Review Index:
Feedback

NVIDIA GeForce GTX 680 2GB Graphics Card Review - Kepler in Motion

Author:
Manufacturer: NVIDIA

The Kepler Architecture

Join us today at 12pm EST / 9am CST as PC Perspective hosts a Live Review on the new GeForce GTX 680 graphics card.  We will discuss the new GPU technology, important features like GPU Boost, talk about performance compared to AMD's lineup and we will also have NVIDIA's own Tom Petersen on hand to run some demos and answer questions from viewers.  You can find it all at http://pcper.com/live!!

NVIDIA fans have been eagerly waiting for the new Kepler architecture ever since CEO Jen-Hsun Huang first mentioned it in September 2010. In the interim, we have seen the birth of a complete lineup of AMD graphics cards based on its Southern Islands architecture including the Radeon HD 7970, HD 7950, HD 7800s and HD 7700s.  To the gamer looking for an upgrade it would appear that NVIDIA had fallen behind; but the company is hoping that today's release of the GeForce GTX 680 will put them back in the driver's seat.

This new $499 graphics card will directly compete against the Radeon HD 7970, and it brings quite a few "firsts" to NVIDIA's lineup.  This NVIDIA card is the first desktop 28nm GPU, the first to offer a clock speed over 1 GHz, the first to support triple-panel gaming on a single card, and the first to offer "boost" clocks that vary from game to game.  Interested yet?  Let's get to the good stuff.

The Kepler Architecture

In many ways, the new 28nm Kepler architecture is just an update to the Fermi design that was first introduced in the GF100 chip.  NVIDIA's Jonah Alben summed things up pretty nicely for us in a discussion stating that "there are lots of tiny things changing (in Kepler) rather than a few large things which makes it difficult to tell a story." 

View Full Size

GTX 680 Block Diagram

Continue reading our review of the new NVIDIA GeForce GTX 680 2GB Graphics Card!!

The chip that the GeForce GTX 680 is built on  GK104  is seen in its block diagram form above.  Already, you can see a big difference between this and the GTX 580 flagship card before it.  There are 1536 stream processors / CUDA cores on GTX 680 compared to the 512 cores found in GTX 580 cards.  The divisions of the GPU still exist in NVIDIA's design the GPC is a combination of SMs though they have changed as well.  A GPC now includes two SMX units (seen below) where the GTX 580 GPC included four SMs each. 

View Full Size

With the SM increasing from 32 cores to 192 cores each, NVIDIA is claiming a performance per watt metric improvement of 2x which is becoming a crucial factor as designers focus on the thermal limits and power consumption of GPUs.

View Full Size

Kepler SMX Block Diagram

The SMX unit consists of 192 CUDA cores, an updated PolyMorph Engine, 16 texture units, thread scheduling, among others.  Further, the cores are arranged differently than we saw in Fermi with six cores per special function unit (SFU) instead of four.  Warp (thread) count has gone from 48 to 64 in Kepler.

With the 128 total texture units on the GTX 680 (twice what we had on the GTX 580) and an increase in cores of nearly 3x, you might be wondering how it all balances out.  You may also be curious whether Kepler is really 3x as fast as Fermi.

View Full Size

Gone away is the "hot clock" of NVIDIA GPUs where the cores would operate at twice the clock rate of the base GPU.  Instead Kepler now runs the entire chip at the same clock rate.  The reasoning is a trade off in terms of die space and power consumption.  Engineers were able to reduce the clock power by half and logic power by 10% at the expense of some die area, but with a focus on power efficiency on this design it was a change they were obviously willing to make. 

View Full Size

Another change in Kepler is found in the scheduling component where much of the process is actually moved from hardware to software to be run in the NVIDIA driver.  Because the software is already handling so much of the decoding process from DirectX, CUDA, OpenCL, and more NVIDA found it to be more power efficient to continue to increase the workload in the software rather than on the chip itself.  Some items remain on die though because of latency concerns, such as texture operations. 

View Full Size

Because of a reduction in the number of SMX units per chip, NVIDIA had to double up on the performance of individual PolyMorph engines.  But because we have half the SMX units on Kepler as you did on Fermi, total chip performance hasn't changed much.

