Sign in with
Sign up | Sign in
Your question

Do Two GeForce GTX 680s Beat Three GeForce GTX 660 Tis In SLI?

Last response: in Reviews comments
Share
February 18, 2013 3:00:04 AM

What a problem, right? You have the money for high-end graphics, but don't know whether to buy a couple of GeForce GTX 680s or a trio of GeForce GTX 660 Ti boards for three-way SLI. We run through a number of games and help you come to a conclusion.

Do Two GeForce GTX 680s Beat Three GeForce GTX 660 Tis In SLI? : Read more
February 18, 2013 3:20:07 AM

off topic:

when titan's review comes out?

on topic:

frankly i'm no multi gpu nor multi monitor user. but for the green camp if you want multi monitor gtx 660 ti 3Gb is a good choice if you want the cheap option. if you want more horse power while still on the cheap side then 7950 is the answer. (the latest game bundle from amd might also lean some people to buy amd cards)


Related resources
February 18, 2013 3:35:20 AM

From my own experience I would rather go with 2x 680 as dual sli scales very well on about _all_ titles while tri/quad sli can be more of a gambling in performance - For titles where all the cards scale well its sure nice as hell - Where it does not you can even loose performance by adding a third/fourth card (even thoo that scenario gets less and less frequent).
February 18, 2013 3:37:57 AM

id like to see some benchies on metro 2033 with physx and DoF just for fun hahaha
February 18, 2013 3:41:02 AM

Nice review, makes me consider taking 3x7950 instead of 2x7970. Would equal in price due to sales that are going on. But, maybe grab a Titan? I hope to see another review in 8 hrs featuring Nvidia GeForce Titan.
February 18, 2013 3:52:22 AM

Not the point of this article, but i noticed that a single gtx680 is a complete beast at 1080P.
Also, it would have been interesting to see the CPU usage during the tests.

Edit : A repeat of this article with the HD7000 cards should be performed, specially after the 13.2 driver are released.
February 18, 2013 3:57:23 AM

mayankleoboy1Not the point of this article, but i noticed that a single gtx680 is a complete beast at 1080P.Also, it would have been interesting to see the CPU usage during the tests.Edit : A repeat of this article with the HD7000 cards should be performed, specially after the 13.2 driver are released.


Yes same review for HD 7000 series should be given. Can't wait for that.
February 18, 2013 4:00:30 AM

I'd still go for a GTX 690 Hydro Copper. Overclock it and pwn while only taking two slots. SLI configs which take up all your slots seem highly inefficient and unappealing.
February 18, 2013 4:02:57 AM

when are you going to switch to the new 3dmark?
February 18, 2013 4:20:33 AM

Why do you say "higher/lower is better" in some graphs? Which is it - higher or lower?
February 18, 2013 6:21:40 AM

Interesting article, however, it'd be very nice to see a comparable Radeon article.
February 18, 2013 7:52:42 AM

What about 4 X crossfire 7870LE VS 3 X crossfire 7950 VS 2 X crossfire 7970 VS Devil 13 7990
February 18, 2013 9:24:31 AM

A review with SLI and no microstutter and frame latency tests?? Seriously?
February 18, 2013 9:34:34 AM

670s are still a better value, you've shown that numerous times.
February 18, 2013 9:37:20 AM

PadaV4A review with SLI and no microstutter and frame latency tests?? Seriously?


Frame latency has only been a major issue for AMD setups, and mostly around the 7950, not so much the higher end cards.
February 18, 2013 10:25:12 AM

TheMadFapperFrame latency has only been a major issue for AMD setups, and mostly around the 7950, not so much the higher end cards.


Actually no, all cards have a problem with inconsistent frame latencies, including Nvidia cards.
The problem is even bigger on multicard setup, where microstuttering can be a visual annoyance
February 18, 2013 11:25:02 AM

kristi_metalActually no, all cards have a problem with inconsistent frame latencies, including Nvidia cards. The problem is even bigger on multicard setup, where microstuttering can be a visual annoyance


Actually no, AMD cards till now recently have had poorer frame latencies that the comparable Nvidia cards, mainly because of poor drivers. It does not mean Tahiti < Kepler. It only means AMD driver team is lagging.
February 18, 2013 11:28:27 AM

Could see the conclusion coming, 2x 4GB GTX680s seem to make more sense than the Ti's.

