GTX680 pics

[Skurge]

200 3Dmarks in it. Looks like there are equal clock for clock. Which is pretty amazing from a technical standpoint. Looks like a 75mhz OC will bring a 7970 to parity. Although the GTX looks to have a better feature set. Not so excited about a 670 anymore.

 

[Tempered81]

200 3Dmarks in it. Looks like there are equal clock for clock. Which is pretty amazing from a technical standpoint. Looks like a 75mhz OC will bring a 7970 to parity. Although the GTX looks to have a better feature set. Not so excited about a 670 anymore.

I think a 7970 will need a 175mhz oc to around 1100mhz to match a 1006mhz stock 680 in most popular game titles across the board.

3dmark11 extreme is one of the most favorable 7900 benches around when you're comparing red to green. 11 and Crysis 1 and Metro. The 680 will likely kick butt in BF3 and Batman

 

[cusideabelincoln]

Power viruses not withstanding, do you have proof of a 480 drawing more than 250 watts continuous?

http://tpucdn.com/reviews/Gigabyte/HD_7770_OC/images/power_average.gif
http://www.techpowerup.com/reviews/Gigabyte/HD_7770_OC/24.html

This is in a real game with power being measured directly at the card level. It averages 257W.

GTX 480 draws about 40 more watts than the 7970, let's get back to reality.

Nope, the discrepancy is much bigger than 40 watts.

The reality of the situation is this is worse than that even... This is a GTX 480 that draws more power, and is slower than the 5870. Can you imagine that?

Because that is what the 7970 is shaping up to be compared to the 680, slower and draws more power.

Nope. The 480 was dogged on because of the size of the margin of discrepancy at the time: Huge power difference for relatively little performance boost. Even in current DX11 titles, in which there is still only a select few where the 480 does have a significant performance boost, it still draws a lot more power to basically bring its performance per watt up to par with the 5870. Oh and Nvidia charged a premium for the product at the time.

This isn't like the 480 situation at all. Some of the qualifiers might be the same, but the quantifiers are different and not as severe. In simpler words, these current high end cards are more equal according to these leaked benches. We still haven't been seen how the card performs in games where Nvidia hasn't already had the natural advantage.

 

[taltamir]

Power viruses not withstanding, do you have proof of a 480 drawing more than 250 watts continuous?

I really hate the term power virus.
Viruses are programs that self propagate, infecting machines. Power viruses overclock your hardware until it burns out.

Those sniveling PR guys are pushing hat label on genuine testing tools accusing them of unrealistic stressing of the hardware and use that to justify limiting power consumption ONLY on those... and that is absolutely BS.

If their claim that ONLY those testing tools cause the GPU to consume more power then the amount they set it to limit at, then they would have lost nothing by making the limit universal.

But the limit is only for select benching programs. Because they know of they made it universal many games will hit the limit. I have played many games that stressed the GPU further then my benchmarking and stress testing programs.

 

[BallaTheFeared]

The problem is no game will actually load up a gpu like furmark can, continuously. Some may spike past normal power draws from time to time and a limiter would cause problems in those games, however there is no game that will do what furmark and other tests like it do.

It's an unrealistic test and sets a bad precedence. For example a 480 will draw upward of 270 watts in furmark, wow that's scary as the 5870 will only draw 190ish. Man that 480 is an awful card! The flip side of this is the 480 only draws about 430 watts in Heaven, while the 5870 still draws about 190ish but the 480 can nearly double the performance of the 5870 with tess, so which one looks inefficient and awful now?

So it goes back to the problem of tests like Furmark creating unrealistic results without actually providing any meaningful data. The only thing Furmark type programs are good for is testing your cooling solution, any other information obtained through it is worthless.

 

[BallaTheFeared]

Yes thanks ballatheFeared for your lesson, however I know what a core and shader clock are. That is what a hotclock is. 1.1ghz core shows 2.2ghz shader in that program, because the program reads the video card's core speed from the I2C bus, and outputs twice that measurement into the "Shader" box because the uncore & core domains in Fermi are 1:2.

There is no speed measurement derived from the ALU domain, the program just says xxx mhz times two for Nvidia. amirite?

