GTX680 pics

[BallaTheFeared]

wtf just happened?

GeForce GTX reached 680 in 3DMark 11 in the Extreme Preset 3165 points and is almost 20 percent faster than a Radeon HD 7970 (2669 points) and around 50 percent quicker than a GeForce GTX 580 (2077 points).

Gotta give them props if this is true, what an amazing little card.

 

[MrK6]

But what is the benefit? Shouldn't the more demanding games already tax the GPU and so max the TDP?

My guess would be to even-out disproportionate loads on the GPU. It looks like a more efficient rendering process within the confines of a TDP. For example, if a certain scene loads a particular portion of a GPU more heavily (and more uneven) than the rest, for instance a shader heavy scene, you could then use the power "saved" by the under-used portions of the GPU and dump that into higher clocks for the currently bottle-necking portion of the GPU (in this example, the shaders). Now I'm not sure which parts of the GPU can be overclocked or how flexible/dynamic this technology is, but it's definitely a step in the right direction. From that review and what Tempered81 is posting, it looks like it's at least the core that can be dynamically overclocked.

Otherwise, at first look, it seems AMD GPGPU toll is something like -30% performance in games.

How do you figure?

 

[Tuna-Fish]

TXAA sounds awesome then if it is indeed a real AA mode and not more post AA junk. Hoping it can be forced on in any game and work via the CP then. A game like BF3 could really benefit from something like that.

No real aa (except FSAA) can be forced on in every game. Simply because programmers are sneaky and with the modern flexible pipeline they can use it in some way it's not originally designed to be used, resulting in the card and driver having no clue where the object boundaries are in the final picture. This will probably work with BF3 simply because NV has had time to speak with it's devs, but there will be future games where it won't work.

I expect TXAA to be a combination of MSAA and post-processing. Essentially, the MSAA provides more depth values for the post-processing so that it can do a better job figuring out where it shouldn't blur things.

 

[Termie]

Almost 9200 at stock?

Someone refresh my memory, what does a 580 get?
I found this but I have to feed my son so I can't look into it right now.

I'd really like to see this card with more rops, more shaders and a bigger bus, power be dammed.

 

[Dribble]

Mmm, prices too high. The only bright spot is they are only too high because nvidia milking it. Card is pretty similar to a GTX 560Ti in die size, mem bus, etc which suggests they could charge about the same amount if they wanted too (other then the higher cost of 28nm which should drop).

Just need 28nm to mature so supply > demand then prices can fall.

 

[Tempered81]

Some power consumption. Very nice!

http://img819.imageshack.us/img819/9607/16235232521973668591.png


@groove, TXAA looks like Multisample + post. Seems to be Nvidia's version of SMAA 1 and 2

 

[OILFIELDTRASH]

My guess would be to even-out disproportionate loads on the GPU. It looks like a more efficient rendering process within the confines of a TDP. For example, if a certain scene loads a particular portion of a GPU more heavily (and more uneven) than the rest, for instance a shader heavy scene, you could then use the power "saved" by the under-used portions of the GPU and dump that into higher clocks for the currently bottle-necking portion of the GPU (in this example, the shaders). Now I'm not sure which parts of the GPU can be overclocked or how flexible/dynamic this technology is, but it's definitely a step in the right direction. From that review and what Tempered81 is posting, it looks like it's at least the core that can be dynamically overclocked.

How do you figure?

Wow. If that technology works the way you describe it then all AMD can do is tip their hat to Nvidia and admit defeat. That's a game changer. Nvidia is the intel of gpu's.

 

[GaiaHunter]

How do you figure?

Die sizes and performance expectations when compared to performance and die sizes difference between GTX460/GTX560Ti and 5870/6970.

Only definitive and clearer with reviews, of course.

Will be interesting to see what kind of GPGPU performance the GTX680 will be able to pull (not that I'm currently interested in any GPGPU application).

 

[tviceman]

Some power consumption. Very nice!

http://img819.imageshack.us/img819/9607/16235232521973668591.png


@groove, TXAA looks like Multisample + post. Seems to be Nvidia's version of SMAA 1 and 2

Wow huge turn around for Nvidia since... well, since they started making compute-focused GPU's.

 

[sontin]

Wow huge turn around for Nvidia since... well, since they started making compute-focused GPU's.

Fermi was the first step. It was the architecture. Kepler is Fermi on drugs with much better Perf/Watt. It's like G71. Which was nearly only a die shrink of g70. But this time nVidia will not have the worse architecture.

 

[Termie]

If all of this is true (and we don't know for sure until next week), this is an absolute revelation.

nVidia will have just beaten AMD at its own game, namely performance per die size and performance per watt. If nVidia wanted to, it could shut down all of AMD's sales over $300, but it won't...