Just like the other folks, would like to see the Radeon equivalent article too! And Titan as well :D 
February 18, 2013 11:36:21 AM

kristi_metalActually no, all cards have a problem with inconsistent frame latencies, including Nvidia cards. The problem is even bigger on multicard setup, where microstuttering can be a visual annoyance

I don't experience microstutter with my two 680s in SLI. I had it constantly with my 2 5850s and still with 3 5850s. I never had a problem with microstutter with 2 580s in SLI either.

I challenge everyone to try both multi-card technologies for themselves and then talk about their results. Don't just take the word of others as the truth.
February 18, 2013 12:23:29 PM

Somewhat happy to see the results, for reason that 660 Ti didn't seem to be bottlenecked by it's 2 GB VRAM (except perhaps in BF 3). Guess the guys who told that unless you plan to go 3-way SLI your cards don't have enough raw power to use 4 GB VRAM anyway were right and back when I bought my 2 670 cards it would have cost me 150$ extra for a bit weaker cards. So I have no regrets. PS. Gaming at 5760x1080 is awesome but even more so is video editing.
February 18, 2013 12:47:52 PM

Or just grab used GTX 690 for about £530-600 from ebay UK (duno about US price, sorry), for 1080p gaming it is just enough to ultra everything and even on 1440p you'll be just fine, games will drop under 60FPS in some games but still the most power and price efficient card (when buying used one).
February 18, 2013 1:20:06 PM

I always go for the 2nd best card of the day when first released, then SLI or Xfire a year later when the price has dropped significantly. Then you can skip the next gen and repeat again after.
February 18, 2013 2:24:04 PM

uhm... biostar mobo? i'm genuinely surprised
February 18, 2013 5:49:54 PM

PadaV4A review with SLI and no microstutter and frame latency tests?? Seriously?

According to wikipedia, Tests have shown that Nvidia has made major improvements on microstuttering, tests with dual 680's showed only a 7% variance in frame delay in BF3 compared to 5% with a single 680. As for 3-way SLI, Tom's did a test awhile back that showed that 3-way crossfire greatly improved microstuttering over 2-way crossfire, so so I assume the same would hold true for SLI. As for AMD, Tom's did a test on the 7990 with the free program RadeonPro awhile back, which greatly improved microstuttering. Here it is: http://www.tomshardware.com/reviews/radeon-hd-7990-devi...

So, in short, microstuttering isn't as big a deal with SLI/CF as it used to be.
February 18, 2013 8:28:54 PM

downhill911Or just grab used GTX 690 for about £530-600 from ebay UK (duno about US price, sorry), for 1080p gaming it is just enough to ultra everything and even on 1440p you'll be just fine, games will drop under 60FPS in some games but still the most power and price efficient card (when buying used one).


Two 670s could be cheaper for pretty much the same performance. Last I checked in the US prices, the 690 is still around $1000 whereas you can get 670s under $400 each.
Anonymous
February 18, 2013 8:32:39 PM

Loving my 2 GTX 680s 4gb each in SLI at 2560 x 1600, good review. Its good to see the king of cards (nvidia side) best 3 el cheapo versions at higher resolutions. I wouldnt do 3 1080 displays though, just too small in my book. 3 30 inch ones thats what I want to see next.
February 18, 2013 8:53:38 PM

These tests should be redone using 3DMARK's 2013 suite of benchmarks and use nVidia's just released 314 drivers
February 18, 2013 8:55:04 PM

blazorthon said:
Two 670s could be cheaper for pretty much the same performance. Last I checked in the US prices, the 690 is still around $1000 whereas you can get 670s under $400 each.

