• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

3D Vision Surround & 5760x1080 benchmarked

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

funsoul

Senior Member
Joined
May 3, 2004
Location
NJ, USA
Hi again folks!

Recently re-built my gaming pc and thought it'd be interesting to finally take an in-depth look at 3D Vision Surround and Widescreen gaming. In particular, wanted to test the following:

- impact of AA/quality settings on overall performance
- pcie 2.0 vs. pcie 3.0
- sli scaling between 2x and 3x 680's
- the penalty associated with using 3D Vision Surround

The short story is that if your motherboard supports pcie 3.0, you should definitely enable it as, on average, it provides a 4% 'free' performance gain. Bear in mind that this only translated to +2.6fps but could be improved with additional overclocking of the cpu, ram and graphic cards.

Going from two to three vga cards increases performance by 33% and enables running higher quality settings at good framerates. Although even two 680's delivered an average of 60 fps at maximum settings (excluding tweaks made through Nvidia Control Panel), adding a third card takes performance from the fringe to more butter-like even at maximum, tweaked settings.

If you're thinking of going 3D Vision Surround, be prepared for the ~50% performance penalty. 3D Vision Surround will take as much muscle as you can throw at it. Even with three gtx 680's, running games with all settings maxed out just isn't happening. Keep that in mind when planning for a new system. Haven't done the research or testing but anyone who says they want a 3D setup 'that can run everything at ultimate, maximum settings', best be ready to shell out the bucks for 3-4 Titans (if anyone has a few, spare Titans they want to send, I'll be happy to test them out if I can keep them hehe). It'll be interesting to see how long it takes for 2-3 mainstream graphic cards to deliver reasonable fps at 5760x1080 with good quality settings and 3d enabled. Also, when looking at non-3D game results, estimate a penalty of 51% or 54% (for 2x or 3x sli, respectively) to get the expected 3D Vision Surround results.

All that said, let's move to the actual testing and results. If you find anything wrong or messed up in my post, please let me know so the issue(s) can be fixed. tia!

Testbed:
i7 3930k @ 4600 GHz
asus rampage iv formula
16g gskill ddr3 2133
samsung 830 256gb ssd
2x 1TB wd black RAID0
3x zotac gtx 680 4gb @ 1202 MHz
3x asus vg278h
cpu/vga watercooling

cpu-z.jpg

PCIE 2.0 and enabling PCIE 3.0:
Although the RIVF supports pcie 3.0, enabling it requires edits to the registry. A bit of a pita (and the settings reset whenever the drivers are re-installed) but the fix works without any issues.

gpu-z.jpg

Benchmarks:
Aliens vs. Predator http://downloads.guru3d.com/Aliens-vs.-Predator-DirectX-11-Benchmark-Tool-download-2553.html
Bioshock Infinite http://www.bioshockinfinite.com/
Lost Planet 2 http://www.techpowerup.com/downloads/1854/lost-planet-2-benchmark/
S.T.A.L.K.E.R. Call of Pripyat http://downloads.guru3d.com/S.T.A.L.K.E.R-Call-of-Pripyat-benchmark-download-2433.html
Sniper Elite 2 http://downloads.guru3d.com/Sniper-Elite-V2-Benchmark-download-2935.html
Tomb Raider http://www.tombraider.com/
Unigine Heaven 4 http://unigine.com/products/heaven/
Unigine Valley http://unigine.com/products/valley/
Unigine Sanctuary http://unigine.com/products/sanctuary/

OS, drivers, tools, etc.:
Adrenaline Action Benchmark Tool http://benchzone.adrenaline.uol.com.../adrenaline-sleeping-dogs-benchmark-tool.html
AVG anti-virus http://free.avg.com/
FRAPS http://www.fraps.com/download.php
MSI Afterburner http://event.msi.com/vga/afterburner/download.htm
Nvidia GeForce 320.49 http://www.nvidia.com/object/win8-win7-winvista-64bit-320.49-whql-driver.html
Windows 7 SP1 64-bit http://www.microsoft.com


General testing notes:
Unless otherwise noted, all tests were run a minimum of three times to generate each data point and the results were those displayed by the benchmark or tool itself. After compiling the results, outliers and other data points that appeared unreasonable were fully re-tested. If the system became unstable, the Nvidia drivers were re-installed. As/when a driver re-installation did not help, the build was restored using an image created during the initial setup process. The build used included only windows, drivers, the apps/games/tools (including steam) and anti-virus. The general quality settings used include; no AA, 2xAA, maxAA and the maximum settings available via Nvidia Control Panel (NVmax). For a number of applications, 3D Vision Surround limits AA to a maximum of 2x. In those situations, the maxAA results were interpolated using a straight line. Maximum values were used for all other available settings.

