Review Index:

Frame Rating: GeForce GTX 660 Ti and Radeon HD 7950

Author: Ryan Shrout
Manufacturer: PC Perspective

Crysis 3

Crysis 3 (DirectX 11)


Master the power of the Nanosuit. Armor, Speed, Strength and the ability to cloak are the ingredients of the most effective tactical combat armor ever created. Suit up! It's all yours in Crysis 3.


View Full Size

View Full Size

Crysis 3 starts out much in the same way that Battlefield 3 did - CrossFire looks much different in FRAPS than it does in our Frame Rating-based observed FPS result.

View Full Size

The reason is pretty simple to understand when you view a frame time graph like this.  The single HD 7950 does outperform the GTX 660 Ti by a small margin but with SLI's smoother multi-GPU result makes a dramatic impact for anyone that might be considering pairs of either card.

View Full Size

The minimum FPS percentile information shows another view of performance where the SLI result is obviously the only one that scales with two GPUs.  The GTX 660 Ti cards goes from about 25 FPS (total run average) to nearly 45 FPS.  While both single cards have consistent frame rates you can see the sudden drop off at the end for SLI.

View Full Size

Interesting results here - the HD 7950s in CrossFire are clearly the most troublesome with the frame time variances.  The GTX 660 Tis in SLI also show more variance than the single cards after the 90th percentile but it is much less dramatic than the 20+ ms jumps you see from AMD.


View Full Size

View Full Size

While all card performance starts to fall off at this resolution and these quality settings, the same problem persists with the HD 7950s in CrossFire.

View Full Size

With a slower frame rate due to performance issues, the problem of AMD's CrossFire gets worse as we see frame times alternate between ~5 ms and ~85 ms!  SLI does have some hitches and spikes in frame times as well and only the single Radeon HD 7950 3GB card can run through without a problem.  Obviously the 2GB vs 3GB frame buffer is starting to make a difference in Crysis 3 at Very High settings.

View Full Size

While the GTX 660 Ti definitely scales going from one card to two, the frame rate definitely starts to drop towards the end of the graph, and more suddenly than we would like.  Looking at just the single GPU cards, the Radeon HD 7950 starts out with an average just a couple of FPS higher than the GTX 660 Ti but it stays about that much faster throughout our test.

View Full Size

Our 2560x1440 frame variance results again paint the HD 7950s in CrossFire in another negative light with frame time variances that are high from beginning to end.  Note that the GTX 660 Ti, in both single GPU and SLI configurations, actually has more variance than we would like at these settings as well with only the Radeon HD 7950 keeping things within check.


View Full Size

View Full Size

Things aren't looking pretty for any of these cards at 5760x1080 at these quality settings - future runs will likely be run at lower IQ options.  We still see a ton of dropped frames from the AMD Radeon HD 7950 CrossFire config. 

View Full Size

View Full Size

Both CrossFire and SLI show some varied frame times in our plot here though CrossFire continues to look much worse with a lot of frame occillation.  The black line, slightly hidden by the SLI result, is clearly the best performer.

View Full Size

Interestingly, even the GTX 660 Ti in SLI can't really keep up with the results of the single HD 7950 that puts in the most consistent work.  At about the 85th percentile the GTX 660 Ti SLI results actually fall below that of the single GTX 660 Ti and the CrossFire HD 7950s.  This is definitely showing the cases where gaming scenarios that are heavily limited by the CPU and other factors can be bottlenecked in other ways.

View Full Size

The Radeon HD 7950 3GB is definitely the best option of the four card configurations tested here with both CrossFire and SLI resulting in high variance levels.


April 2, 2013 | 03:24 PM - Posted by technogiant (not verified)

Absolutely god awe-full again for AMD....a couple of 7950's in crossfire was meant to be about the best bang for buck high performance system you could build....oh how deluded or even worse misled we have been.

April 3, 2013 | 11:59 AM - Posted by Anonymous (not verified)

Deluded, misled, fanboyed out, raging radeon rightist, screaming psychotic frame rate fanboy, totally full of it, amd marketing borg bot, COMPLETELY FOOLISH IDIOT WHO CANNOT SEE THE GAME ON THEIR OWN ULTRA HIGH REZZ BRAGGADOCIO DISPLAYS WITH MEGA BUCK AMD CARDS....