Are you saying there are no hot clocks with Kepler?

I'm actually confused as to what you're trying to convey here as the condescending tone was encrypting your actual meaning.

As far as how GPU-z works that's a question best left to asking techpowerup since they wrote it and I've never seen the source code.

 

[f1sherman]

He's saying don't look at GPU-Z for shader clocks. Hot-clock is there.

 

[BallaTheFeared]

He's saying don't look at GPU-Z for shader clocks. Hot-clock is there.

I'm confused then, because GPU-Z is showing a hot clocked shader speed.

Maybe I'll just leave this part of the conversation because I'm pretty much lost at this point /:|\


Is he saying he doesn't believe Kepler has a 1:2 hotclock, and that it's different? 1:1.5 or something? I dunno why I'm having such a hard time figuring out what he's actually saying.

 

[f1sherman]

Spoiler

1:2

scratch that - no hotclock

Spoiler

I think :hmm:

GPU-Z shader clock implementation is lacking - DONT LOOK AT GPU-Z 😀

 

[BallaTheFeared]

/🙁\Ughhh/🙁\

 

[notty22]

I'm confused then, because GPU-Z is showing a hot clocked shader speed.

Maybe I'll just leave this part of the conversation because I'm pretty much lost at this point /:|\


Is he saying he doesn't believe Kepler has a 1:2 hotclock, and that it's different? 1:1.5 or something? I dunno why I'm having such a hard time figuring out what he's actually saying.

Because only certain aspects of gpu-z actually do hardware polling.
An easy example is the die size measurement. It is just pulled from a database.
GPU-z may not know how to deal with Kepler currently and in fact can measure core speed from a clock sensor and then has been just doubling for the shader speed, because that was what was happening.
But may not be now.

 

[maddie]

I'm confused then, because GPU-Z is showing a hot clocked shader speed.

Maybe I'll just leave this part of the conversation because I'm pretty much lost at this point /:|\


Is he saying he doesn't believe Kepler has a 1:2 hotclock, and that it's different? 1:1.5 or something? I dunno why I'm having such a hard time figuring out what he's actually saying.

He's saying that GPU-Z calculates the hot clock by saying 2 x core clock. There is NO actual independent verification of shader clocks. The program simply assumes the ratio 2:1 for Nvidia just as it assumes a 1:1 ratio for AMD.

By the way, no one commented on the 706 MHZ default core clock for GTX680. What is happening there?

 

[f1sherman]

706 - Base clock for 3D, which shoots up to Boost clock (and beyond depending on TDP/app profile).

http://abload.de/browseGallery.php?gal=uXJ1d1Mb&img=1618490343259459683yku5t.jpg

PS

CharLIE caught... AGAIN 🙄
http://semiaccurate.com/forums/showpost.php?p=149618&postcount=249

 

[blackened23]

Someone at chiphell apparently has been testing the 680 :

Running points, the clock frequency is stronger than 7970 with 680 by 5%.
However, battlefield 3,680 is slower by 15%.
Or 1080, running 1600, 680 is estimated to lose by 20%.
That is, run 3d11, 680 super, games have to wait for the driver.
Also note that, Why run eight times aa. Not open aa leading 580 afraid not.

Whoa if thats true, wonder what will happen. D:

 

[f1sherman]

At chiphell?

Then why all google comes up is your post at [H]
http://hardforum.com/showpost.php?p=1038506419&postcount=253

Golum...Golum?

 

[blackened23]

Chiphell is not an english website. Translated, original is in cantonese. http://www.chiphell.com/thread-387232-1-1.html

Not sure if the translation is correct, but he does mention something about 3dmark11 being "super" and games needing a new driver. This line in particular:

That is, run 3d11, 680 super, games have to wait for the driver.

Translation could be wrong, I dunno, but his comment about games needing a new driver is curious. Does anyone here speak/read cantonese? HE also states that availability will be the 27th or 28th.

 

[f1sherman]

does this really sound like someone who has been testing GK104

Or 1080, running 1600, 680 is estimated to lose by 20%.
That is, run 3d11, 680 super, games have to wait for the driver.
Also note that, Why run eight times aa. Not open aa leading 580 afraid not.

driver stories are bull... both amd and NV have had chip in house long enough.

