GK104 = GTX 680 (confirmed), paper launching soon [SA]

Man if I was GoldenTiger I wouldn't verify crap for anybody. He doesn't owe any of you anything. He stated his performance and if I was him I wouldn't care if people believed me or not.

That's about all it boils down to, plus why bother when it is meaningless anyway as Zarathustra demonstrated :) :


Zarathustra[H];1038472648 said:
I'm 1337:er than all of you.

Got my GTX680 almost two months ago from a super secret Newegg sale :p

6965436241_206d99b3d3_b.jpg
 
I just hope this is a tight race, as mentioned before. Excited to see if the extra CUDA cores will help with distribute computing that I enjoy so much. This coming from someone who currently has 2 7970's in Xfire running Eyefinity and 3 x 580's in SLI running Surround. All cards are under H20.

I am just now finding time to test against each other and can't wait to pick up some 680s.

Edit: So you don't think I am making things up;) 7970 rig: http://www.evga.com/forums/tm.aspx?m=1422017 Mods rig shows my other setup.

I really do think that Nvidia changed their naming scheme for this generation after the 7970 came in low (to their expectations). Maybe they took what WAS going to be the 660TI and renamed it the 680 since it performed more in line with AMD's 7970? That's just what I think:)
 
As I am a forum noob, despite being a member for so long. What, may I ask, is a paper launch?
 
As I am a forum noob, despite being a member for so long. What, may I ask, is a paper launch?
it means the product is not actually available for purchase when it is reviewed.

you could have have just googled it...
 
As I am a forum noob, despite being a member for so long. What, may I ask, is a paper launch?

paper launch just means they remove the NDA and launch the card to the media and retailers to build up hype but the cards won't actually release for another week or two. a good recent example was AMD removing the NDA for the 7970 in December but not actually releasing the card til the first week in January.
 
We now return you to your regularly scheduled flame fest ...carry on....and at least try and talk about video card related stuff...:)
 
Its funny how NV says the 7970 is low to their expectations yet it sounds like they can't even live up to their own expectation.

I get to try out the 680 paper launch on a blind test against the 7970. I have to wear loud earphones so I can't tell by the sound of the cards. Single screen I presume unless we get lucky and get 2 for 3 screen.
 
Its funny how NV says the 7970 is low to their expectations yet it sounds like they can't even live up to their own expectation.

I get to try out the 680 paper launch on a blind test against the 7970. I have to wear loud earphones so I can't tell by the sound of the cards. Single screen I presume unless we get lucky and get 2 for 3 screen.

That is pretty hilarious
 
I have to wear loud earphones so I can't tell by the sound of the cards.

Yes, you wouldn't want the loud AMD cards to give themselves away. Or are you forgetting that all the Nvidia 5xx cards are quieter than their AMD counterparts?
 
my 7970 is whisper quiet at stock. If and I said if the 680 is like the 480 then yes it will be easy to tell the difference
 
my 7970 is whisper quiet at stock. If and I said if the 680 is like the 480 then yes it will be easy to tell the difference

All the tests show the 7970 being louder than the GTX 580 also, so Nvidia certainly has the ability to make a powerful, yet quiet, card.

You seem to be the only person who thinks GK104 will be hot and/or loud, but I guess that's to be expected given your post history.
 
All the tests show the 7970 being louder than the GTX 580 also.

You seem to be the only person who thinks GK104 will be hot and/or loud, but I guess that's to be expected given your post history.

in my opinion tests don't really tell you anything. great you have a number, big deal. not every persons hearing is the same, not every ones case or hell system location is the same. so its relative to the person which a test really can't tell you anything. for example people cried and bitched about how loud the 8800GT's were back in the day, both my overclocked 8800GS's had the same heatsink and at full load i still couldn't hear either of them even with both fans at 80%.
 
my 7970 is whisper quiet at stock. If and I said if the 680 is like the 480 then yes it will be easy to tell the difference
in the Anandtech review the 7970 was the loudest single gpu card in the test. yes the old 480 and 470 were loud but the 580 and 570 are pretty quiet so hopefully Nvidia learned their lesson.
 
in the Anandtech review the 7970 was the loudest single gpu card in the test. yes the old 480 and 470 were loud but the 580 and 570 are pretty quiet so hopefully Nvidia learned their lesson.

