NVIDIA Kepler speculation thread

[sheepdogexpress]

The 660 is 1% faster at 1080p at Hardware Canucks, and 4% faster at 1080p on Anandtech.

It's not a lot no - but when the rest of the tech press has the 7870 some 5% to 12% faster on average you have to wonder where this 15% or so is disappearing to on these sites.

There are plenty of reviews that show the gtx 660 being faster.

guru 3d had a slight edge for the gtx 660.
Hot hardware had a slight edge to the gtx 660.
Benchmark reviews also favored the gtx 660.

And many sights were basically a tie like pc perspective and hardocp(when you consider the stock overclock on the 7870) and hardwarecanucks.

I think your trying to make a mountain out of a mole hill. We can say the same thing about website which had 12% in favor of the 7870. That is a particularly high number in favor of the 7870 unlike most sites. Are those guys shills now?

 

[boxleitnerb]

Differences up to 15% on average are negligible. Why? Because different architectures favor different games/engines and every site has a different selection of games they test. It is only natural that there is some variation.
All results are perfectly valid, you just have to look at the games you play to the the ones that are relevant to you.

 

[UniversalTruth]

Marketing got crazy or what? WTH, this card sucks for the asked price (230 for 660 http://www.newegg.com/Product/Produ...ption=gtx 660&bop=And&Order=PRICE&PageSize=20, and 210 after MIR for the much longer present on the market Radeon HD 7870 http://www.newegg.com/Product/Product.aspx?Item=N82E16814121624) but they open sooooooooooo big mouth about how "good" it is...

Compared to what exactly?

 

[boxleitnerb]

Marketing got crazy or what? WTH, this card sucks for the asked price (230 for 660 http://www.newegg.com/Product/Produ...ption=gtx 660&bop=And&Order=PRICE&PageSize=20, and 210 after MIR for the much longer present on the market Radeon HD 7870 http://www.newegg.com/Product/Product.aspx?Item=N82E16814121624) but they open sooooooooooo big mouth about how "good" it is...

Compared to what exactly?

8800GT. Yes, honestly. The 8800GT appears on the official marketing slides for the GTX660. That is just crazy.

 

[jimbo75]

There are plenty of reviews that show the gtx 660 being faster.

guru 3d had a slight edge for the gtx 660.
Hot hardware had a slight edge to the gtx 660.
Benchmark reviews also favored the gtx 660.

And many sights were basically a tie like pc perspective and hardocp(when you consider the stock overclock on the 7870) and hardwarecanucks.

I think your trying to make a mountain out of a mole hill. We can say the same thing about website which had 12% in favor of the 7870. That is a particularly high number in favor of the 7870 unlike most sites. Are those guys shills now?

Any site using Lost Planet 2 isn't even worth reading, and what a surprise all 3 sites you mentioned used Lost Planet 2.

HotHardware has 4 TWIMTBP games first (including Lost Planet 2), followed by 2 Gaming Evolved games. Not only that but they somehow managed to get the 660 faster than the 7870 in Metro.

Benchmark Reviews is run by Olin Coles, the guy who went nuts claiming AMD was cheating about something a couple of years ago (memory is scratchy on that one, but there was a huge spat over AMD apparently seeding the press with 6870's that had more shaders than the normal cards?). Guess what, he benchmarks 3 TWIMTBP games and only 1 Gaming Evolved game. He also used the 660 SC and didnt bother with reference benchmarks - is it any wonder that he proclaimed the 660 almost as fast as the 7970?

So sure if you trawl the depths of the tech press you will find a couple where the 660 is "faster". If we stick to what is generally thought of as the semi-trustworthy press then the consensus is that the 660 comes in-between the 7850 and 7870. Sites which have the 660 beating out the 7970 need never be mentioned again.

 

[silent_guy]

Jimbo,
Just to make sure I get this novel observation of you right. Are you suggesting that some games should not be benchmarked because it puts one company in a better light than the other? And that some sites are not following your advice? Maybe you should repeat that a couple of times to make sure we totally get this complicated concept.

 

[Sinistar]

Seems to work for Nvidia

We've omitted DiRT Showdown because the vast gulf in brand-based performance there skews the results pretty wildly

http://techreport.com/review/23419/nvidia-geforce-gtx-660-ti-graphics-card-reviewed/11

 

[jimbo75]

And it's not just techreport either - HardwareCanucks refuses to benchmark Showdown because - in SKYMTL's words - "it uses proprietary shader subroutines". Didn't stop him from benching Lost Planet 2 back in the day and I'll eat my hat if we don't see Borderlands 2 in their suite soon.