 

[taltamir]

The problem is no game will actually load up a gpu like furmark can, continuously. Some may spike past normal power draws from time to time and a limiter would cause problems in those games, however there is no game that will do what furmark and other tests like it do.

Furmark is above average but there are certainly games that will load a GPU more then it, continuously.

It's an unrealistic test and sets a bad precedence. For example a 480 will draw upward of 270 watts in furmark, wow that's scary as the 5870 will only draw 190ish. Man that 480 is an awful card! The flip side of this is the 480 only draws about 430 watts in Heaven, while the 5870 still draws about 190ish but the 480 can nearly double the performance of the 5870 with tess, so which one looks inefficient and awful now?

So it goes back to the problem of tests like Furmark creating unrealistic results without actually providing any meaningful data. The only thing Furmark type programs are good for is testing your cooling solution, any other information obtained through it is worthless.

Except it doesn't. It creates extremely realistic and accurate data and that is what scares them. This is why they limit just those programs instead of everything.

Also I think you meant to say "480 only draws about 130 watts in Heaven" otherwise the context doesn't make sense

 

[sontin]

Chiphell is not an english website. Translated, original is in cantonese. http://www.chiphell.com/thread-387232-1-1.html

Not sure if the translation is correct, but he does mention something about 3dmark11 being "super" and games needing a new driver. This line in particular:


Translation could be wrong, I dunno, but his comment about games needing a new driver is curious. Does anyone here speak/read cantonese? HE also states that availability will be the 27th or 28th.

And he said that the price will be lower than 7970. OBR said the same for the czech market.

It seems that nVidia blocked the driver so that there will be no real leaks until the end of the nda.

 

[BallaTheFeared]

Furmark is above average but there are certainly games that will load a GPU more then it, continuously.



Except it doesn't. It creates extremely realistic and accurate data and that is what scares them. This is why they limit just those programs instead of everything.

Also I think you meant to say "480 only draws about 130 watts in Heaven" otherwise the context doesn't make sense

lol ok then! I guess that's why AMD has had them blocked in their drivers for years now too! It all makes sense, Nvidia and AMD are conspiring against the greatest program ever!!!!!!!!!!!!!!!!!!oneoneoneone1

 

[Grooveriding]

32

lol ok then! I guess that's why AMD has had them blocked in their drivers for years now too! It all makes sense, Nvidia and AMD are conspiring against the greatest program ever!!!!!!!!!!!!!!!!!!oneoneoneone1

Pretty sure AMD does not throttle Furmark, at least in my experience. When the 480s were initially released, they didn't either.

When running Furmark to check stability my power supply's fan actually becomes audible, so yeah, I would agree it increases power consumption. It's no different than running Prime95 though, it should be able to handle the load. Would not surprise me if some people crash when running Furmark because it pushes their PSU beyond capability.

 

[maddie]

And he said that the price will be lower than 7970. OBR said the same for the czech market.

It seems that nVidia blocked the driver so that there will be no real leaks for the end of the nda.

I guess that no review site will have a valid review when the NDA lifts.

 

[sontin]

I guess that no review site will have a valid review when the NDA lifts.

They got the driver last week. 🙂

 

[blackened23]

And he said that the price will be lower than 7970. OBR said the same for the czech market.

It seems that nVidia blocked the driver so that there will be no real leaks until the end of the nda.

This can't be true, reviewers won't have enough leeway to get reviews up when the NDA is lifted.

 

[BallaTheFeared]

32

Pretty sure AMD does not throttle Furmark, at least in my experience. When the 480s were initially released, they didn't either.

When running Furmark to check stability my power supply's fan actually becomes audible, so yeah, I would agree it increases power consumption. It's no different than running Prime95 though, it should be able to handle the load. Would not surprise me if some people crash when running Furmark because it pushes their PSU beyond capability.

What experience is that exactly Grooveriding?

Are you saying your 480's now throttle in furmark?

Prime95 isn't that bad, an unfair comparison actually. Test with AVX Linpack, overclocked with your chip at 5GHz let's see how "realistic" that test is.