7970 have a odd problem, some card runs hotter and some card don't..
The thermal paste on some of them are completely off...

Gladly, the 7970 I had is quiet as my Arctic Xtreme Plus.... It's always @ 35-40% on heavy load in OC..
 
7970 have a odd problem, some card runs hotter and some card don't..
The thermal paste on some of them are completely off...

Gladly, the 7970 I had is quiet as my Arctic Xtreme Plus.... It's always @ 35-40% on heavy load in OC..
yeah now I do remember hearing about thermal past issues on some of the review cards.
 
Mine is real quiet and cool even at my high oc. I keep meaning grab the arctic freezer but I use earphones and never hear the thing.
 
I hope the 680 has 50 blinking leds and ribbon tasles on the exhaust to make it all pretty
 
in my opinion tests don't really tell you anything. great you have a number, big deal. not every persons hearing is the same, not every ones case or hell system location is the same. so its relative to the person which a test really can't tell you anything. for example people cried and bitched about how loud the 8800GT's were back in the day, both my overclocked 8800GS's had the same heatsink and at full load i still couldn't hear either of them even with both fans at 80%.

They tell you an empirical comparison between the cards, which is all any testing can do. You are right that hearing sensitivities vary (as well as sensitivity/tolerance to different frequencies), but at least it gives you some idea of the relative noise of the card. Some people are more sensitive to FPS, but that doesn't mean benchmarking doesn't tell you anything.

I do agree that the sound tests aren't as useful as they used to be, given how close together and quiet most of the new cards are.
 
Last edited:
Zarathustra[H];1038472538 said:
This would make GK104 ~170% faster than it's predecessor, the GF104 at launch.

I can't help but think that this is highly unlikely.


Do you mean 70% faster?

170% faster is 2.7X as fast.
 
Just looking at architectural design, the consensus on Fermi as I recall was that it was an incredibly powerful design, but targeted mostly at computing applications with gaming as a sideline. It had enough capacity to give good gaming performance by brute force, but it wasn't optimized for it.

If Kepler is a return to gaming-centric optimization in the design, that could explain the seemingly unbelievable performance gains. Sort of a Fermi-meets-debut-of-8800GTX moment. What can I say, I'm a dreamer...
 
They tell you an empirical comparison between the cards, which is all any testing can do. You are right that hearing sensitivities vary (as well as sensitivity/tolerance to different frequencies), but at least it gives you some idea of the relative noise of the card. Some people are more sensitive to FPS, but that doesn't mean benchmarking doesn't tell you anything.

I do agree that the sound tests aren't as useful as they used to be, given how close together and quiet most of the new cards are.

More like 'semi'-empirical, because not every review site tests in the EXACT same way; some test at 2", some a 4", and others at 1'-3' (normal range), some test on an open bench, while others test in a case (which itself can compound testing).
 
More like 'semi'-empirical, because not every review site tests in the EXACT same way; some test at 2", some a 4", and others at 1'-3' (normal range), some test on an open bench, while others test in a case (which itself can compound testing).

Within the same test at least you can get a decent idea... by the deltas. Compare it to other review deltas and you get an even better idea.
 
Except that testing on an open bench might not have any correlation to testing in a case.
At least an open bench gives an unfair advantages to internal exhaust coolers.
 
Last edited:
I assume by GF104 you meant GTX 560? Maybe this card is better compared to the GTX 560 Ti 448.

I was using the GTX460 as it represented what the GF104 was capable of at launch.

I feel it is most appropriate to compare early silicon to early silicon, as they are similarly mature processes.

But it being the dual card would imply that GK104 is 85% of the performance of the GF104, using your numbers. Which seems even more unlikely.

I don't see where your math comes from.

Two GTX460's in SLI roughly tie a GTX580 give or take a tiny bit.

So if dual GK104's are 45-50% faster than this, then it would follow that a single GK104 at launch would be 45-50% faster than a single GF104 at launch, which even that seems a little bit optimistic, but might be true.