I cannot even fathom the utter SHAME the CF amd fanboys must be feeling now.

A thousand times on every website we were told lies, by the reviewers, by the fanboys raping away in the comments section.

Now we know they were all deluded idiots, period.

AMD has claimed utter ignorance, and I certainly do not believe that since it's worse than being just incompetent in writing drivers and making hardware, and extends the incompetence out ever further, into "not knowing what is going on at all" instead of covering it up because monetary concerns rule the day.

It's one giant can of fail, and AMD has gulped the entire 55 gallon drum in one fail swill.

I wonder, will they pay the full price refund for one of the amd cards for every crossfire user, because indeed they certainly owe them. Hopefully a giant lawsuit will make the amd fanboys pocketbooks right, and I could see the penny pinching fanboy scrooges getting behind that 100%.

The next thing I want to know is how many of the LYING BUFFOON IDIOT articles from the hundred review sites that show amd fraps! in crossfire winning will be CHANGED, UPDATED, and in big red letters at the top:


(ROFL - Oh man, they ALL need to resign)

That should only take them a few months to fix on all their websites since the amd underdog fanboyism is so freaking strong it could destroy all the rainforests of the world and induce global warming and nuclear winter concurrently.

O M G - thanks for the years and years of total lies

April 3, 2013 | 02:58 PM - Posted by Anonymous (not verified)

Dude...are you okay?

You have issues.

April 3, 2013 | 07:51 PM - Posted by Anonymous (not verified)

poor bitter guy.. go f*ck yourself and try getting a life. these guys at pcper are doing their job right unlike you causing misery to others making them feel bad..

kudos pcper for a well done review..

kudos to amd and nvidia

in the end. consumer wins the deal!! ^_^

April 4, 2013 | 05:05 AM - Posted by Anonymous (not verified)

poor bitter guy is spot on.

they all deserve to feel bad for spreading their ignorance like a virus.

kudos to amd ? get outta here.

April 5, 2013 | 02:53 AM - Posted by thinkbiggar (not verified)

Assuming data collection method has validity which on the surface seems to make sense, then the best we can say here is AMD and Nvidia are neck and neck in single card performance and AMD has some real crossfire performance issues. However I play some games that generate a fair amount of lag with with single card performance and I add a card in crossfire and see a huge improvement. Secondly, in crossfire you see time in ms is all over the place, not as tight, and us guys know tighter is better right? Well how is your data going to as low as 0 ms on the crossfire setup. This is a clear indication that something in your formulation is wrong. For all we know this variance is actually improving gaming visual effects and reality.

April 9, 2013 | 11:05 PM - Posted by Anonymous (not verified)

If you are not satisfied... get two GeForce GTX 660 Ti and SLI it. AMD cards are not designed to handle two cards (CrossFireX).

April 2, 2013 | 03:32 PM - Posted by Shambles (not verified)

Aha, starting to get into cards in the price range I'm looking for. I'll be greedily going through this article and waiting for the next one with the GTX 660 and 7870 before buying my next card.

April 3, 2013 | 02:41 AM - Posted by alwayssts (not verified)

Unless something is shown to be massively off the rails, the results will repeat the pattern.

That said, I own a 7870 and while I could argue all the clear merits, metrics, and reasoning vs. 660, the jist of the matter is it too, while slightly better, is at most (overclocked) sitting right at to slightly under where you want high quality 1080p performance now-a-days (and moving forward considering the ps4 spec). I would recommend you look closer at this article; buy a Tahiti LE if these are out of your budget and overclock it to 7950B speeds...or wait for the inevitable Hainan SKU if length/size is a factor (which should also drive down the price of 660ti to around ~250, in theory).

If you think that performance level is overpriced now, like I did, just wait. It's going to be a very important market soon because of the analogous to new consoles (1080p 30/60hz) so the competition will be incredibly fierce, if not then stagnant for quite a while.

April 2, 2013 | 03:45 PM - Posted by Anonymous (not verified)

Thanks for yet another great article, I can only imagine the amount of work that has gone in to testing all these cards and games at different resolutions. I do hope AMD will fix their broken CrossFire system to balance out the competition, even though I'm not a fan of their cards myself.