 

[Tempered81]

Major change in the shader, good for them. It's a 28nm cayman with 2x warp. Smart design. Really happy to see nvidia beating amd in the design field.

AMD has nothing now, unless oc'd 7970's trounce oc'd 680s. And the 3gb vs 2gb frame buffer argument, until 4gb hits.

 

[Tempered81]

Actually, it kinda sucks cuz they have no reason to launch 110 now. ugh

 

[MrK6]

Die sizes and performance expectations when compared to performance and die sizes difference between GTX460/GTX560Ti and 5870/6970.

Only definitive and clearer with reviews, of course.

Will be interesting to see what kind of GPGPU performance the GTX680 will be able to pull (not that I'm currently interested in any GPGPU application).

Bolded the agreement, as there's still too much in the air. For example, die size numbers have ranged from 360mm^2 all the way down to 294mm^2, however these pictures show definitely that the die measures 18.55mm x 17.18mm, or 318.689mm^2. I remember 320mm^2 being thrown around previously, so one of the leaks was correct. Tahiti was 365mm^2, so 14.5% larger. If the GTX 680 and the 7970 end up trading blows (which seems likely), that's not a bad disadvantage for making a competitive GPGPU chip, IMO.

 

[Tuna-Fish]

nVidia will have just beaten AMD at its own game, namely performance per die size and performance per watt. If nVidia wanted to, it could shut down all of AMD's sales over $300, but it won't...

It's unlikely the 28nm process has that kind of throughput yet. NV could price their GPUs at 300, but they'd have to sell empty shelf space.

But looking forward to summer and autumn, it does indeed look grim to AMD.

 

[BallaTheFeared]

Actually, it kinda sucks cuz they have no reason to launch 110 now. ugh

Maybe for the return of the $800 Ultra? D:

 

[tviceman]

Bolded the agreement, as there's still too much in the air. For example, die size numbers have ranged from 360mm^2 all the way down to 294mm^2, however these pictures show definitely that the die measures 18.55mm x 17.18mm, or 318.689mm^2. I remember 320mm^2 being thrown around previously, so one of the leaks was correct. Tahiti was 365mm^2, so 14.5% larger. If the GTX 680 and the 7970 end up trading blows (which seems likely), that's not a bad disadvantage for making a competitive GPGPU chip, IMO.

I asked about this and these were the answers I got:

http://forum.beyond3d.com/showpost.php?p=1628882&postcount=2669
http://forum.beyond3d.com/showpost.php?p=1628903&postcount=2685

 

[Grooveriding]

Some power consumption. Very nice!

http://img819.imageshack.us/img819/9607/16235232521973668591.png


@groove, TXAA looks like Multisample + post. Seems to be Nvidia's version of SMAA 1 and 2

That was my first guess when I saw it. It sounded too good to be true to deliver real AA for less performance :sneaky:

It is most likely their own copy of the latest SMAA 4x, which uses 2xMSAA 1xTRSAA and POST AA in combination. Makes me wonder if it will be possible to force it in games. You cannot force the newest SMAA 4x in games using the injector because of the way it uses some MSAA/TRSAA. From my understanding of what I've read the game has to be coded to allow for its use.

Meh, more post AA. The developers of SMAA claim the newest 4x has significantly less blur, but since you can't inject it and use it in games, there is no way to experience it first hand.

 

[Tempered81]

Gonna have to get you some 680 sli on wednesday!

 

[Ajay]

Some power consumption. Very nice!

http://img819.imageshack.us/img819/9607/16235232521973668591.png


@groove, TXAA looks like Multisample + post. Seems to be Nvidia's version of SMAA 1 and 2

So, how can it have just two 6 pin PCIe power connectors and hit >350 Watts?

 

[BallaTheFeared]

So, how can it have just two 6 pin PCIe power connectors and hit >350 Watts?

Because there is a cpu and other parts to power as well?

Any word on what Nvidia has done to reduce cpu speed dependency?

 

[tviceman]

So, how can it have just two 6 pin PCIe power connectors and hit >350 Watts?

That graph is entire system power consumption

 

[Grooveriding]

Gonna have to get you some 680 sli on wednesday!

Even if these cards are 10% across the board faster than a 7970, if they manage to do that with this die size, would make me loathe to buy them with the GK110 coming. ()

When anand has their review up I am hoping there will be some mention of future products. I don't mind to wait until the fall for a 500mm2 die x2 setup with this sort of perf/mm2 While I doubt it will have the same perf/mm2 this chip does, I would like to assume some of the improvement will be there.

 

[railven]

Are we getting the GTX 680 on Wednesday or just a paper launch?

So no worder on the 670?

Eitherway, glad we're finally getting something!

 

[footballrunner800]

If this card overclocked ends up faster than my 7970 overclocked, I will definitely be going green. As somebody mentioned in an earlier post, Nvidia seems to be taking all of AMDs features and improving on them.