I really can't wait to see what Titan will bring to the table with the rumored $899 price tag.
February 18, 2013 11:10:43 PM

we are in that dead zone where everything means less till we see what Titan can do. I have never spent more than $300 for graphics but if my budget works out soon it may be time to take that elite jump.
I am married so whatever I spend my wife will expect equal spending privileges. That will be an expensive upgrade.
February 19, 2013 12:37:16 AM

unknown9122These tests should be redone using 3DMARK's 2013 suite of benchmarks and use nVidia's just released 314 drivers


Not really... the new suite is pretty disappointing.
February 19, 2013 1:31:20 AM

ubercake said:
I don't experience microstutter with my two 680s in SLI. I had it constantly with my 2 5850s and still with 3 5850s. I never had a problem with microstutter with 2 580s in SLI either.

I challenge everyone to try both multi-card technologies for themselves and then talk about their results. Don't just take the word of others as the truth.


You can't use both CF and SLI and get useful results from just that context. Different GPUs (especially of different architectures) can vary in effectiveness with the technologies even with the same drivers (granted the drivers are a part of it). Radeon 5000 cards and the other graphics adapters that have GPUs that are VLIW5 are all known to have had above-average stutter issues in Crossfire compared to a VLIW4 or GCN GPU of a similarly performing graphics card. For example, this is why Radeon 6850 Crossfire is usually said to be much more stutter-prone than Radeon 7770 Crossfire regardless of the drivers being the same or different no matter what version they are. This is despite the two cards having comparable performance in single GPU situations.

Also, faster GPUs tend to reduce stutter compared to slower ones. For example, this is why GTX 560 SLI is usually reported to be much more stutter-prone than GTX 580 SLI. Yes, Nvidia does and has had issues with this as well. All graphics cards, especially dual-GPU graphics (be they on a single card or on two cards), fail to have perfectly consistent frames. A single GPU card usually keeps them far below any perceptible level, but not always (regardless of the GPU's designer). Dual-GPU setups pretty much always make this issue worse, but not always so much so that it is perceptible by most people. Extremely high end cards such as the GTX 680 rarely have significant issues, but that is not a guarantee to never have issues in any situation.

The drivers are also a factor. For example, Catalyst 13.1 is usually reported as smoother than previous versions and Catalyst 12.7 is usually reported to be smoother than previous versions. AMD is also not the only one that has had driver issues with this even recently. For example, for a few months after their launch, Nvidia's new drivers had some issues with stuttering. It might not have been micro-stutter, but instead a different form of stutter, but it was stutter nonetheless. Another aspect of drivers is that not everyone has the same issues even with the same hardware and driver version. So, not everyone had this issue. However, there was a large number of people (large enough that Nvidia had to recognize it and a few other issues and make a statement about fixing it) with the issue.

These are the main graphics-related factors. Software/operating system, other hardware, and much more can also have an impact. Except for significant bottle-necks such as pairing a GTX 680 with an Athlon 64 x2 CPU or some other huge bottle-neck such as using extremely slow RAM, the rest of the system is usually not of huge significance. However, it's still a factor.

My point is that if you want to make a comparison for micro-stutter, you have to take in all of the factors and make several apples to apples comparisons to isolate many of these factors.
February 19, 2013 1:32:51 AM

unknown9122 said:
I really can't wait to see what Titan will bring to the table with the rumored $899 price tag.


I'm interested in it too.
February 19, 2013 1:53:53 AM

Here's to hoping that this site is wrong, but here's a disappointing perspective on Titan:


Link:

http://www.egypthardware.com/?p=7804

If this is accurate, then Titan isn't going to be worth buying in most situations. Two 670s could beat it by huge margins in most situations while being much cheaper.
February 19, 2013 2:52:01 AM

Titan is for people with lots of money to spend on less gains, and dont care for SLI much.
The plus i see is tat it is a "halo product" for Nvidia. Just see the nuzz it has created!
February 19, 2013 2:59:27 AM

BigMack70 said:
We already knew that a pair of 670s/680s/etc was going to be faster.

Titan has always been for:
1) People who hate multi-GPU but need more than current single GPU performance
2) People who want to drop a TON of cash for multiple Titan cards in SLI


Yes, but most sources that I've seen said that the difference was expected to be greater than this.
February 19, 2013 3:20:43 AM

ssd_pro said:
Not really... the new suite is pretty disappointing.

