Search

MTBS3D RT @IfcSummit: We are excited to welcome Director in Privacy and Security, Paul Lanois, for @Fieldfisher as a speaker at #IFCSummit. Paul…
MTBS3D Jim Jeffers talked about @intel’s efforts to enable over a billion users with creative and computing tools.… https://t.co/Z9fi0pS8xp
MTBS3D RT @IfcSummit: Adshir to discuss ray tracing at #IFCSummit. Adshir is strategically important to the #ClienttoCloud Revolution because it’…
MTBS3D RT @IfcSummit: We are honored to welcome Linda Sellheim, Education Manager for @EpicGames, to the upcoming #IFCSummit #IFCSummit2019 speake…
MTBS3D RT @tifcagroup: Enabling Breakthrough Innovations in the #ClienttoCloud Revolution. https://t.co/dglxcxLO3D
MTBS3D RT @tifcagroup: TIFCA’s Client-to-Cloud Vision document has been published. We are meeting during #SIGGRAPH2019 to jointly address the cont…
MTBS3D It’s been a major boon for the Client-to-Cloud Revolution at #E3. #E32019 #E319 #GoogleStadia #BethesdaE3https://t.co/IqIrR81D8o
MTBS3D RT @IfcSummit: Sixth International Future Computing Summit Moves to Silicon Valley November 5-6, 2019. Open Call for Visionary Speakers and…

Stereoscopic 3D Benchmarking: DDD, iZ3D, Nvidia

By Neil Schneider

While traditional gaming media has every benchmark under the sun, it’s very rare to find ratings for modern stereoscopic 3D drivers like DDD, iZ3D, and Nvidia’s GeForce 3D Vision.  This is completely understandable because it’s much more time consuming, the drivers don’t work equally well with the available measurement tools (e.g. FRAPS), and game setting expectations are different from one solution to the next.

We’ve been getting our share of graphics cards for game testing, and it got us thinking.  What’s stopping us from doing some benchmarking on MTBS?  Could we share information that other sites don’t?  Let’s find out!

Using both an Nvidia and AMD graphics card, combined with all three driver solutions, we wanted to determine:

1.    Can games be fairly benchmarked in stereoscopic 3D?
2.    What is a realistic game efficiency expectation in 3D?
3.    How does antialiasing impact S-3D game performance with the different driver solutions?
4.    Does deeper access to the graphics card directly impact performance?

To be fair, this article is more of a media experiment than a diverse collection of S-3D gaming performance results.  Similar to M3GA’s regimented rules, benchmarking should be based on a fixed platform that meets certain criteria.  It’s for this reason that we limited benchmarking to just two graphics cards and a handful of tests.  We want to get the process down pat, get the required benchmarking equipment, and then we will be able to follow through with a full-fledged service on MTBS.

Benchmarking Process


Under normal circumstances, we would use FRAPS to benchmark a selection of video games in 3D.  Unfortunately, FRAPS is not compatible with iZ3D’s stereoscopic 3D drivers – at least, not consistently.  The best way around this problem is to find games with benchmarks built in.

We chose Metro 2033, Resident Evil 5, Crysis Warhead, and Battleforge.  These are great selections because all driver developers have been aware of these titles for some time, and most have active profiles in their software offerings.

We wish to make it clear that this is not a test of quality – just performance.  So if one solution is a little faster than another with a given title, don’t jump to conclusions!  Like the tortoise and the hare, it’s important to pay equal if not more attention to the quality of the 3D gaming experience – and there can be big variations from one driver developer to the next!

This round of benchmarking is only focused on DirectX 9 performance.  DDD and iZ3D have just started supporting DX10 and DX11 in an official capacity, and we want all solutions to have adequate game profiles in these modes before testing.  We also limited ourselves to 1280 X 720 and 1920 X 1080 resolutions for testing.  Anything less or more would fall out of current generation display standards – especially in 3D.

We also turned off auto-convergence features and “virtual 3D” or “2D+depth” modes.  Auto-convergence cuts down on performance, and “Virtual 3D” skews frame rates because it doesn’t actually render a second camera view and display a complete 3D experience.

We tested for three things: first, we measured 2D performance with the S-3D drivers completely deactivated.  Then we tested each S-3D driver with multiple antialiasing options.  Finally, based on these results, we determined the S-3D efficiency relative to the 2D performance in each benchmark.


