Review Index:

Frame Rating: GeForce GTX 660 Ti and Radeon HD 7950

Battlefield 3

Battlefield 3 (DirectX 11)


Battlefield 3™ leaps ahead of its time with the power of Frostbite 2, DICE's new cutting-edge game engine. This state-of-the-art technology is the foundation on which Battlefield 3 is built, delivering enhanced visual quality, a grand sense of scale, massive destruction, dynamic audio and character animation utilizing ANT technology as seen in the latest EA SPORTS™ games.

Frostbite 2 now enables deferred shading, dynamic global illumination and new streaming architecture. Sounds like tech talk? Play the game and experience the difference!

Our Settings for Battlefield 3

Here is our testing run through the game, for your reference.

View Full Size

View Full Size

Under our FRAPS data you see that the HD 7950s in CrossFire appear to be faster than the GTX 660 Tis in SLI, but after taking out runts our observed average frame rates are much lower, more in line with the single Radeon HD 7950. 

View Full Size

Our now familiar frame time plot data demonstrates the problem with AMD's CrossFire technology and runt frames - frames taking up such a small area of the screen that they adversely affecting animation performance.  While both single GPU results look pretty good the HD 7950s running in parallel produce the mass of orange alternating long/short frames.  NVIDIA's GTX 660 Tis in SLI are definitely slightly more variant than the single GPU options but appear to be within reason.

View Full Size

The minimum FPS data shows the GTX 660 Tis in SLI running well ahead of any other option with average frame rate of about 122 FPS while the rest hoever in the 70 FPS range.

View Full Size

The GTX 660 Ti cards in SLI definitely exhibit some more frame time variance than the single GPUs but the Radeon HD 7950s in CrossFire definitely skew things dramatically.  While this doesn't definitely detect stutter it is a good indication that you will see some with results like this.


View Full Size

View Full Size

Again the Radeon HD 7950s look great in CrossFire on the outset, ahead of the GTX 660 Ti cards in SLI.  But after removing those dreaded runt frames the "real world" performance of the GTX 660 Ti cards is much better.

View Full Size

The Radeon HD 7950 combo actually looks worse at 2560x1440 with more variance on a frame to frame comparison where as the NVIDIA GTX 660 Ti cards appear to be tighter, resulting in a better multi-GPU environment. 

View Full Size

Our minimum FPS percentile data shows great results from both the single GeForce GTX 660 Ti and the Radeon HD 7950 but adding a second HD 7950 doesn't change much for AMD.  Only the GTX 660 Ti is scaling.

View Full Size

Again at 2560x1440 we see the frame variance ratings are much higher with the HD 7950s in CrossFire.


View Full Size

View Full Size

Eyefinity with CrossFire continues to be an even worse problem for AMD as we can see here there are tons of not only runts, but flat out dropped (never shown at all) frames that bring the observed frame rate down considerably. 

View Full Size

Ech, a mess of color here.  The blue lines are telling us that the 2GB version of the GTX 660 Ti are struggling with 5760x1080 in SLI mode though they are reliably showing frames to the gamer. Single GPU results (though kind of cut off at the top, sorry!) are much smoother but are definitely faster on the single HD 7950 than the single GTX 660 Ti.

View Full Size

The minimum FPS chart shows the single HD 7950 ahead of the GTX 660 Ti, as we expected based on the plot of frame times above (22 FPS average against 27 FPS average).  But while CrossFire sees basically no observed frame rate increases the SLI result jumps up to 38 FPS or so at the 50th percentile.  Note though that SLI in this case does follow a curve that brings the frame rates back down in line with a single HD 7950 at the tail end.

View Full Size

Our frame time variance chart has really interesthing things to tell us, starting with the expected "AMD CrossFire isn't working that well" kind of thing.  AMD's CrossFire clearly sees the most frame rate differences though NVIDIA's SLI isn't immune here with frame variances crossing the 15 ms in some fringe cases.  To put that in perspective though, CrossFire HD 7950s see variances larger than that for more than 20% of all frames rendered!


April 4, 2013 | 09:10 AM - Posted by Noah Taylor (not verified)

I really don't understand how incompetent you must be to accuse Ryan's last few articles of being all Nvidia and meant to destroy AMD. He explains his testing methods, his setups, and creates repeatable, measurable data for us to analyze. If you don't believe it's true, don't read it. I for one have been utterly displaced from my own enthusiasm for AMD's crossfire. I just couldn't understand why i had such a rough gameplay experience at 1440p when my frames were so high. What a mess.

BF3, Crysis 3, and Even Bioshock Infinite: anytime i'm in battles or high character count areas it becomes choppy and skippy, even though it says 100 fps :S

April 8, 2013 | 12:51 AM - Posted by raulduke

meh, useless trolls..

April 3, 2013 | 11:10 AM - Posted by Maester Aemon (not verified)

Was the tinfoil hat bundled with Radeon card or did you buy that separately?