 

[ECH]

And it's not just techreport either - HardwareCanucks refuses to benchmark Showdown because - in SKYMTL's words - "it uses proprietary shader subroutines". Didn't stop him from benching Lost Planet 2 back in the day and I'll eat my hat if we don't see Borderlands 2 in their suite soon.

Wait, Borderlands 2 is an AMD game?

 

[AlexV]

NVIDIA always wrote motivational emails better, not sure why you guys are surprised. Ask yourself if it is AMD's fault for not scratching the right backs, or NVIDIA's for doing it. But take a long time wondering, and factor in that this is business, not a beauty contest. And IMHO, certain people in the press should show class and not just re-gulp what ends up in their inboxes...ninja bombad omagad shader subroutines oh noes!

silent_guy: a few more nice and informative posts on tech topics like process tech and far fewer posts in which you snipe awful awful people that don't like your IHV of choice would be quite appreciated, thanks. Let he who is without sin cast the first stone or something along those lines...

 

[Alexko]

And it's not just techreport either - HardwareCanucks refuses to benchmark Showdown because - in SKYMTL's words - "it uses proprietary shader subroutines".

What does that even mean?

NVIDIA always wrote motivational emails better, not sure why you guys are surprised. Ask yourself if it is AMD's fault for not scratching the right backs, or NVIDIA's for doing it.

Sometimes, the line between back-scratching and corruption is pretty fine.

 

[I.S.T.]

Seems to work for Nvidia



http://techreport.com/review/23419/nvidia-geforce-gtx-660-ti-graphics-card-reviewed/11

To be fair, they also omitted HAWK 2 back in the day due to how insanely biased towards Nvidia it was.

 

[AlexV]

Sometimes, the line between back-scratching and corruption is pretty fine.

We'd have to define corruption before making such claims. What would constitute corruption in context?

 

[Dave Baumann]

And it's not just techreport either - HardwareCanucks refuses to benchmark Showdown because - in SKYMTL's words - "it uses proprietary shader subroutines". Didn't stop him from benching Lost Planet 2 back in the day and I'll eat my hat if we don't see Borderlands 2 in their suite soon.

Linky please?

 

[Alexko]

Linky please?

Probably this: http://www.hardwarecanucks.com/forum/649231-post84.html

The guy seems to be dismissing any kind of fishiness with Crysis 2's use of tessellation too.

We'd have to define corruption before making such claims. What would constitute corruption in context?

The act of making someone do or accepting to do something ethically wrong in exchange for money or otherwise financially valuable favors.

 

[UniversalTruth]

The act of making someone do or accepting to do something ethically wrong in exchange for money or otherwise financially valuable favors.

Correct,

or can be back-scratching in... the bed... sleeping...

 

[jimbo75]

Linky please?

http://www.hardwarecanucks.com/foru...60-ti-review-comment-thread-9.html#post649231

edit - just noticed alexko already linked it.

 

[iMacmatician]

So I stumbled upon this TPU forum post which linked to a PCHome page (translated) containing some purported specs for some of the GTX 700 series.

GTX 770: GK110, 950 MHz core, 980 MHz boost, 2 GB 6.25 GHz GDDR5, 256-bit bus, 32 ROPs.

GTX 780: GK110, 3 GB 5.2 GHz GDDR5, 384-bit bus, 32 ROPs.

GTX 790: (Apparently 2x) GK110, 6 GB 4.06 GHz GDDR5, 2x 384-bit bus, 2x 32 ROPs.

Really weird specs and I'm not really believing them (only posting them for interest/discussion/speculation). Fairly recent though—the poster says they were updated 12 days ago.

Not far away on PCHome are purported specs of the GTX 655, the GTX 665 Ti, and the GTX 670 Ti.

GTX 655: GK106, 768 CCs, 2 GB GDDR5, 192-bit.

GTX 665 Ti: GK106, 3 GB GDDR5, 192-bit.

GTX 670 Ti: GK104, 1536 CCs, 900 MHz core, 2 GB 5 GHz GDDR5, 256-bit, 112 TMUs (how are 112 TMUs with 1536 CCs possible?), 6+6-pin.

 

[DuckThor Evil]

GTX 670 Ti into between 680 and 670 sounds quite weird when the performance difference is quite small between them and a GK106 based 665 Ti on top of GK104 based 660Ti sounds even more unlikely.

 

[AnarchX]

According to sources of German c't magazine GTX 650 Ti will be launched in the second week of October.
Then Nvidia will have a full 28nm line-up from GT 640 DDR3 up to GTX 690. OC version fill-up the gaps.

Realignments will probably happen with GK110 launches in early 2013.

Are refreshes of GK104/106/107 likely? Instead of GF10x these chips seem to reach the maximum clock rates of 28nm and also the IMC of GK104 seem to be able to reach 6Gbps+.
The chips just need higher production volumes.