Review Index:
Feedback

Frame Rating: GeForce GTX Titan, GeForce GTX 690, Radeon HD 7990 (HD 7970 CrossFire)

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

While all three cards are able to keep Battlefield 3 running well at 1920x1080, in the FRAPS based information the HD 7970s in CrossFire that are emulating the HD 7990 are clearly the performance leader followed by the GTX 690 with the GTX Titan rounding things out. 

View Full Size

Well, things change quickly around these parts and you can see that for the HD 7990 the observed frame rate after removing any runts or drops has come down considerably. 

View Full Size

Our plot of frame times from the Frame Rating capture technology shows two interesting items.  First, the HD 7970s in CrossFire result in an alternating fast/slow frame times, usually indicative of runts, frames that take up so little of the screen's scanlines that they aren't positively affecting apparent performance.  Also, even though the GTX 690 has better frame rates than the GTX Titan, it definitely has more frame time variance as evident by the wider blue band of color on the image above.

View Full Size

Minimum frame rates after taking out the runts result in an observed FPS average of about 102 FPS for the HD 7970s, 120 FPS for the GTX Titan and 140 FPS for the GTX 690. 

View Full Size

But once we look at the variance picture again we find that the GTX 690 and the GTX Titan have swapped places, with the single GPU performance of the Titan resulting in a smoother overall experience than even the GTX 690.  Both NVIDIA solutions are drastically better than the HD 7990 / HD 7970s in CrossFire.

 

View Full Size

At 2560x1440 the FRAPS results look pretty similar to those above...

View Full Size

But once we take away the runts and drops we find the HD 7970s in CrossFire fall behind the performance of both of the NVIDIA GeForce cards.

View Full Size

Ouch, another blanket of color from the Radeon solution that indicates unsmooth and inconsistent frame rates!  If we look just as the NVIDIA side of the equation we again see the a thinner band of color on the GTX Titan results that indicates tighter and more consistent frame times throughout the benchmark run.

View Full Size

Here is an individual run graph for the HD 7970s in CrossFire to help demonstrate how the runts cause the observed frame rates to be lower. 

View Full Size

View Full Size

And the two runs for the GTX 690 and the GTX Titan do not indicate any runts at all...

View Full Size

These minimum FPS percentile charts show some pretty dramatic differences, even between the competing NVIDIA options.  The HD 7970s in CrossFire average around 60 FPS, the GTX Titan at 75 FPS and the GTX 690 at 95 FPS. 

View Full Size

But the frame variance results, our ISUs (International Stutter Units), once again prove that the single GPU solution has a more consistent and fluid frame time result with the HD 7970s in CrossFire really seperating themselves (not in a good way) starting at the 80th percentile.

 

View Full Size

View Full Size

Even though we only have NVIDIA results for 5760x1080 due to the extreme amount of dropped frames on the HD 7990 / HD 7970s, comparing these two options is interesting.  In both FRAPS and observed average frame rates per second, the GTX 690 is showing as the faster of the two options, running faster than the Titan the entire time.

View Full Size

But the plot shows an interesting story - the frame times on the GTX 690 are not as consistent or as smooth as on the GTX Titan.  They are averaging much lower, based on the where the bulk of the blue band resides in comparison to the green band, but the spikes that show themselves on the GTX 690 are gone completely with the GTX Titan. 

View Full Size

If we look at only the minimum FPS marks we find the GTX 690 to be 33% faster in average frame rate over the entire run, but based on the graph above (and the one below) that isn't the whole story.

View Full Size

Here we see the result of all of those "spikes" in frame times - a pretty sizeable difference in frame variance going from the GTX 690 to the GTX Titan. While the Titan never has more than 2.5 ms of variance from one frame against the running average of the past 20, the GTX 690 has 8-9 ms jumps at times, which will likely cause some noticeable stutter.