April 2, 2013 | 04:31 PM - Posted by Daniel Masterson (not verified)

I'm rooting for AMD. I like to have a 2 card solution but I just can't do that with Xfire like I can with SLI. Here is to hoping they come out with a decent driver that really shows Xfire can work! Then I might get an 8000 series.

April 2, 2013 | 04:40 PM - Posted by Nigel (not verified)

I'm pleased with the frame rating results for AMD's single GPU solutions. I'll never do crossfire anyway but I'm hoping they fix their drivers for scaling... I can see people staying clear for AMD otherwise.

April 2, 2013 | 06:04 PM - Posted by Anonymous (not verified)

AMD is garbage. We are almost to the point of refusing to sell AMD sh*t. It really has gotten that bad.

April 2, 2013 | 06:12 PM - Posted by Shambles (not verified)

You must have read a different article since this one clearly shows that AMDs single GPU products are better than nVidia at this price point. Since the dawn of SLI/Crossfire it's been a bad idea to go multi-gpu on mid-range (lol $300 mid) parts rather than go with a single higher end GPU.

April 2, 2013 | 06:05 PM - Posted by db87

I Like to know how long these Crossfire problems exist. How much is the GCN Architecture to blame? How long have we been stupid in believing that Crossfire was just as good as SLI?

Take a HD6990 and a HD4870X2 for example and also add some old games (UT3, Doom 3, Half Life 2).

I have been running SLI for years in the old days. I never used Vsync because it was known to not work great for SLI (also was my experience) Since I have been using Radeon HD5850 OC in Crossfire I use Vsync for all singleplayer games and using framecap for multiplayer games. I always thought it had something to do with the Crossfire implementation and that it might have been optimized for Vsync primarly. I also did notice that Crossfire had some additional tearing. Now these days I know better...
Thanks for these great articles!

April 2, 2013 | 06:41 PM - Posted by bystander (not verified)

I know that FRAPS was showing problems in the 6000 and 5000 series. I also have heard SLI had problems back in the 200 series. I believe it was with 400 or 600 series that they started the frame metering improvements, since they said they started this a couple years ago.

April 2, 2013 | 11:34 PM - Posted by Ryan Shrout

This is a really interesting idea though I don't know how much time I'll be able to devote to older GPUs...

April 3, 2013 | 11:26 AM - Posted by Maester Aemon (not verified)

That would be an interesting test just to see if this issue was introduced with the GCN architecture or it's an underlying problems of ATI/AMD multi GPU implementation.

Maybe without going to far back, a test with a pair of 69xx cards shouldn't take too long ;)

April 3, 2013 | 02:38 PM - Posted by Luciano (not verified)

The next step, after AMD put an option for evenly distributed frames in the drivers, is going to be test input lag compared to SLI.
If the results of this future test for both are closer to vsync + triple buffering, then AMD could claim that there never was a problem and the option (of smoothing) was already there for all users.

April 2, 2013 | 09:21 PM - Posted by Anonymous (not verified)

The BF3 discussion mentioned GTX 680 SLI twice above and below the 4th plot. I think GTX 660 SLI was intended.

April 2, 2013 | 11:36 PM - Posted by Ryan Shrout

Thanks, fixed!

April 2, 2013 | 09:39 PM - Posted by Rick (not verified)

I know this is a ton of work, but....

I would like to see how VLIW4 compares to GCN. Is this a new problem with AMD cards, or has this been happening for a long time?

Thanks Ryan for doing all this work, it is a very interesting read.

And keep up the good work on the podcasts!!

April 2, 2013 | 09:44 PM - Posted by Rick (not verified)

Even if it is a quick game review (BF3?) of 6970's in crossfire.

April 2, 2013 | 10:18 PM - Posted by Foosh (not verified)

Crossfire and SLI both are unbearable without using something like Afterburner to meter the frames. I'd like to see that tested.

April 3, 2013 | 12:10 PM - Posted by Anonymous (not verified)

Nope. Only amd's crossfire is unbearable.

SLI is awesome and rocks the smoothness.

Frame rate target is ultra awesome on single card nVidia.

Only amd's crossfire is an utter failure and a giant fps lie.

CF fails completely on half the games, too.
On the other half they have false frame rate double pumped when in reality they display single card rates with massive stutter and tearing.

amd cf is epic fail incarnate

Now all the amd fans can scream once again, "you can't see anything beyond 30 fps, the human eye is incapable of it ! "
Then they can move their human eye benchmark to 60fps
Then to 120 since the 120hz monitors have come out...

In any case, is easy to say on amd crossfire you will get 30 frames MAX with massive stutter, so it is true that the amd fanboy human eye cannot see beyond 30 fps.
It's okay to be on disability with amd cards.