View Full Size

Compared to AMD's Radeon HD 7970 the GTX 680 is actually a bit slower at lower expansion factors and it's not until we hit 11x that we start to see the advantages NVIDIA once claimed to have throughout the scale.  Both companies debate which factors are most important though to game developers with AMD claiming that the lower factors are much more often used.

View Full Size

For the new memory design NVIDIA has gone with a 256-bit controller (compared to the 384-bit found on Fermi) though the clock speeds are running at 6 Gbps (1500 MHz)!  The total memory bandwidth provided by this design is 192 GB/s, which is basically identical to that of the GTX 580.  ROP count has decreased from 48 on the GTX 580 to 32 on Kepler/GTX 680, however.  

Today's GTX 680 will ship with a 2GB frame buffer and some users may lament of expectation for NVIDIA to match AMD's 3GB memory configuration on the HD 7900 cards.  While we are never one to say we don't want MORE memory on our GPUs, in our testing we have not seen detrimental effects of 2GB versus 3GB of memory even on multi-display gaming.

The GTX 680 is indeed a PCI Express 3.0 compatible card and the GPU does support DX11.1 features as well, but it isn't really anything to get excited about just yet.

View Full Size

One interesting change is the addition of NVENC, a dedicated video encoding engine that is built (essentially) to rival the QuickSync technology found in Intel's Sandy Bridge processors.  The logic is completely fixed function now it is no longer using the CUDA cores to encode video and NVIDIA claims that it is even more power efficient than Intel's implementation.  In fact, I was told by designers that the NVENC feature could actually be used while the GPU was powered off.

View Full Size

Another important change is found in the display support on Kepler as NVIDIA has finally moved away from the two display limit on single GPU cards.  You can now run up to four displays on a single card, and run three of them in an NVIDIA Surround or 3DVision Surround configuration for multi-display gaming.  This is obviously a feature that NVIDIA has needed for quite some time, and we are glad to see it in Kepler.  DisplayPort 1.2 support is included as well.

View Full Size

And here she is, the Kepler die in all her glory.  The 28nm GPU is built with 3.54 billion transistors and is 294mm^2. 

There is quite a bit more to Kepler and the GeForce GTX 680 though.

March 24, 2012 | 02:37 PM - Posted by Steven (not verified)

Here's a review of overclocked cards. GTX 680 comes out top.

http://www.overclockersclub.com/reviews/nvidia_gtx680/

March 23, 2012 | 07:50 AM - Posted by Ryan Shrout

1. That's good to know, but we aren't testing overclocked cards. We are testing reference cards in this review. We do review OC model cards (like the XFX model 7970 we tested) and will have a test of the Lightning next week. I think your estimate of 25% better improvement is an overestimate.

2. That's crap. Metro 2033 was an "NVIDIA game" and Dirt 3 was an "AMD game" yet those titles don't show any bias. Both vendors can write driver fixes to ANY game.

3. Fair enough - but we can only test based on the real user experience that you get when you install the cards and play the games. Saying that the HD 7970 is underclocked to me simply means that AMD didn't have the balls to release a card at the CORRECT speed then.

4. I think you misunderstand the feature then - I think it is really a compelling addition to get users the full capability of their GPU regardless of the game/application being played. When our video interview with Tom Petersen is online, you should take a listen?

5. I was able to hit a 150 MHz offset on the GTX 680 which is basically a 15% overclock.

6. Pricing is THE MAJOR factor other than performance. Yes, it is always changing, but to pretend it isn't a crucial factor is simply insane.

Thanks for you comments man, I love to hear what others are thinking, even if we disagree! :D

March 23, 2012 | 07:18 PM - Posted by nabokovfan87

1. It was just one of those things that glared out to me, and the 25% is probably 20% based on the HardOCP review, but that is a bunch more then 2%, which would make things extremely close in the end.

2. I don't know if AMD helped with metro, or if all amd cards run better on it, but the point is, Nvidia clearly had helped with physx and some filtering stuff in bf and others, it would be a good idea to add some variety to the benches, I would add R.U.S.E. and WoP for sure (they are different types of games then the ones you have listed, and as I said, they have some very nice high end dx10/11 features).