 

There are two take aways from this first page of results.  First, the AMD Radeon HD 7990 or HD 7970s in CrossFire are not going to compare well to the GTX Titan or GTX 690 in many cases because of the runt and dropped frame issues we have detailed.  Second, while the GTX 690 may be "faster" than the GTX Titan in Battlefield 3, at higher resolutions and especially for multi-monitor situations, the GTX Titan looks to provide the better overall experience.

 

March 30, 2013 | 08:45 AM - Posted by bystander (not verified)

Good read, I like the new charts. They are easy to follow. I think these are the best looking graphs I've seen on the subject, so keep up the good work.

I happen to be a 3D Vision user. I play it with about any game I have and when it doesn't work, I use a Helix mod in most cases.

Anyways, I'd be very interested in seeing how Crossfire/SLI works in 3D, be it HD3D or 3D Vision. Theoretically, I believe their results should be similar to a single card setup in variance, but much faster, though some confirmation would be nice. This is because two images are made for each frame, so each card will start a frame at the same time for delivery at the same time, while something like the Titan would have to create two separate images for each frame.

Maybe just one article on it at some point would be nice just to paint a picture. We wouldn't need a lot of them if my theory is correct.

March 30, 2013 | 11:26 AM - Posted by Ryan Shrout

Hmm, not sure how we could use the overlay for that but we will experiment. 

March 31, 2013 | 12:17 PM - Posted by Luciano (not verified)

Running Tridef 3D in passive mode instead of Shutter Glasses the strangest thing I've discovered is that the same FPS feels faster.

3dVision shutter glasses in surround with 400/500 series (it had to be SLI for surround to work with reference boards) according to simracers had stutter.
But that was fixed using .ini configs that PC racing simulators had documented for quite some time.
Those .ini configs, years later I discovered, lead to the same results as nVidia's smoothing techniques.

This is a 2005 racing simulator:

Render Once Per VSync="0"
Max Framerate="0.00000"
Steady Framerate Thresh="0.00000"
Flush Previous Frame="0"
Synchronize Frame="1.00000"
Delay Video Swap="0"

April 2, 2013 | 03:55 AM - Posted by Swolern (not verified)

Hey Bystander! Long time no talk to. Hope you are doing good!

I second the 3d testing please. I'm running 3d vision surround with Titan SLI and would love to see frame time effects with 3d enabled. To me 3d with Lightboost is much smoother than non-Lightboost so I believe the display used will affect your perceived stutter also.

April 2, 2013 | 03:55 AM - Posted by Swolern (not verified)

Hey Bystander! Long time no talk to. Hope you are doing good!

I second the 3d testing please. I'm running 3d vision surround with Titan SLI and would love to see frame time effects with 3d enabled. To me 3d with Lightboost is much smoother than non-Lightboost so I believe the display used will affect your perceived stutter also.

April 2, 2013 | 03:56 AM - Posted by Swolern (not verified)

Hey Bystander! Long time no talk to. Hope you are doing good!

I second the 3d testing please. I'm running 3d vision surround with Titan SLI and would love to see frame time effects with 3d enabled. To me 3d with Lightboost is much smoother than non-Lightboost so I believe the display used will affect your perceived stutter also.

March 30, 2013 | 09:18 AM - Posted by Xen (not verified)

I think most of us understand that 2x 7970s aren't the same as the upcoming 7990, but we appreciate that Ryan and his team used what is available currently from AMD as a comparison to Nvidia's high end solutions.

Thanks for the review!

Guys, please keep the trashy, immature comments to your selves.

April 2, 2013 | 10:54 AM - Posted by truth seeker (not verified)

Yes, but the article is deviously worded to portray the unreleased 7990 in a bad light. It's true the 7970x2 may be pretty bad, but that should be talked as it is, not trying to portray an unreleased card in a bad fashion!

March 30, 2013 | 09:43 AM - Posted by Brokenstorm

I'm curious, have you considered using a dual socket LGA-2011 with 128GB of RAM for RAMDisk as a temporary storage for the capture system?