I'm still betting that this product everyone is talking about is a dual mid-range GPU product based on the new GK104 chip tying or just barely beating the 7970 when both are overclocked.

If I had to chose here, I'd chose a single GPU product every time.


Also, keep in mind when I said that it is unlikely that the new mid range chip is 170% faster than the last mid range chip, that's 170% faster, not 170% of the performance, so this corresponds to 270% of the performance. This really is unlikely, so something is really wrong here. either the rumors are false, or this new product is a dual GK104 board.
 
I don't see why you would think it had to be the dual-chip card to get 7970 levels of performance. After all, AMD got 7970 level performance from the 7970, and this chip appears to be about the same size. Makes sense that performance could be about the same.
 
I don't see why you would think it had to be the dual-chip card to get 7970 levels of performance. After all, AMD got 7970 level performance from the 7970, and this chip appears to be about the same size. Makes sense that performance could be about the same.

But you seem to forget that the GK104 was designed as a mid range chip to replace the GF104/GF114, 460/560/560TI, not as a top end chip to replace the 480/580 based on the GF100/GF110.

Some sort of problems with the GK100 must have made them try to use mid range chip as a stop gap until GK100 is ready. This can not be without its consequences.

Either it's going to be a lot slower than the 7970, or it's going to have to be a dual GPU product. I don't see how it could be any other way.
 
I was going to say... heh. 70% faster = 1.7x as fast. 170% faster, as you said, is 2.7x as fast.

This is exactly what I meant.

If a single GK104 is 45-50% faster than a GTX580 as stated in this thread (which I highly doubt) then it is 170% faster, 2.7x faster than the GF104 (GTX460) was on launch.

This just doesn't sound likely to me, which leads me to believe we will be seeing a dual GPU board or something a lot slower.
 
Zarathustra[H];1038475671 said:
This is exactly what I meant.

If a single GK104 is 45-50% faster than a GTX580 as stated in this thread (which I highly doubt) then it is 170% faster, 2.7x faster than the GF104 (GTX460) was on launch.

This just doesn't sound likely to me, which leads me to believe we will be seeing a dual GPU board or something a lot slower.

AMD nearly pulled it off with a chip around this size, and it's been TWO new chips once Kepler hits since GF104 (GF104--->GF114---->GK104), so I don't think it's that impossible.
 
Zarathustra[H];1038475663 said:
But you seem to forget that the GK104 was designed as a mid range chip to replace the GF104/GF114, 460/560/560TI, not as a top end chip to replace the 480/580 based on the GF100/GF110.

Some sort of problems with the GK100 must have made them try to use mid range chip as a stop gap until GK100 is ready. This can not be without its consequences.

Either it's going to be a lot slower than the 7970, or it's going to have to be a dual GPU product. I don't see how it could be any other way.

It certainly appears that the card everyone is expecting to be announced soon is the single GPU card - the PCB pics alone are enough to show that. Likewise, all the rumors point to that card being called the GTX 680 because it performed well enough that Nvidia felt comfortable targeting it towards the 7950/7970. Couple that with the veiled comments coming out of Nvidia for the past few months (their surprise at the lack of 7970 performance, for example) and you start to get the impression that this re-design is performing a little better than everyone expected. It's not unheard of for a new design on a new process to provide pretty astonishing performance gains (think 9700 Pro), so I don't think it's impossible that they have figured out a way to make the mid-range part compete with AMD's high-end. The GTX 560 Ti and the GTX 560 Ti 448 competed pretty well with the 6950/6970, and let's face it, at stock clocks the 7970 isn't really all that much of a leap in performance. The GTX 560 Ti 448 is, what, only 10-15% slower than the GTX 580, so a 50% performance bump on that card puts it in 7970 territory. Heck, even a 50% bump on the 560 Ti puts it in the 7970 ballpark - are you saying Nvidia couldn't have pulled off a 50% improvement over the 560 Ti?

I'm not saying we aren't in for another GTX 480 disappointment, but even the hints Kyle has dropped make it sound like a pretty good card.

Edit: Even a dual chip GTX 560 Ti card would out-perform the 7970, so why would it take two next-gen chips to perform the same feat?
 
Last edited:
Back
Top