That said, it would be nice to see how two Titans would perform in SLI. I swore to stick with a single card, but two of them are starting to tempt me as SLI appears to perform with smooth frame times.

April 2, 2013 | 03:54 PM - Posted by seravia

Is AMD implementing the summer solution for the previous HD6000 series too?

April 2, 2013 | 11:21 PM - Posted by arbiter

Its gonna be a software fix less they re-release their chipset which i doubt will happen. Nvidia apperently seen this issue few years ago and do it via hardware on their boards. Would expect AMD to address it properly with their next generation chip.

April 2, 2013 | 04:07 PM - Posted by Bob Jones (not verified)

Wht's up with the FRAPS FPS for Far Cry 3 at 1440p in Crossfire. It's showing 4x the performance of a single card - why do no other review sites see that kind of error?

April 2, 2013 | 05:25 PM - Posted by Ryan Shrout

The numbers you see on that metric are based on re-adding the runt frames as well as the frames completely dropped by the game/GPU driver.  Because all of those dropped frames are put back into the compilation the of average frame rate per second, the original or FRAPS result can sometimes be much higher than is really shown.

We saw the same result at 5760x1080 in many of our tests in this article as well as in the HD 7970 vs GTX 680 piece.

April 2, 2013 | 06:11 PM - Posted by Anonymous (not verified)

It's because pcper and nvidia have together invented a reason to throw out frame numbers using their subjective and bias opinion to then create an illusion metric they call 'perceived fps'

pcper cannot be taken at face value due to their heavy nvidia involvement.

the bright side is that more credible hardware review websites will soon be giving their results using similar methods without the nvidia influence that pcper suffers from.

April 2, 2013 | 06:38 PM - Posted by bystander (not verified)

What about Toms Hardware, TechReport and AnandTech? Are they all creating a conspiracy? Use your own head, see the data, and tell me it doesn't matter.

April 2, 2013 | 07:18 PM - Posted by truthobfuscated (not verified)

Have been following much of this new method closely. Take your own advice. Tom's has already called into question that the smaller frames may have value and their worth or not is subjective and requires further study. pcper has hopped on board with nvidia and is towing their line on them having no value.

techreport too is saying the data from fcat is insufficient and more information from the api must be exposed for draw calls. you hear none of that here at pcper/nvidia (which is it, hard to tell any more)

anandtech has reserved any results at all until they can do a full and proper investigation with the fcat tools and will then release their findings.

pcper has attempted to get attention by cowtowing to nvidia in order to release data and riding the gravy train of amd bashing to garner a few page clicks from nvidia fanboys who swallow anything so long as it comes in nvidia positive or amd negative flavor.

much more investigation to be done, by reviewers unlike this obvious nvidia biased and sponsored one

April 2, 2013 | 07:30 PM - Posted by bystander (not verified)

While most are not giving final verdicts, they all see it as a problem. How severe of a problem is what they are holding back on.

There are a few things that they aren't even considering in all this either. Having evenly spaced frames, also means evenly spaced input. This gives a smoother feel as well.

How much latency really is added to the metering technology, if all it does is make one frame that is faster than the previous, wait a little. Once they are at the correct offsets, they should remain fairly close in line, with only occasional adjustments needed.

You also must have to consider that while the crossfire setup still gives a decent playing experience, is it giving as good of a playing experience? Reviews are meant to help us pick what is the better experience for the end user, not to just to find what is acceptable.

And if you've followed a lot of review sites articles, you probably have noticed many notes about different games giving stuttery results in Crossfire, and much fewer cases where SLI has been mentioned. They never made a big stink about it in their articles, because they are trying to give unbiased articles, and using quantifiable tests to gauge what is better. Now we finally have quantifiable evidence of crossfires problems, so it can given more attention.

I found I always had to run with v-sync with my 6950's. I was ok with it, until I learned that the added latency was the cause of my nausea.

April 2, 2013 | 07:51 PM - Posted by Shambles (not verified)

The trolling is strong in this one.

April 2, 2013 | 07:36 PM - Posted by bystander (not verified)

I will agree with one thing though. Pcper is definitely not holding back on their opinion, and would probably be more well received if they delivered the results without as much judgement.