April 2, 2013 | 10:45 PM - Posted by Anonymous (not verified)

Guys!there are too many variable involved in somethi ng this massive .I all brain an angle most forget.what is needed to generate graphic?don't say GPU .next ?is control the vast majority of what is going on .Google sRGB 15 and read his view on color profile.this is just the tip of the iceburg .then you have data fragmentation everywhere always along the way and supposedly seemless.give me a break.nothing cause more trouble then the various cache level at network level at GPU level.add to this the new .net .then we have bufferbloat everywhere which cause more fragmentation.I tell you this because the main cause of my issue have been at the is level I have been trying to get rid of all those fragmentation point but so far the task is so big its insane .I ain't even started on the subject of color profile converting data for x y z to ?all along with the way wiggle room get by and at the end cause all the chaos pcper notice.but I bet if you ask ms they all say it is lower the human perception (85 ms)so it is a non issue.sadly ms is wrong .a fix?lol try getting the exact value of the black point of you monitor for fun.the white point?etc etc etc.there are a lot of value that shouldnt be on the eye ball guess or random yet a huge amount are.this is a good idea but without doing the same for all variable related to the os it is futile exercise because most of the issue come from the os because of all the compromise .imagine when 4k ITU standard is added to scrgb (?)lol it will be even worst.this is why tablet are popular like nexus pay 200 $and will look as good most computer at 1080p!ya there are issue .they are the same as they were when I was using xp 10 years ago .ms took 10 years to fix the universal better be patient on the issue you discover here cause most will be caused by the is (window)

April 2, 2013 | 11:36 PM - Posted by Ryan Shrout


April 3, 2013 | 01:40 AM - Posted by alwayssts (not verified)

Some people will never learn that online translators do not work in conversation. ;)

April 3, 2013 | 09:11 AM - Posted by Lord Binky (not verified)

Has anyone really been far even as decided to use even go want to do look more like?

April 2, 2013 | 11:51 PM - Posted by Anonymous (not verified)

You know for a site that's supposed to be hardware centric, you sure dropped the ball.

NOT A SINGLE WORD on the mini GTX 670.

HERE, I'll do some of your homework FFS.

April 3, 2013 | 11:13 AM - Posted by Ryan Shrout

You mean words like this?

April 3, 2013 | 01:37 AM - Posted by Trey Long (not verified)

latency and frame time delivery as more and more sites are testing will only improve ALL products. Obviously there is a big problem with Crossfire at the moment, and I have no doubt it will eventually be fixed.
My Titan btw, has the lowest latency of ANY solution, so that makes me feel even better about my addiction.

April 3, 2013 | 02:57 AM - Posted by Number1 (not verified)

We get the deal, AMD and nVidia both fairly well matched single card perf. AMD crossfire has problems if you for some idiotic reason don't use a frame limiter like radeon pro.
Blowing things out a little bit here mate. Time to start looking at the quality solutions that people have been using for a long time and that have been reported on for quite some time. Even Tom's Hardware did a bit on radeon pro some months ago and how it eliminates micro stutter. Pretty sure the end result was a win for the unofficial 7990 against the gtx690 in terms of latency and smoothness.

By quality solutions I don't mean ticking the "vsync" box in games then running the game at settings that only get you 40fps. I mean using radeon pro setting frame limit ~59 playing at settings that get you an average of 60+ and watching how smooth your cards look compared to the competition. Oh and how input lag free they remain.

April 3, 2013 | 10:21 AM - Posted by bystander (not verified)

The Toms Hardware article is here:,3329-...

You will notice that they used Dynamic v-sync within RadeonPro to fix it at different FPS limited points. It appears to be a v-sync/FPS limiter combo they are using.

It is worth using, but you shouldn't be required to use a FPS limiter or v-sync to fix stuttering either, but it is a solution for now.

April 3, 2013 | 03:22 AM - Posted by Pete (not verified)

Wow what a complete blunder for AMD!!! No wonder they are giving out amazing free games!! To sell their broken CF hardware!! What an eye opener!!

Looks like Im buying Nvidia this round.

April 3, 2013 | 03:31 AM - Posted by ReWind (not verified)

I have to agree with bystander here, seeing even just a slim article about CF/SLI performance with previous generation cards would have great value, as it would let a lot of people (myself included with my HD5770 CF setup) get some information on what an upgrade to a new CF setup will mean for us.

Having just ordered a Tahiti LE I would love to know if I will get the same amount of stuttering with two cards, just more FPS, or if the stuttering is worse, and in turn will make the upgrade of a lower perceived value than what FPS numbers imply.

April 3, 2013 | 05:17 AM - Posted by rrr (not verified)

Kind of proves Techreport wrong on 660Ti vs 7950, according to whom SINGLE 660Ti was supposed to be much better than 7950.

Then again, flawed conclusions lead to improvements in methodology and finding REAL problems, so I can't complain about that.

Is problem with CF hardware in nature? In last article you mentioned nVidia having built in hardware elements just to handle the issue, whereas AMD cards just output frames ASAP. So it means issue is unresolvable in drivers and AMD has to improve it hardware-wise in next generation? Looks like we have to disregard CF as viable option until it happens, then.

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.