The Graphics Cards

Based on our current inventory, we tested the AMD HD 6870 and the Nvidia GTX285.  This article isn’t intended to be a comparison of two graphics cards.  We are more interested in seeing how the different drivers perform with two competing brands and architectures.

With the exception of DX11 support, the AMD 6870 and GTX 285 are directly competitive in the DX9 mode we are testing today.


The Test Platform

AMD Phenom 9850 X4 2.5Ghz
MSI K9A2 Platinum 790FX
4GB RAM
Windows 7 64 Bit
Zalman 24” 3D Monitor
Creative Labs XFI Soundcard

EVGA GTX285SC

The GTX285 (EVGA GTX285SC) is clocked at 675MHz.  It also features 1GB of DDR3 RAM clocked at 2,538Mhz.  It was tested with version 258.96 of the GeForce drivers because this is the last Zalman compatible 3D driver on record.  There is no evidence that later drivers would make a significant difference to the 2D or S-3D gaming performance.

AMD HD 6870

The AMD HD 6870 is factory clocked at 900Mhz and features 1GB DDR5 RAM clocked at 1050Mhz and was tested with Catalyst 10.10.

The stereoscopic 3D drivers included DDD’s TriDef Experience 4.3.2, iZ3D’s 1.12, and Nvidia’s 258.96 Zalman release.

Zalman 24

We acknowledge that Zalman is an interlaced solution (half vertical resolution per eye), and are open to the possibility that there could be a modest drop in performance with full resolution page flipping displays.  Future benchmarking will use a different platform.  We chose this route because there was no other way to test all driver solutions with a single benchmarking specification.

Resident Evil 5


While the Resident Evil 5 benchmark was released as a marketing tool to promote Nvidia’s GeForce 3D Vision compatibility, it runs equally well on all three S-3D driver solutions.  All settings were maxed out for the test.
Resident Evil 5 1280 X 720 FPS Resident Evil 5, FPS, 1920 X 1080

We were very surprised to see that Resident Evil 5 consistently performs better in stereoscopic 3D on DDD and iZ3D drivers compared to Nvidia’s GeForce 3D Vision on both the 6870 and GTX285.

On the GTX285, the iZ3D drivers maintained a modest two to three FPS advantage over DDD’s TriDef Experience – and this gap widens at 8AA or more.  In contrast, DDD did much better on the 6870 with an eight to twelve FPS lead over iZ3D.

Resident Evil 5, 1280 X 720, 3D Efficiency Resident Evil 5, 1920 X 1080, 3D Efficiency

While Resident Evil 5 was a great marketing push for Nvidia’s GeForce 3D Vision, it isn’t a great contender for demonstrating GPU performance in 3D.  Compared to 3D Vision’s competitors, it doesn’t once breach the 60% efficiency grade on the GTX285.

In contrast, the DDD and iZ3D drivers are able to reach anywhere from nearly 50% to almost 80% efficiency on the GTX285.  On the 6870, the DDD drivers are easily 10% more efficient than iZ3D’s.

Metro 2033

Metro 2033 has also been promoted as Nvidia GeForce 3D Vision certified, but the game and benchmark also works with the DDD and iZ3D stereoscopic 3D drivers.  All settings were maxed out, but we left PhysX off because that’s a proprietary feature unique to Nvidia.

Metro2033, 1280X720, FPS Metro2033, FPS, 1920X1080

Metro 2033 demonstrates a very modest performance advantage on the GTX285 with Nvidia GeForce 3D Vision drivers.  iZ3D is a close second, and DDD is third.

For some reason, to get the DDD drivers to run with Metro 2033 with the GTX285, we had to have the Nvidia 3D Vision drivers installed in the background.  Otherwise, it had a slideshow performance.  This caveat was unique to Metro 2033 and isn’t a trend.
Metro2033, 1280X720, Efficiency Metro2033, 1920X1080, Efficiency

Metro 2033 has a minor efficiency advantage with Nvidia GeForce 3D Vision drivers on the GTX285.  iZ3D’s efficiency rating did very well on the 6870 – even surpassing that of Nvidia’s at 1280X720 resolution.  DDD did ok, but given their need for the Nvidia 3D Vision drivers to be resident, there is probably room for more optimization.