The NVmax settings applied include:

NVCPSettings.jpg

The results shown here are based only on the hardware described above and ymmv.

Testing and Results:
Unigine Heaven 4

heaven4-screenshot.jpg

For testing, all Heaven 4 settings, except AA, were kept at their maximum values. As noted above, each test ran three times to generate the data points. In the case of any re-testing, the application was again run three times. MaxAA results were interpolated using a straight line. The Heaven 4 results, like the other Unigine apps included here, were very stable with minimal variance across test runs.

Heaven4-Main.jpg

Heaven4-charts.jpg

Heaven4-sub.jpg

As expected, Heaven 4 absolutely crushed the test system. No single test point shows an actual 'playable' setting. This benchmarking app exhibits well above average (12%) gains in moving from pcie 2.0 to pcie 3.0. As shown above, the pcie3/2 gains are most pronounced in tri-sli (16%) particularly in 3D mode (19%). Sli scaling is also above average at 48% while the 3D Vision Surround penalty is slightly above average at 51%.

Unigine Sanctuary

sanctuary-screenshot.jpg

As with the other Unigine benchmarks, all settings, except AA, were kept at their maximum values. Each test ran three times to generate the data points. In the case of any re-testing, the application was again run three times. MaxAA results were interpolated using a straight line. The Sanctuary results, like the other Unigine apps included here, were very stable with minimal variance across test runs.

Sanctuary-main.jpg

Sanctuary-charts.jpg

Sanctuary-sub.jpg

In contrast to the Heaven 4 results, the test system delivers sufficient horsepower to yield playable results at every setting for non-3D. With 3D enabled, the third card provides the pixels to run the app at good fps. With regard to pcie 3.0 vs. 2.0, the gains are higher using 3 graphic cards. Sli scaling is above average with maximum benefit at pcie3 and 3D off. Sanctuary has the lowest 3D Vision Surround penalty of any application included in this testing (-35%).

Unigine Valley

valley-screenshot.jpg

All settings, except AA, were kept at their maximum values. Each test ran three times to generate the data points. In the case of any re-testing, the application was again run three times. MaxAA results were interpolated using a straight line. The Valley results, like the other Unigine apps included here, were very stable with minimal variance across test runs. Subjectively, Valley looks absolutely gorgeous in 3D.

Valley-main.jpg

Valley-charts.jpg

Valley-sub.jpg

Valley's another benchmark that brings my system to its knees. The results are very similar to those for Heaven 4. Above average gains from moving to pcie 3.0 with the largest benefit coming from pcie3 with 3D enabled (17%). Valley shows the highest sli scaling of anything tested here at 61.5% when running at pcie3. The 3D penalty is slightly above average at 53%.

Aliens vs. Predator

avp-screenshot.jpg

Perhaps not the best benchmark to include here as there are no user-adjustable settings. AVP testing was run at default and NVmax settings only.

avp-main-replacement.jpg

AVP-charts.jpg

AVP-sub.jpg

Another game that can run anything non-3D using two cards but requires a third card to generate playable results in 3D. Running three cards yields the greatest percentage gains with pcie3 enabled (8% vs. 3% under pcie2) particularly with 3D disabled (9.5%). AVP has the best sli scaling of any games included here at 35% and although poor in the big picture of things, the third 680 still gives the system enough oomph to run maximum settings in 3D at reasonable levels. The 3D penalty is the expected 50%.

Bioshock Infinite

BioShockInfinite-screenshot.jpg

Pretty cool new game (and it looks nice, too). Testing used the Adrenaline Action Benchmark tool.

bi-main-replacement.jpg

BioshockInfinite-charts.jpg

BioshockInfinite-sub.jpg

For people not running 3d, two 680's can deliver playable framerates at the standard 'High' setting while higher settings requires three cards. For 3D Vision Surround users, get ready for Low settings with some tweaking (NVCP and otherwise) even with three 680's. Once again, the best pcie3 vs. 2 improvement uses the tri-sli configuration. The 3D penalty is worse than average at 53%.

