A few days ago I told you that I’ve managed to finally get my hands on GeForce GTX 580 – the new top model GPU from Nvidia, and I was quite happy by the fact that the full cover water cooling blocks from the older GTX 480 fit nicely on the GTX 580. This was enough to convince me to upgrade to 580s, so I used this opportunity to also do some testing and compare the performance of GTX 480 and GTX 580 in a single and dual configurations (SLI). Of course there are a lot of reviews out there that have already done that, but not quite the way I wanted to test and by that I mean to test in stereoscopic 3D mode in some of the more recent and popular and of course more demanding games. So I’ve chosen 5 different games and started testing, I was eager to try the GTX 580, that is why I didn’t go for a lot of games on one side and on the other, I was having some trouble finding games that do not max out at 60 frames per eye with SLI in stereo 3D mode.
Below you will find the results from the five games I tested with, including Metro 2033 which is still one of the heaviest games, especially if you want to play it with max details and in stereo 3D mode as you will see from the charts. There are four different test scenarios in the charts – single GTX 480, single GTX 580, dual GTX 480 in SLI and dual GTX 580 in SLI, with the idea that you can compare not only the single card performance, but also to see how it scales in dual GPU configuration. The results in frames per second listed in the charts are for each eye, with 60 fps being the maximum per eye, because of the Vsync required to be forced when in stereo 3D mode. So if you see 60 fps in the chart, that actually means 120 fps average framerate for both eyes and since you cannot go further, maxing out at 60 fps per eye average simply means that the system can pretty much supply even higher fps. All the tests were done in 1920×1080 resolution with maximum detail levels and some AA/AF enabled as the specific game supports and these settings are also mentioned in the charts graphs. You may notice that the difference between single card configurations and dual card configurations are quite different with the single cards showing much bigger difference at times. The reason for that is quite simple, because of the Vsync in stereo 3D mode the top framerate is capped, so the difference in performance becomes less apparent for games where the 60 fps per eye is very close to the average achieved. And that is pretty much clearly visible in the SLI results for the first 4 games, so you should use the Metro 2033 results for judging the performance in SLI and how well does dual GTX 480 scale as compared to dual GTX 580. Also do not forget the fact that 5 frames in the chart are actually 10 frames difference, because the chart lists only the framerate achieved per eye and the actual fps is doubled, because in stereo 3D mode both eyes see different frames.
I’m starting with Battefield: Bad Company 2, a game that is still a bit challenging for single card configurations in stereo 3D mode as you can see from the results, however with dual GPU configurations the average fps almost hits 60 fps per eye. And as I already mentioned this is not good for comparing the scaling of the dual GPU configurations, although you can compare the single vs dual results. The difference between GTX 480 and GTX 580 in terms of performance is about 18% and the difference between single GTX 480 and dual GTX 480s is almost hitting 50%.
Moving to Formula 1 2010, a quite demanding game for single GPU configurations in stereo 3D mode, but not a challenge for SLI setup even with GTX 480 as the game is hitting 60 fps per eye average in both SLI setups. And this pretty much means that with two GTX 480 or two GTX 580 in SLI the minimum framerate pretty much does not drop below 60 fps per eye, so comparing between these two in the F1 game is pretty much pointless as the difference cannot be measured in stereo 3D mode. In single GPU configurations the GTX 480 is just about 7% slower than the GTX 580.
The next game in line, Fallout: New Vegas does pretty much the same as F1 2010 in terms of results, just about 8% difference between the single GPU configurations in favor of the GTX 580 of course and almost 60 fps per eye average for both SLI configurations.
Mafia II is another quire demanding games that can be a tough nut to crack for a single GPU in stereo 3D mode, however when in SLI with dual GPUs it again almost reaches the 60 fps average per eye, which makes it hard to compare the difference between SLI setup with 480 and 580. Around 14% difference between GTX 480 and GTX 580 in favor of the later of course.