Battleforge


A wildcard RTS game, Battleforge has no official recognition for being S-3D compliant.  However, Nvidia rates this title as “excellent” in their games list, and DDD already has a Battleforge profile.  iZ3D also has compatibility, and they have been informed that we were looking to review this title some time ago.

The only limitation is that Battleforge doesn’t have a 1280X720 mode, so we only tested at 1920X1080 resolution.  All settings were at maximum.

Battleforge, 1920X1080, FPS Battleforge, 1920X1080, Efficiency

From a performance point of view, Battleforge consistently ran best with the Nvidia GeForce 3D Vision drivers.  On the GTX285 graphics card, DDD and iZ3D remained neck and neck, but iZ3D inched ahead once antialiasing was set to 8X or more.

DDD and iZ3D performed better on the AMD 6870 than they did on the GTX285 both in frame rates and efficiency.  The performance gap between solutions was practically non-existant until they hit the 8AA setting. 

Crysis Warhead


While Crytek has been promoting Crysis 2 as supporting 3D without a drop in frames per second, it’s going to be a pseudo 2D+depth solution that doesn’t render a second camera view.

Crysis Warhead combined with DDD, iZ3D, and Nvidia stereoscopic 3D drivers does it the old fashioned way - complete left/right rendering.  This makes Warhead a great performance measuring tool.  We ran an external benchmarking utility because it let us go through a fixed testing series in batch form.

The only setting we turned off was shadows because it causes major problems on all three driver solutions.

Crysis Warhead, 1280X720, FPS Crysis Warhead, 1920X1080, FPS

At 1280X720, the GTX285 didn’t really have any clear performance winners.  iZ3D had a one or two frame advantage here and there, but that was all.  The 6870 had a similar result, though DDD held the lead this time around.

On the GTX285, while DDD and iZ3D had a minor performance advantage over Nvidia’s drivers at lower resolutions, this edge diminished at 1920X1080 as the antialiasing settings increased.

DDD seemed to have some trouble with antialiasing compared to iZ3D on Crysis Warhead with the 6870.  The moment antialiasing was activated, its performance dropped like a rock with each settings increase.

Crysis Warhead, 1280X720, Efficiency Crysis Warhead, 1920X1080, Efficiency

Nvidia’s GeForce 3D Vision drivers have average but consistent stereoscopic 3D efficiency with Crysis Warhead.  While they are easily outperformed in 1280X720 on the GTX285, their efficiency is unmatched at 1920X1080.  In contrast, DDD and iZ3D do well at the beginning of the race, but putter out at high resolution on the GTX285.

What is interesting is that while DDD and iZ3D performed better on the 6870, their efficiency grades both dropped at 2X AA.  They faced a modest drop of efficiency at 4X AA, and become nearly useless at 8X AA.

Conclusion


We think it’s fair to say that all stereoscopic 3D drivers have a competitive offering, and consumers should have confidence that they aren’t going to miss out on any major performance gaps.

While Nvidia was green with envy over their competitors’ scores with Resident Evil 5 (a “GeForce 3D Vision Ready” game), their 2D/3D performance was a little more efficient in three out of four titles tested on the GTX285.  AMD’s 6870 combined with DDD and iZ3D drivers proved to be an even match, however.

DDD and iZ3D had nearly equal performance in most cases, and tended to perform better with the AMD HD 6870.  iZ3D inched ahead of DDD’s results when antialiasing was set high enough.  We are taking an educated guess that iZ3D’s memory footprint is smaller than DDD’s when in 3D mode, so antialiasing impacts DDD’s performance sooner than it does iZ3D’s.  We would be interested to see if additional GPU memory would make a difference in this outcome for either graphics card maker.

It’s also interesting to point out that while 3D is often accused of cutting FPS in half, these tests show otherwise with efficiency scores ranging from 50% to 70% of the 2D equivalent.  In cases where efficiency dropped below the 50% mark and the FPS was on the slow side, it’s perfectly reasonable to expect decent game speed with some eye candy reductions on mid-range graphics card equipment.

All things being equal, measuring an extra frame here and there isn’t going to determine the winning combination.  Stereoscopic 3D game quality and visual experience is much more important.  Please visit MTBS' 3D Game Analyzer regularly to track that based on member submission.  You can also read our S-3D game reviews.

In conclusion, we plan to do more tests like this so consumers are informed of which GPUs will work best with their favorite drivers and games.