While they may be right, a lot of people see it as an attack, and seem to be very defensive about it.

April 2, 2013 | 09:47 PM - Posted by ThorAxe

I realise the need to justify your purchasing decision but accusing PC Per of bias is one helluva stretch.

I have used 6870s in Crossfire and noticed these problems and yet I don't have the same issues with SLI.

The numbers don't lie however much people dislike the results.

Bravo Ryan for calling it as you see it.

April 2, 2013 | 10:40 PM - Posted by bystander (not verified)

I was agreeing with Pcper. The guy I originally responded to was the one who was trying to discredit it.

April 2, 2013 | 11:13 PM - Posted by ThorAxe

Sorry about responding to you bystander, it was intended for the guy you mentioned. :)

April 3, 2013 | 11:35 AM - Posted by Anonymous (not verified)

ROFL !!!!

"Smaller frames may have value" HAHAHHAHAHHAHAAAAAAAAAAAA

Please, you and the amd fanboys at Tom's play with runts and ghosts, that's valuable. A value that could not be understated.
Here let me show you what you game will look like:

_______________________________________ ___________
------------------------- -------------------------
____________ ___________________________________
- -----------------------------------------------


April 2, 2013 | 11:24 PM - Posted by arbiter

Maybe you didn't notice a lot of AMD sponsor ad's on the site? As for cannot taken at face value, if a site is biased in their reviews, no one will trust them. Yes the tools were developed originally by Nvidia but PcPer is working on coding their own tools to get away from those.

July 6, 2013 | 03:53 AM - Posted by Rod (not verified)

Hmm iѕ anyоne elѕe encountering ρrоblems with the imаges on this blоg
lοаding? І'm trying to find out if its a problem on my end or if it's the blog.

Any suggеstіonѕ woulԁ be greаtly aρpreсiated.

my webρage :: Southold eye eхam (Rod)

April 2, 2013 | 11:33 PM - Posted by Ryan Shrout

Try not to be such a troll.

How about this theory?  Because I have been using these tools, regardless of who developed them, for MUCH LONGER than Tech Report, Anandtech or Tom's Harware, I have more experience and am much more confident in the results they are showing me.  

Accusing me of NVIDIA bias while AMD ads run continuously on our site is also pretty stupid.

April 3, 2013 | 03:15 AM - Posted by andrei (not verified)

Well done Ryan !

You have destroyed all AMD's credibility as planed, without any real proof just with the help of Nvidia and I mean FCAT ( and who knows what else... )

" I have more experience and am much more confident in the results they are showing me" - You know just what Nvidia wants you to know... AND by the way You are a liar! First time when you mentioned about this "new frame metric" You told that it is developed by you and your team but on the first day of your benchmark you say the converter is made by Nvidia and I have to find from other sites that this whole benchmark is actually an Nvidia's product???

Now you can go and count your green money, upss I mean tell your readers Nvidia's propaganda, upss I mean the real objective stuff, you do care about your readers don't you??? as it can be seen you DON'T!

From a site with AMD in his name you became Nvidia's marketing guy (puppet ) but that is understandable because AMD does not give bribe to sycophants as Nvidia does ( How about this theory? ). Reading your articles it gives me the impression that they are written by Tom Petersen.

PS:go have fun with project shield, this is the future of gaming!- Nvidia marketing.
By the way you should rename this site to

"Accusing me of NVIDIA bias while AMD ads run continuously on our site is also pretty stupid." - you are not doing this freely! you receive good money for that, now who is pretty stupid?

April 3, 2013 | 06:38 PM - Posted by Josh Walrath

Yeah, well... except that AMD has publically admitted that they are having issues with CF and will have a fix in place by this summer (in Beta form).  Damn guys must be biased against themselves!

April 3, 2013 | 09:59 PM - Posted by ThorAxe

It's no use Josh, these zealots won't listen to reason.

I have a 4870x2 and 2 6870s still running in my kids' PCs but I am not blind enough to call Ryan a liar for exposing Crossfires deficiencies. Instead I greatly appreciate all the hard work that has gone into these tremendous articles.

It turns out that my eyes were not lying to me.

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.