Obviously such write speeds aren't needed right now, but should a capture card capable of 4k@60Hz or 1600p@120Hz ever be made, then it might be cheaper to use a RAMDisk than to buy the entrepise SSDs that would be needed to capture at such speeds.

March 30, 2013 | 09:45 AM - Posted by Anonymous (not verified)

PCPER still shilling hard for nvidia. 7990 isn't even released and he is 'reviewing' it here by using two 7970s.. what a joke this site is.

Without knowing what the 7990 actually has on board or what the drivers will be like for it, he is 'reviewing' it now and using the 7990 name. Never mind that the 7990 is using two of the new Malta cores, not the Tahiti cores in the 7970.

You're a joke, dude. Shill harder for nvidia why don't you. When is the next nvidia card release where you'll once again have their PR man and yourself shilling away in your live stream (free nvidia advert)

Same guy who was releasing information using nvidia's toolset without disclosing everything was being done with the help of nvidia. While decent review sites didn't do anything until they first came out and said the tools are from nvidia.

PCPER is a worthless nvidia shill site.

March 30, 2013 | 10:01 AM - Posted by billeman

The 7990 is 2 7970 GPU's with a PCIe bridge, doesn't sound too different to 2 7970's on different PCIe slots

March 30, 2013 | 10:45 AM - Posted by Josh Walrath

Yeah, the 7990 is just two 7970 GPUs connected via a PCI-E 3.0 bridge chip that then communicates with the system at x16 PCI-E 3.0 speeds.  It might show a small improvement in communicating with each other, but it is going to be minimal.  The problems that AMD have will require a pretty hefty driver revision to smooth things out.

March 30, 2013 | 11:28 AM - Posted by Ryan Shrout

Cool story, bro.

March 31, 2013 | 11:30 AM - Posted by Luciano (not verified)

I was questioning that but as I discover it wouldnt make no difference.

And PCPER never backed nVidia in the articles Ive read.
I had AMD 6900 and nVidia 570 and Frame Limiter + Adaptive vSync or x-buffering eliminates stutter in the racesims I play to experience the same as with nVidia 500.

But as PCPER says: although available solution (with Radeonpro) this is NOT AMD's control panel option and is not supported by AMD.

April 3, 2013 | 09:55 AM - Posted by Anonymous (not verified)

The maker of msiafterburner and evga precision has written the fcat type color bars into the new releases.

The problem for you is amd is fail, and the new evidence is unassailable.

The last cry of the dying, lying cf breed, total annihilation is moments away.

None of us will be awaiting your certainly never forthcoming apology, I however will be enjoying your delicious tears, as the full implications of total epic CF amd fail sinks home into the thick, biased, crank of amd fanboy water carrying bloated skull.

BWAHAHAHAAAAAAAAAAAAAAAAAAAAAA !

You may buy the newly released book at online shops everywhere:

: "Death of the amd fanboy"
The lies, the fantasies, the obstinate denial, years in the stuttering darkness, and the gruesome ending when full exposure and half the frame rate ripped their living guts out. Delicious amd fanboy tears ending. Don't miss it !

March 30, 2013 | 11:28 AM - Posted by John Doe (not verified)

Ryan and Josh are doing a pretty damn fine job at deleting EVERY single comment related to JonnyGuru people.

That should tell you about their psyche.

March 30, 2013 | 01:00 PM - Posted by technogiant (not verified)

The bit I don't understand is why the crossfire configuration doesn't have any problems in....I think it was Dirt 3.

I mean if this is supposed to be something that is fundamentally broken in crossfire it should be the same in ALL games.....so whats going on differently??

March 30, 2013 | 04:33 PM - Posted by Ryan Shrout

My theory is it does not happen as often when the CPU is more of the bottleneck in the game. 

March 30, 2013 | 08:55 PM - Posted by bystander (not verified)

It certainly would be a good thing to test.

March 31, 2013 | 12:25 AM - Posted by Anonymous (not verified)

Isnt that where GPUView from Microsoft comes into play. That seams to be a better tool then this is.