Lost Planet 2

lostplanet2-screenshot.jpg

Pretty fun game and 3D looks better than Nvidia's 'Fair' rating. Each test (A and B) within the benchmarking app ran once. Data points were generated from MSI Afterburner log files.

lp2-main-replacement.jpg

LostPlanet2-charts.jpg

LostPlanet2-sub.jpg

Lost Planet 2 is another game that the 2x-sli, non-3D folks can expect excellent results at all but the best (NVmax) setting but where 3D users get horribly punished. Not sure what's up with the in app quality settings but this game shows the worst performance hit (-49%) going from the max in-game settings to the NVmax setting in non-3D. Pcie3 vs. 2 gains, sli scaling (vs. the other games tested) and 3D penalty were all in line with the averages.

Sniper Elite 2

sniperelite2-screenshot.jpg

Another benchmark with no user-adjustable settings.

SniperElite2-main.jpg

SniperElite2-charts.jpg

SniperElite2-sub.jpg

This games appears to be completely un-playable at 5760x1080 with tri-sli 680's. The three card, non-3D settings showed the best Pcie3 vs. 2 performance improvements. This game scales horrendously (23%) when going from two to three graphic cards. An above average 3D penalty just puts the cherry on top of this one.

S.T.A.L.K.E.R. Call of Pripyat

stalkercop-screenshot.jpg

Ran through the standard benchmark once for each data point and compiled the results from the MSI Afterburner log files. This one gave me some stability issues at the NVmax settings.

cop-main-replacement.jpg

STALKERCOP-charts.jpg

STALKERCOP-sub.jpg

Not a lot of surprises here as the results are essentially average. 3D Vision Surround users will need more graphics horsepower to run this one at decent settings. Non-3D users will be fine up to 2xAA (plus a little tweaking with 2 680's or equivalent) or full out maxed with three 680's or better.

Tomb Raider

tombraider-screenshot.jpg

One of the most boring benchmarks ever seen. That said, some 3D users will undoubtedly smile when they see Lara's swelling breasts.

tr-main-replacement.jpg

TombRaider-charts.jpg

TombRaider-sub.jpg

Another game that shows more or less average numbers and some despair for 3D folks. Interesting that with 3D off, sli scaling reaches a very solid 47%.

Composite Results:
The composites below are comprised of each game and/or application in equal portions unless noted otherwise.

Unigine Composite

Unigine-main.jpg

Unigine-charts.jpg

Unigine-sub.jpg

Comprised of Heaven 4, Sanctuary and Valley benchmarks. The average framerates in the table are the average of the results for each application and test point. Tri-sli delivered the best pcie3/2 gain, pcie3 the best sli scaling. The 3D penalty reduces significantly at the maxAA setting.

Games Composite
Comprised of the six games. To accommodate variability in quality settings, a simple 1 through 3 plus NVmax quality setting scale was used. As there were only pairs of results for Aliens vs. Predator and Sniper Elite 2, those results were only included in the lowest quality setting (1) and NVmax figures.

Games-main.jpg

Games-charts.jpg

Games-sub.jpg

Compared with the Unigine Composite results, the Games Composite shows 50% worse sli scaling and a 30% worse 3D penalty. Am not sure but it makes sense to reason that at least part of the reason for the difference is that the folks at Unigine have put a lot more time and effort into tweaking their code for sli scaling and 3D than the game developers. Hopefully the situation improves over time but I'd guess multi-card scaling will get better before 3D support does (based on the niche nature of 3D atm). For now, though 3D Vision Surround gaming at full maximum settings is pretty much out of the question with the hardware equivalent to that used for this testing and may not be attainable with the current generation of hardware with the possible exception of quad Titans.

Overall Results and Conclusion:

Again, to accommodate variability in quality settings a simple 1 through 3 plus NVmax quality setting scale was used.

Composite-main.jpg

Composite-charts.jpg

Composite-sub.jpg

To wrap it all up, the average expected gain in going from pcie 2.0 to pcie 3.0 is 4% (2.6fps based on what's in the composite). It's not worth upgrading an entire platform but if you can get it working with your current hardware, definitely do so. Tri-sli configurations, particularly with 3D disabled, reap the most performance rewards.

Regarding sli-scaling, the law of diminishing returns definitely comes into play so it's up to people's budgets how far they want to go. Running 3 or more cards provides very nice bang for the buck in that it can allow for higher quality settings and/or make the difference between playable and un-playable at a given setting. Hopefully driver updates and game coding levels improve over time to extract better scaling. Again, tri-(or more)sli configurations running with 3d off at pcie 3.0 gain the most benefits.