The DX11 benchmark looks suite. I think it is disappointing because they are trying to focus on tablets and mobile OS so it takes away from the desktop development.
February 19, 2013 3:22:25 AM

blazorthon said:
Here's to hoping that this site is wrong, but here's a disappointing perspective on Titan:


Link:

http://www.egypthardware.com/?p=7804

If this is accurate, then Titan isn't going to be worth buying in most situations. Two 670s could beat it by huge margins in most situations while being much cheaper.

Only by 100 bucks though. The titan will use less power though and produce less heat.
February 19, 2013 3:27:06 AM

BigMack70 said:
You're forgetting the part where the 670s are like 50% faster than the current rumored performance...

Let's see... 50% faster for $100 less at maybe the expense of 50W more power draw. Real close call there...... :sarcastic: 

Titan is for people with boat loads of money ready to drop on multiple Titan cards or for people who are tinfoil hat paranoid about multi-GPU.

Now maybe we get reviews tomorrow and find out it's actually much faster or much cheaper than currently expected, but it's a fail product according to current info.

Forgot about the core count right :lol: 
February 19, 2013 3:32:06 AM

unknown9122 said:
Only by 100 bucks though. The titan will use less power though and produce less heat.


Actually, many 670s (granted usually just reference models, but still) can be had for around $350 when on sale, so the difference can be much more than $100. Like BigMack70 said, the power consumption difference is also unlikely to be significant whereas the performance difference is likely to be huge.
February 19, 2013 3:40:51 AM

BigMack70 said:
85% of 690 performance according to TPU's summary charts put it at 7970GE +30% at 1600p, so I guess this is just about what I expected.

It just sounds more impressive to say "85% of a 690" than "30% faster than a 7970 GE" when you're trying to sell a $900 card :lol: 

Anyways, all should be revealed in a few hours now...


That is true, but I was seeing many reports saying 60% over the 680. Instead, it's closer to the smaller number of reports that said around 40%. I find it surprising considering the difference in specs seem to imply a greater performance difference IMO. If it's only 40% over the 680, then it'll also be the first card that I failed to accurately predict the performance of based on specifications and that's saying something considering how well my math has worked for several generations since I started using it. It looks like I'll have to re-work my math for this card...
February 19, 2013 3:45:43 AM

BigMack70 said:
Also to be fair, we don't really have any leaks from a broad test suite yet... I expect that something (either performance or price) is meaningfully different from what I'm seeing now.


I hope so you're right about that.
February 19, 2013 3:45:48 AM

how about hd 7850cfx just to see as i have one glad i got it looks dirt cheap compared to these guys
February 19, 2013 3:53:04 AM

also why any onew sli 680s is crazy sli gtx 660 is enough one gtx 660 or h d7870 are enough for 1 monitor
February 19, 2013 3:53:15 AM

masmotorshow about hd 7850cfx just to see as i have one glad i got it looks dirt cheap compared to these guys


Two 7850s would be around a reference-clocked 7970 GHz Edition in average performance, maybe a little higher.
February 19, 2013 3:58:06 AM

masmotors said:
also why any onew sli 680s is crazy sli gtx 660 is enough one gtx 660 or h d7870 are enough for 1 monitor


The point of the article was an ~$1000 Nvidia graphics budget comparison. What I got from it was that although both are poor values at that point, two 680s are still better values than three 660 Ti cards and three 660 Ti cards are without a doubt generally somewhat faster than two 680s.

Your claim about what is worth buying seems to leave out a ton of important context such as the graphics intensity settings, game, intended performance (IE if you have a 120Hz refresh rate display and want frame rate delivery to consistently keep up with it, especially for 3D gaming where it's even more important) and resolution, so it's pretty inaccurate in addition to being subjective.
February 24, 2013 12:50:31 PM

Can it play crysis 3 at 300 frames per second? lol what a ridiculous comparison. Spend more time rustling jimmies at MSI and the like, TOMS. I can't be bothered with this nonsense.
February 25, 2013 7:10:24 AM

So proud of my two 660 Ti's.
!