Kepler Nvidia GeForce GTX 780

[kromosto]

something is not right about the scores imo can be premature driver. it is a good card thats for sure compute power may not be great but enough. so as it is nvidia buying turn for me i may be my next card but i am not satisfied with the price if rumors are true.

[Tha Last Meal]

Sorry AMD.........But this is what happened

[Kristoferr]

http://www.theinquirer.net/inquirer/...70-gpu-compute

[GoldenTiger]

Sorry AMD.........But this is what happened

LOL, love it .

[RealTelstar]

*subscribed*

[Tim]

http://www.theinquirer.net/inquirer/...70-gpu-compute

Can't say that I personally care. I know some people buy these cards for crunching but lets face it, it's all about the games for the majority of people.

[tajoh111]

The type of performance they would need to be competitive I would think it basically a 7870 but double everything and possibly double the size and even then it might not be guaranteed. Such a huge die I could imagine would consume possibly more than twice as much energy as well, with the power leakage and all. The problem with this idea, is also AMD would be making something 400nm2+, with its GPU compute features thrown out and it would be a pure gaming card. That a really big chip to sell for purely gamers. And with the R and D and small volume of enthusiasts, it might not be worth it for AMD. AMD has avoided the monolith chips for the longest time and I think they would be more content with being second place in the single chip arena. They could have captured the single chip performance in the past I think, but they didn't(they sort of did with the 5870 in somewhat recent times, but this was because the competition was late trying to make the big chip. I think AMD would rather avoid this).

[BeepBeep2]

Looks like I need to take some VerdeTrol.

[[XC] Oj101]

I don't think this will overclock far at all using sub zero cooling, only four phase VRM :/

[865pe]

http://www.techpowerup.com/162765/GK...roximated.html

[Rollo]

Umm, because the 680 has 2GB of VRAM opposed to the 7970's 3GB...

It's fairly amazing to me how last year (3 months ago) 2GB of 256 bit access VRAM was "the thing to have" and "what you need for Eyefinity and 25 X16", and the 3GB 580s were "overkill".

Case in point, HardOCPs review of the 3GB 580s:

http://hardocp.com/article/2011/06/2...card_review/11

While this was a big improvement over standard GTX 580's, it is less of an improvement over two 6970 video cards, which already have a good amount of RAM on them. For the most part, these two setups will perform very similar, except in some specific cases depending on the game.

And he goes on to say in FEAR 3 the 3GB helped at the setting run.

This "gotta have 3GB VRAM" stuff is out of control, and very inaccurate.

I have gamed a TON on 57X10 and 25X16 with 1.5GB/2GB/3GB NVIDIA cards and can tell you 1.5GB is enough for 25X16, and most of the time for 57X10 4X16X. I don't think I've ever run out of VRAM running 57X10 with 2GB, but I've never played SkyRim with the texture pack and 8X MSAA, nor have I played BF3 on Ultra settings. (don't have either game)

My thought is 95%+ of the people saying you need 3GB for 25X16 or 57X10 don't have either, and a review of how many people actually have those resolutions on the STEAM survey would back that up. I'll post some 57X10 benches of the 680s after they go one sale and hopefully put this to bed for good.

People need to remember you can exceed the VRAM on any graphics solution with things like texture packs on the handful of games that have them, or super high AA settings, so proclamations like "I want to play Skyrim at 57X10 8XMSAA with the texture pack settings maxed!" need to be tempered with "and I'm willing to have use a slower, louder card that doesn't have PhysX, good 3d support, CUDA apps, forced AO and needs more power to do so".

A graphics card investment is more than the ability to play one game at one setting, unless that is all you plan to do with it.

[Leeghoofd]

I don't think this will overclock far at all using sub zero cooling, only four phase VRM :/

There are ways to circumvent that

[Vardant]

If these benchmarks hold true, GTX 680 is going to be one heck of a nice card.

Oh common...

All THG links removed by request of THG, so please no more THG links.

[DilTech]

Already taken care of Vardant.

Guys...NO LINKS TO THE LEAKED NUMBERS FROM THG.

[Piotrsama]

Sorry AMD.........But this is what happened

Would be funny if AMD surprise-launched their 7990 the same day

[SubZero.it]

You got me.....something isn't quite right there...

Mmm

[Andrew LB]

Already taken care of Vardant.

Guys...NO LINKS TO THE LEAKED NUMBERS FROM THG.

Oops. I didn't see this when it was posted earlier in the thread. My apologies.

[DilTech]

Oops. I didn't see this when it was posted earlier in the thread. My apologies.

Mistakes happen, don't fret.

[thematrixhazune]

I am just assuming here... but all this talk about gk110 being more efficient than gcn, depending on benches gk110 is clocked 14% higher then 7970. So all this "AMD is in trouble stuff" is really irrelevant. Unless however, 680 totally surpasses 7970. All I want is one card to be a little faster so the price goes down on the other!

[Leeghoofd]

All I want is one card to be a little faster so the price goes down on the other!

Indeed competition is good for pricing !! And any card you choose, does the job perfectly. We are so spoiled with the latest generation in GPU's

[Borden]

ok, now AMD can release the "PE" BIOS update for all the 7970...

[SKYMTL]

http://www.theinquirer.net/inquirer/...70-gpu-compute

After going through the synthetic GPGPU programs over the last week, I've come to the conclusion that they aren't worth squat.

If they don't support a particular architecture (case and point: Kepler) or use OpenCL / Directcompute extensions that aren't yet supported at the driver level (ie: Sandra, ComputeMark, etc), then there will be absolutely horrible performance.

Not only that but certain architectures are specifically designed for certain tasks. For example, Tahiti is geared towards Double Precision tasks in a big way but will that superiority actually translate into faster compute performance in actual applications? Only if the applications in question are actually optimized for AMD's implementation of OpenCL.

[spicypixel]

Agreed, maybe for CUDA you could use 3dsmax and glu3d on a standard candle scene and just bench simulation times. CUDA only though.

[Gaul]

ok, now AMD can release the "PE" BIOS update for all the 7970...

what is PE BIOS ?

[GoldenTiger]

what is PE BIOS ?

It's a joke about the old Phantom Edition... errr ahem, I mean Radeon Platinum Edition cards back in the day.