Nvidia FCAT grabs the information overlay at the point where Fraps take it at the beginning of the pipeline and merges it to the Frame output at the end of the pipeline.
Its supplying two pieces of information at different points of the pipeline and presenting them as one.

Thats an odd way of measuring things to say the least. Especially if your discounting what goes on in the middle.

March 30, 2013 | 01:33 PM - Posted by Anonymous (not verified)

Wow people can be assholes huh? Anyway great job Ryan!!!

March 30, 2013 | 01:52 PM - Posted by Braveheart (not verified)

Loving the framerating reviews in comparison with standard benchmark based testing (Which clearly still has its place)

It's like the difference between testing a car on a rolling road and taking it for a thorough test drive.

Well done to all the team.

March 30, 2013 | 03:13 PM - Posted by technogiant (not verified)

Ryan...if your theory is correct that this problem only occurs with crossfire when the gpu is the primary bottle neck...then this should be very easy to test out just by dropping the graphics settings/resolution and seeing if it goes away? I'm sure you've already thought of doing this.

Cheers...I'm off now to see if granny has learned to suck eggs....lol

March 30, 2013 | 11:44 PM - Posted by technogiant (not verified)

Or visa versa use dirt 3 and drop the cpu clock speed and see if the runts come back.

March 30, 2013 | 07:31 PM - Posted by Anonymous (not verified)

EPIC FAIL. Where is Titan SLI comparison?

March 30, 2013 | 07:41 PM - Posted by Ryan Shrout

Not in this article...  I don't see how that affects our results at all??

A pair of Titan cards prices out at $2000, not the $1000 that all three of these cards in our comparison today were around.

March 31, 2013 | 12:27 PM - Posted by renz (not verified)

why is it fail? the point is to compare single gpu frame time vs multi gpu configuration. btw since you want Titan in SLI action why don't you donate one so PCPer can make the test?

March 30, 2013 | 09:04 PM - Posted by Maester Aemon (not verified)

Great job Ryan! Kudos to you and PCPer for going through all this testing.

This confirms what I suspected. I have been using ATI/AMD cards in single and xfire configurations for many years and always felt there was something amiss with dual gpus, but could not pinpoint it before now.

March 31, 2013 | 12:56 AM - Posted by Pendulously (not verified)

The Frame Variance Graph: "...What this does NOT really show are the large hitches in game play seen as the spikes in frame times. Another stutter metric is going to be needed to catch and quantify them directly..."

As an example: If Average FPS (over 60 seconds) is 100, then Total Frames observed over 60 seconds is 6000.

If ONE SINGLE FRAME is above 100ms, then for the y-axis value '100' (milliseconds), the x-axis value will be '99.9983' (percentile), i.e. one minus (1/6000).

If FOUR FRAMES are above 30ms, then for the y-axis value '30' (milliseconds), the x-axis value will be '99.9933' (percentile), i.e. one minus (4/6000).

If TEN FRAMES are above 20ms, then for the y-axis value '20' (milliseconds), the x-axis value will be '99.8333' (percentile), i.e. one minus (10/6000).

Therefore, instead of PERCENTILE on the X-AXIS, you should put NUMBER OF FRAMES on the X-AXIS.

Following our example, for the y-axis value of '100' (ms), the x-axis value will be '1' (frames), for y-axis '30' (ms), the x-axis will be '4' (frames), for y-axis '100' (ms), the x-axis will be '10' (frames), and so on.

March 31, 2013 | 01:01 AM - Posted by Pendulously (not verified)

EDIT: Therefore, instead of PERCENTILE on the X-AXIS, you should put NUMBER OF FRAMES on the X-AXIS. Following our example, for the y-axis value of '100' (ms), the x-axis value will be '1' (frames), for y-axis '30' (ms), the x-axis will be '4' (frames), for y-axis '20' (ms), the x-axis will be '10' (frames), and so on.

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.