For 3D Vision Surround users, the huge 3D penalty can be wallet and/or experience killing if one wants to play with reasonably high graphic settings. That said, my normal gaming experiences are much better than the numbers and quality settings suggest and, imho, is outstanding from an immersion perspective. If you're planning a build from scratch, you may be best off waiting for the next generation of cards. It's interesting that the penalty goes down the higher the AA/quality setting (from -53% at the lowest quality setting to -44% at NVmax).

Best Regards,
-funsoul

PS- As an aside, if you plan on heavy, mmo and/or highly modded widescreen gaming, particularly if running 3D Vision Surround, I'd recommend cards with a minimum of 3GB gddr.
 
Last edited:

manu2b

Member
Joined
Feb 6, 2011
Thanks for the huge amount of time spent Funsoul.

And for the accuracy of the results of course!

Great work:clap:
 

MattNo5ss

5up3r m0d3r4t0r
Joined
Aug 11, 2008
You know, this kind of stuff makes good frontpage articles :chair:

Thanks for all the time spent testing, and good work :thup:
 
OP
funsoul

funsoul

Senior Member
Joined
May 3, 2004
Location
NJ, USA
This is an absolutely massive amount of testing, thanks for the report!

wow
great job

Thanks for the huge amount of time spent Funsoul.

And for the accuracy of the results of course!

Great work:clap:

You know, this kind of stuff makes good frontpage articles :chair:

Thanks for all the time spent testing, and good work :thup:

Thanks ya'll! Glad I could give something back. :) (I'll fix the 4 charts with cosmetic errors tomorrow. update- charts now fixed)
 
Last edited:

carlosg

New Member
Joined
Sep 15, 2013
Thanks for posting + number verification

FunSoul:

Finally, after scouring the interweb for 5780 X 1080 benchmarks for hours I managed to stumble into your post. This is the most comprehensive test results I've seen, and I salute you for putting so much time and effort into doing this.

However, I do have one question regarding your results:

I have two AMD 7970's in crossfire mode. No 3D. PCIe 2 (it's the older Rampage III motherboard). My Samsung triple monitor set up has its resolution at 5800 X 1080 (bezel compensation). I have an Intel 980x @ 3.33GHz and 24GB of DDR3 Ram. So nothing fancy here. I ran the Heaven 4.0 test with everything maxed out (including AA) and got average FPS of 19. On the other hand your 'interpolated' Max AA at the 'comparable' 2x SLI Pcie 2 no 3D clocks in at 14fps according to your table... which seems odd because from everything I've heard 2x 680 SLI should be roughly equivalent to 2 x 7970 crossfire... where as this implies a 35% performance premium for AMD setup . So unless the recent 13.10 catalyst control center beta driver update I did this weekend did wonders for us AMD GPU users (which I seriously doubt) - perhaps I'm missing something here or may be your interpolated max AA for Heaven 4 over-penalized your FPS? Will post my results in a bit so you can see what happened in my case...
 
OP
funsoul

funsoul

Senior Member
Joined
May 3, 2004
Location
NJ, USA
hi carlosg!

Welcome to OCF! :welcome:

First off...thanks for your kind comments...definitely took a while to get through everything.

Hmmm...the 14fps is the actual test results as displayed by Heaven4 (ran the test 3 times and took the averages). The only interpolated values were the 3D max AA ones. Your 19fps is excellent and definitely surprising that the result is somewhere between my 'no AA' and '2x AA' results.

Am at work so don't have all the info in front of me but guess the first thing to confirm is that you ran the test with all available Heaven4 settings maxed (dx11, ultra quality, extreme tesselation, max AA).

Are your 7970's overclocked, at all? If so, how much? I ran all the tests with my cards slightly overclocked. Agreed, your results are definitely better than expected! It's possible that at certain settings the 7970's outperform the 680's but 35% seems a bit out of line (particularly given the difference in our cpus/speeds).

Anyone else have any thoughts on this?
 
OP
funsoul

funsoul

Senior Member
Joined
May 3, 2004
Location
NJ, USA
Hiya EarthDog

Hmmm...no idea. The 7970 has 3GB on a 384bit bus, my cards have 4GB on a 256bit bus. Could that really be the source of the fps difference? I can test and measure but am still pretty clueless as to what difference the bus speed would make at that resolution. Is it possible that my bus was saturated or something?
 

