of Titan justify the higher price tag?

 

[Soccerman06]

Doesn't the compute performance of Titan justify the higher price tag?

Not really, unless your slot limited

 

[Soccerman06]

double

 

[BallaTheFeared]

Doesn't the compute performance of Titan justify the higher price tag?

The DP?

 

[3DVagabond]

Doesn't the compute performance of Titan justify the higher price tag?

If you need/can use the DP performance and forgo any professional level support from nVidia, then yes, it would. It's a bargain.

 

[wand3r3r]

But even then isn't it only in like 5 pro apps?

Something about it in toms I believe.

 

[RussianSensation]

Doesn't the compute performance of Titan justify the higher price tag?

No. If you say firms that don't need ECC view Titan as a bargain for $1,000 for compute, then might as well argue that NV could have priced it at $1,500 or $2,000 since even at those prices it would be a 'bargain' for those customers vs. K20X. In that case why even call the Titan a consumer gaming card if we consider that it targets business that are cross-shopping K20X and Titan? Compute performance of the Titan is all over the place.

And so it continues....Titan convinces everyone that $800 is the new norm for a single GPU card.:\

It's worse than that. $450-500 GTX680s, $360-380 GTX670s, $290 HD7950s and $410 HD7970GEs 1 year after their launches, with almost no price drops. The entire GPU industry is failing hard. Consumers must be clueless to how GPU cycles work when they go shopping or people are getting nice bonuses and raises at work. Hopefully NV brings the GTX770 at $399.

 

[bleucharm28]

When I have the money(I don't have enough money right now-kinda broke at the moment)..
should I get a GTX 670 or wait for the GTX770?

(I currently have a single GTX480.)

I would wait due to more memory.

 

[Compddd]

Will Titan still be faster than GTX 780?

 

[Rvenger]

I don't believe this is being released on May 23rd at all.

 

[wand3r3r]

Just a hunch?

 

[Rvenger]

Just a hunch?

I can't say but I know its too soon.

 

[UaVaj]

if there is no competition (titan and 780) - there is absolutely no reason for price to drop.

until (1) amd can come up with some competition or (2) until every enthausaist who can afford a titan/780 buys out.

price will stay high. titan and 780 is a premium luxury exclusive product. you want to play - you have to pay.

 

[blastingcap]

No. If you say firms that don't need ECC view Titan as a bargain for $1,000 for compute, then might as well argue that NV could have priced it at $1,500 or $2,000 since even at those prices it would be a 'bargain' for those customers vs. K20X. In that case why even call the Titan a consumer gaming card if we consider that it targets business that are cross-shopping K20X and Titan? Compute performance of the Titan is all over the place.

It's worse than that. $450-500 GTX680s, $360-380 GTX670s, $290 HD7950s and $410 HD7970GEs 1 year after their launches, with almost no price drops. The entire GPU industry is failing hard. Consumers must be clueless to how GPU cycles work when they go shopping or people are getting nice bonuses and raises at work. Hopefully NV brings the GTX770 at $399.

I'll just quote myself below.

That wasn't a "marketing" slide, it was supposed to be a presentation to TSMC, basically. NV also wasn't thrilled that someone leaked it.

Look at the cost curve crossover slide (it's a different slide):

http://www.extremetech.com/computin...y-with-tsmc-claims-22nm-essentially-worthless

It's out of date and maybe 20nm is going better than expected, but Apple is also rumored to be taking a HUGE chunk of TSMC 20nm and Qualcomm and other non-GPU makers want their share, too, so I doubt 20nm wafers come cheap.

Anandtech has also commented that node sizes going down typically doesn't do a whole lot in terms of profit at first, since wafer costs go up to offset whatever you would have saved:

http://www.anandtech.com/print/2937/

"The problem is with any new process, the cost per wafer goes up. It’s a new process, most likely more complex, and thus the wafer cost is higher. If the wafer costs are 50% higher, then you need to fit at least 50% more die on each wafer in order to break even with your costs on the old process. In reality you actually need to fit more than 50% die per wafer on the new process because yields usually suck at the start. But if you follow the foundry’s guidelines to guarantee yield, you won’t even be close to breaking even.

The end result is you get zero benefit from moving to the new process. That’s not an option for anyone looking to actually use Moore’s Law to their advantage. Definitely not for a GPU company.


The solution is to have some very smart people in your company that can take these design rules and hints the foundry provides, and figure out which ones can be ignored, and ways to work around the others."

We used to see bigger jumps in price/perf because there were quicker cost recoveries from going down in process size and a mad dash to stuff more and smaller transistors in the same amount of silicon real estate. But as it becomes more and more costly to transition down in size, price/perf increases also slow down. (Improving architecture usually doesn't net a whole lot, with some exceptions like NVidia's 8800 series.)

Add in increased demand for TSMC wafers from the likes of Apple and you may see even slower price declines AND slower increases in performance if GPU makers elect to stay on one process size for longer. This is not the fault of AMD or NV per se. We're just running into quantum limits soon, and having mobile chipmakers bidding up wafer prices isn't helping. Moore's Law has had a good run but it looks like that run is soon over and we'll have to come up with something new other than more lithography shenanigans. Some people think 3D is the next phase. I don't know. I do know that ultimately this is a process problem. The architecture folks at Intel, AMD, NV, etc. can do only so much by themselves.