3. I completely agree, but it would have been nice as a reader to see some "by the way" type text.

4. I didn't say there wasn't a point to it, but I just don't get it, or better put, agree with it. Why with 1500+ cores you would want to OC them on the fly? I would want to be able to test things and make them stable. I see benchmarks vastly varying with games, it could be related to this "feature". Maybe heat of the case/card forces some bad benches for some sites and better scores for others? IDK, just something to look into. I know for power specifically people have "proof" of shenanigans, especially with 3dmark, so who really knows, but I know there is more then meets the eye with this.

5. But the performance was only 2% better, not much higher, The high binned 7970's go to 1250-1300+ which is around (35-40%, even at 1150 it is 25%)

6. It has more to do with the fact that we already know that amd is going to reduce their pricing, and it is going to be within a week more then likely. If they don't the review stands, but if they do, then it really comes off as overly harsh for no reason in the long run. Let me put it this way, have you ever gone back and revised this, for the people who look at your review AFTER the prices drop? Nope, because it ALWAYS happens and it always has to do with either a competitor releasing something, or a new card being out. In my opinion, it always seems like after a week the pricing on reviews is completely off.

Like I said, I didn't want to come off as a jerk or w/e, but it was stuff I wanted you to discuss or look into. The point with the OC stuff was to bring it up before you do SLI/OC testing, and I know Josh is writing the lightning review, but it would be a good idea to toss those numbers in the mix and get an idea of which one, OC'd or at it's best let's say, actually performs the best. It seems very straightforward to me that the 7970 OCs a lot better.

March 22, 2012 | 08:45 PM - Posted by Michael (not verified)

Looks like we have that annoying "coil whine" here. Ryan, did you experience any in this review? Peeps may want to watch this video review at LinusTechTips here @ 7:31:

http://www.youtube.com/watch?v=IaPYEOUH7LE

Note: Video also includes a cool cat

March 23, 2012 | 07:51 AM - Posted by Ryan Shrout

Haha.

Actually, I know of this phenomenon but I didn't experience with the GTX 680. I have with some HD 7000 cards though. In my opinion this is some kind of odd mix of PSU/GPU that causes it. I have never been able to pinpoint one architecture of PSU as the culprit.

March 23, 2012 | 06:05 AM - Posted by Dave (not verified)

Ryan, can you put that fancy SPL meter on this beast and provide some noise measurements?

I run a Define R3 chassis and use "reference" blower-style fans on my GPUs in order to keep the noise/heat down during idle periods and provide more effective heat dissipation under load without adding more than the 2 fans I currently have. It's loud under load, but I can deal with that since I game with a headset.

Would be interesting to see how this compares to, say, my current 570 blower. Especially considering every new card release comes along with a marketing line regarding lower heat and noise.

March 23, 2012 | 07:57 AM - Posted by Ryan Shrout

Sorry, we had these numbers the whole time but I forgot to post them!

http://www.pcper.com/reviews/Graphics-Cards/NVIDIA-GeForce-GTX-680-2GB-G...

Again, the GTX 680 impressed us.

March 23, 2012 | 07:50 PM - Posted by Hacksaw777 (not verified)

Wow just got the gtx680 installed and all I can say is Wow!!!!. I have bought many of top of the line video card over the years, and this is the first time I have ever been completely satisfied with one. This one is amazing. So powerful, quiet and only running 65c under load in Nvidia cooler master stacker case. Battlefield 3 run so smooth even running on 32inch 1080p ultra with vsync enabled never drops below 60 fps. Crysis 2 smooth as butter on extreme hi res textures. Now I think its time to look into getting into surround setup. if I could only get some good 27inch with small bezels for a good price.

March 31, 2012 | 04:12 AM - Posted by cyow

All I can say it WOW Sweat I Want one so sad it out of my price rang at this time.

Great Review many thnaks Ryan

April 2, 2012 | 04:05 AM - Posted by jewie27 (not verified)

The reason for frame rate capping is to lower the temps of the card. Which also reduces the fanspeed automatically and consuming less power overall.

It works, I tried it in Battlefield 3 with my two GTX 580's in SLI. Normally I get about 100 FPS on Ultra, gpu's at 82C max. With a FPS cap of 80 FPS, I drop temps down to 70's. 10 degree savings and quieter operation.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.