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
Potentially, yes. That could be the source.

Adding AA and such, and I imagine 3D, is not helping. You would want all the bandwidth and vRAM you can get with 3D Surround and AA. More vram is fine, but if its limited by the bus, that can be a reason it is slower, sure. Generally speaking, we all say, for multi-monitor, AMD due to its 384bit bus as well as the 3GB it offers (microstutter issues not-withstanding).

If that is the case, overclocking the ram should help quite a bit (versus when you are not limited and memory overclocking doesn't matter as much).
 
Last edited:
OP
funsoul

funsoul

Senior Member
Joined
May 3, 2004
Location
NJ, USA
I'll overclock the memory a bit and see what it looks like. Will go 3x sli, pcie3 as that's where my setup's at atm. If I see a non-negligible delta, will try at 2x sli, pcie2.
 
OP
funsoul

funsoul

Senior Member
Joined
May 3, 2004
Location
NJ, USA
Hi ATMINSIDE

Don't think so...the 2xsli/pcie2 numbers look consistent with the rest of the Heaven4 results.
 

ATMINSIDE

Sim Racing Aficionado Co-Owner
Joined
Jun 28, 2012
Just making sure because the "SLI 3x/2x gain %" has a legend of "PCIe 2 3D off"
While "pcie3/2 performance gain %" has a legend of "2x SLI 3D off"

Thought that part might be backward.
 

carlosg

New Member
Joined
Sep 15, 2013
Okay, now that I had a bit more time to run the test again - I've gotten a completely different set of numbers. The frame rate went up to 34.7FPS. Not sure why performance has changed so drastically given that I never changed any GPU or system settings. I ran this several times today just to be sure and I kept getting the same benchmark results (34.1-34.9fps avg). See the screenshots below:


But yes, to answer your question my x-fire 7970's are overclocked to 1115MHz and 1600MHz - though it shouldn't cause this much of a difference...
 

Attachments

  • Techpowerup GPUz.gif
    Techpowerup GPUz.gif
    23.1 KB · Views: 726
  • Heaven 4.0 8aa.jpg
    Heaven 4.0 8aa.jpg
    87.8 KB · Views: 700
OP
funsoul

funsoul

Senior Member
Joined
May 3, 2004
Location
NJ, USA
Wow. Let me take a look when I get home tonight. You definitely ran it with all heaven settings maxed?

ATM...will also double-check the items you raised.
 
OP
funsoul

funsoul

Senior Member
Joined
May 3, 2004
Location
NJ, USA
Hmmm...took a look...my 2xsli, pcie2 test results were real tight:
min max avg
8.2 30.2 14.3
8.6 30.0 14.3
8.7 29.9 14.3
8.2 30.2 14.3

As EarthDog raised, the delta between our results could be a memory bus issue but take a look at hwbot.org and look up the Heaven4 Xtreme preset scores. The 7970's whip 680 butt....HARD in Heaven. The best 2x 680 score I could find is 135th place. So....that's the reason you're fps is so much higher on that one...it's a much higher performing card at Heaven. Not sure but can all that difference be explained by the different bus width? My feeling is no...that there must be more to that story...but I've really no clue.

Did try out playing with the gpu and memory overclock a little and (of course) EarthDog was spot on...largest gain came from memory overclocking. Here's what I got with 3x sli, pcie-3:

heaven-gddr-gpu-test-3xsli-pcie3.jpg

Please note that these results are NOT compatible with the main test results as the cpu is now running at 4GHz (not 4.6GHz) and the o/s is loaded with other junk now.

Think I'll keep running the cards at 1274/1654 for those couple fps...they all add up ;) Will give it another run with the cpu @ 4.6.
 
Last edited:

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
The reason why they are pounding on them at hwbot is because the AMD cards can turn off tessellation while the nvidia cards cannot. That is a HUGE difference there.
 

MattNo5ss

5up3r m0d3r4t0r
Joined
Aug 11, 2008
HWBot uses Heaven 2.1, not Heaven 4. Plus, AMD users typically disable tessellation within CCC for Heaven to boost results, so that's a major factor in AMD vs NVIDIA Heaven results on HWBot.

EDIT: ninja'd by 7 minutes... guess I should have refreshed the page :rolleyes: