Reviews - NVIDIA GeForce GTX 680 2GB graphics card

[HEXUS]

Kepler is here. Is Kepler King?

Read more.

[[GSV]Trig]

Whats this, top of the line NVidia card comes in at under 200w...

[watercooled]

but the base clock is higher than anticipated at a tasty 1,006MHz. We aren't allowed to talk about it, but the Boost Clock of 1,058MHz suggests that the card will ramp up the core operating frequency as and when needed - think of it as Intel's Turbo Boost, but on a GPU.

If you hadn't have said that, I wouldn't have thought it would be as simple as that, 52MHz (roughly 0.5% increase) isn't going to make an awful lot of difference.

But I guess we'll have to wait and see, because you're not allowed to say.

[kalniel]

It's hard work being an honest publication, but with NDAs being what they are, we aren't allowed to talk about a certain NVIDIA product that's said to be launching soon. Heck, we can't even tell you when the NDA lifts.

But you're confirming a rumour for us - thanks!

[shaithis]

Whats this, top of the line NVidia card comes in at under 200w...

The general consensus (which I am inclined to believe based on the moniker GK104) is that this is really the 660 or 670 card and the real high-end card(s) is yet to come.

I guess time will tell.

[CAT-THE-FIFTH]

The review of the GT640M confirms the hotclocks are still present.

[Hicks12]

Seems very pointless if the "turbo" style is really only 50mhz... but we shall see .

May i request that when you inevitably do this review, can you try and get alternative graphs, have some linegraphs with the barcharts just so we can see how stable FPS really is (think doing this for just the top cards is best as really it will be the 7950/70 that you will be comparing to, three cards).

Thats if its not to much hassle!. I look forward to reading this review when its out .

[crossy]

The general consensus (which I am inclined to believe based on the moniker GK104) is that this is really the 660 or 670 card and the real high-end card(s) is yet to come.

Be interested to see when Hexus do the benchmarking - especially if this is the 660 - as I'm thinking of treating myself to an upgrade for my current 460SOC later in the year. By which time, hopefully, they'll be widely available and at prices less than the usual ramped up ones for launch day.

Also wonder if XFX etc are going to do a special highly-overclocked version called a "666" or "616" (http://en.wikipedia.org/wiki/Number_of_the_Beast)...

[Hoonigan]

A memory speed of 6.0Gbps doesn't really seem that good. My GTX580 gets about 200Gbps iirc.

I really don't think I believe these slides, the increase in core count, core speed, decrease in TDP, it's all too much difference for just one generation of cards.

[Terbinator]

A memory speed of 6.0Gbps doesn't really seem that good. My GTX580 gets about 200Gbps iirc.

I really don't think I believe these slides, the increase in core count, core speed, decrease in TDP, it's all too much difference for just one generation of cards.

I think you're getting a bit confused. The memory speed of the GTX 580 is slower than the GTX 6xx (4002MHz vs 6000MHz). However, the GTX 6xx has a leaner bus at 256bit compared to 384bit, but faster RAM meaning that it has the same (just less according to the chart) bandwidth (~192GB/s for both cards) as the GTX 580.

[CAT-THE-FIFTH]

The slide seems to state memory speed in GBPS not MHZ it seems. GBPS is memory bandwidth not speed.

[Jimbo75]

Turbo on a graphics card? Give me a break.

Surely the whole point of turbo is to increase the speed on low threaded and short duration apps? This smells like a benchmarking trick to me.

[Terbinator]

The slide seems to state memory speed in GBPS not MHZ it seems. GBPS is memory bandwidth not speed.

I knew someone would throw that up.

The sums still equal the same though - 192GB/s.

Turbo on a graphics card? Give me a break.

Surely the whole point of turbo is to increase the speed on low threaded and short duration apps? This smells like a benchmarking trick to me.

Pfff, my motherboard runs at single-core Turbo speeds on all four cores.

[kalniel]

The general consensus (which I am inclined to believe based on the moniker GK104) is that this is really the 660 or 670 card and the real high-end card(s) is yet to come.

I guess time will tell.

Indeed.

One one hand it does look a little like they intended to release this as a performance (rather than enthusiast) chip, but on getting good results back from the fabs could up the target - supporting evidence is the moniker and the 256bit interface, as well as a moderate TDP.

But on the other that would be a change in tactics - usually nVidia release the high end card first I think - of course, rumour has it gk100 was cancelled completely, so perhaps there is no equivalent high end, and this would explain the delay compared to AMD (gk100 should have launched earlier). The new generation high end (gk110) will presumably be even better, and might even be moved forward given the cancellation of gk100. Question is, does it bear any relation to the performance segment chips, or like AMD, are the enthusiast chips so saddled with compute features that they only offer relatively minor gaming improvements for the increased complexity/cost?

[Hoonigan]

The slide states "Memory speed: 6.0 GBPS", which, when translated into GB/s is 48GB/s, which is still only a quarter of my GTX 580, or is there more calculations to be done?

I'm sorry, but NVIDIA would always state the highest number they can, they wouldn't put 6.0GBPS, they'd put 192GB/s, if that were the case.. So I really don't think I can believe the slide.

[kalniel]

The slide states "Memory speed: 6.0 GBPS", which, when translated into GB/s is 48GB/s, which is still only a quarter of my GTX 580, or is there more calculations to be done?

I'm sorry, but NVIDIA would always state the highest number they can, they wouldn't put 6.0GBPS, they'd put 192GB/s, if that were the case.. So I really don't think I can believe the slide.

Yup slide is a fake, but Hexus are confirming some of the numbers for us, which we can only assume is based on privileged information