And as a final, the Metro 2033 results. As I’ve already mentioned the game can stress well enough even two GTX 580s in SLI setup, and it is even more demanding for single GPU configurations in stereo 3D mode. The average fps of a single GTX 480 can make the game not so comfortable to be played in stereo 3D mode at times, because of the framerate dropping to about 20 fps per eye at maximum detail levels of course and although the GTX 580 performs better, you better go for SLI setup for that game if you want to play it in stereo 3D mode with everything to the max. The single GTX 580 is about 18% faster than the GTX 480, but if you go from a single GTX 480 to dual GTX 480s in SLI you will be gaining not only more comfortable framerate at about 70% scaling. The situation from a single to dual GTX 580s is pretty much similar with about 59% improvement in the framerate when using two cards in SLI, and the GTX 480 SLI versus GTX 580 SLI provides about 10% difference in performance.
So depending on the game when playing in stereo 3D mode you can get between 7% and 18% or an average of about 13% faster performance from the five tested games between the GTX 480 and GTX 580, so there is actually a point in upgrading a single GTX 480 to GTX 580. The scalability that a single to dual GTX 480 or GTX 580 also shows is quite nice, so you can also consider going for a second GTX 480 card to get a SLI setup instead of replacing the single GTX 480 with a single GTX 580 with a more limited budget for upgrades. Of course if you don’t have a limited budget for upgrading, then going for two GeForce GTX 580s in SLI and slightly overclocking them should solve your problem even with Metro 2033 in stereo 3D mode and will make sure you are absolutely ready for the upcoming new Crysis 2 game. And now I’m going to overclock the dual GTX 580s to see how far I can go into upping the framerate in the Metro 2033 game over the above results that were achieved with all of the video cards running at their stock parameters and not being overclocked.
After finally managing to get the AMD HD3D Technology working with the Panasonic 3D HDTV I’ve moved to testing the 3DTV Play functionality from Nvidia, which did not have any trouble working with the 3D TV and making it work was just a simple matter of plugging in the TV to the PC… hopefully AMD will work that out too in the near future to make things work as simple as that and resolve all the issues they have, including providing support for Crossfire configuration with multiple GPUs. And since I did not have trouble making things work with 3DTV Play, I’ve moved to testing with games and as expected I’ve seen a few mostly minor issues that need to be fixed by Nvidia in order for the 3D HDTV support to be further improved. The test system I’ve used is with two GeForce GTX 480 video cards running in SLI (no problems with dual GPUs here), under Window 7 Pro 64-bit and with the latest drivers version 260.99 in order to ensure most up to date functionality and the best possible performance. All of the 20 recent games I’ve tried were set to maximum details with no AA by default, although I’ve also tried them with anti-aliasing if they had available in the game options, just to be sure, but the fps results mentioned below are with no AA. My goal was to find if the games work in the 1080p 24Hz 3D mode, so I’ve started with that, before also trying them out in 720p 60Hz 3D mode. And another thing I was interested in was if they worked in the 1080p 24Hz 3D mode were they comfortably playable in that mode that is originally designed for 3D movies and not for games. So below you can see what were my findings…
Here on the video you can see how the difference looks when playing a game in the 1080p 24Hz 3D mode with vsync enabled and disabled with Bioshock 2 used as an example. In the video it might not be that noticeable, but with vsync enabled the game is pretty much unplayable and disabling the vsync makes with more comfortable to be played at that mode, provided that you can get high enough average fps. You can also see some of the other minor issues I’ve encountered with pretty much only the problem with Darksiders being more serious and thus preventing the game to be played in stereo 3D mode. Another thing that needs to be worked on a bit is the optimization for 3DTV Play to utilize the graphic processors with their full potential, especially in multi-GPU configurations as stereo 3D at high resolution can be quite demanding and getting a good average framerate ensures the good experience. But now lets take a look at each of the twenty games I’ve tested with and see some of the specifics for each title, I do plan to try out some more games, but as it is a quite time consuming task it will take some time.
1. Avatar the Game
The game Avatar works without problems in 1080p 24Hz 3D mode and can be played pretty comfortable even with Vsync enabled, although you better disable it for a bit better experience. The problem however turned out to be the average framerate I was getting from the game which was around 30 fps (per eye) at 1080p mode with maximum details with both GPUs loaded to just about 48%. So definitely some performance optimizations can help get the user better experience from the game.
2. Battlefield: Bad Company 2
The 1080p 24Hz 3D mode worked just fine here too, but the game is playable only with vsync disabled. Bad Company 2 is quite a demanding game even when not played in stereo 3D mode, so with maximum details and no AA and HBAO disabled I’ve managed to get just an average of 35 fps with about 55% load on both GPUs.
3. BlazBlue: Calamity Trigger
This is one of the games that I could not make the 1080p 24Hz 3D mode work as the game tries to force 60Hz refresh rate by default, however there were no problems to play it in 720p 60Hz 3D mode. The game itself is not so demanding anyway, so the GPU load was quite low here.
4. Bioshock 2
Here the 1080p 24Hz 3D mode works just fine, but the game is unplayable with vsync enabled, disabling it however improves the situation a lot and as you can see from the video above with over 100 fps average the game is using both GPUs at maximum with them being loaded at over 95%.
The 1080p 24Hz 3D mode works just fine here too, however the performance the game can provide in that mode is a bit low as at times the framerate drops below 20 fps at some moments. There is no vsync control option in the game and the sync is off by default, but in order to get better experience higher performance is required, so you might need to compromise a bit with the level of details. Both GTX 480 GPUs in SLI were loaded to just about 50%, so some optimizations can definitely be done to improve the situation.
6. Civilization 5
In Civilization V the 1080p 24Hz 3d mode works just fine and with vsync enabled the game is quite playable, although scrolling the map can be a bit choppy, so disabling the vsync can also help here. An interesting issue I’ve noticed here was that the in-game cursor is flashing in the right eye, but only in the menus and not while you play the game. Similar issue with the cursor blinking in just one of the lenses and thus creating a bit uncomfortable feeling I’ve also noticed in a few more games I’ve also tried. And although it is not a serious issue, this needs to be addressed, you can get an idea of what I’m talking about from the video above where I’ve demonstrated the problem.
Here the game switches on in the 1080p 24Hz 3D mode, but there is some issue that is preventing it to be played, also present in the 720p 3D mode and even disabling the stereo 3D mode by pressing CTRL + T does not help. There are some strange gray lines passing over the screen and in order to play the game normally in non-3D mode you have to go to the control panel and completely disable the Stereoscopic 3D mode. It is important to note the fact that I’ve also tried Darksiders with a 120HZ 3D LCD and there it works just fine without the problem with the gray lines, I’ve included a short demonstration of the issue int he video above. I’ve also noticed the same issue with the blinking cursor as in CIV V here too, but the other issue is dfar more serious and needs to be addressed first.
8. Dead Rising 2
In that game the 1080p 24Hz 3D mode also works and the game is not very comfortable when playing with vsync enabled, so you better disable it if you want to play it in the higher resolution mode. The GPU utilization is over 60% for both video cards and most of the time the provided framerate is quite high (over 70-80 fps), but there are some moments when entering big open spaces with a lot of zombies when it drops down to less than 20 fps and surprisingly enough the GPU utilization is also drops to less than 20% at these times. The issue with the flickering mouse cursor in the right lens is also present in this game.
9. Fallout: New Vegas
In this game the 1080p 24Hz 3D mode also works, and with vsync enabled from the game it is kind of playable, but you get a feeling of inertia of your crosshair when panning around, so disabling the vsync again can improve things. You get about 35 fps average framerate per eye which can be a bit on the edge and both GPUs are loaded to just about 45%, so some performance optimizations can help here too.
10. F1 2010
In Formula 1 2010 the 1080p 24Hz 3D mode worked just fine and surprisingly enough the game is quite playable with vsync enabled, although you might want to disable it for better experience. Something interesting I’ve noticed is that with no AA enabled the game behaves weirdly, the performance when the 3D mode is activated drops significantly to just about 0-1-2 fps and the game becomes unplayable, disabling the 3D mode brings things back to normal framerate and enabling 4xAA resolves the issue.
11. Just Cause 2
Here the 1080p 24Hz 3D mode works, with vsync enabled there is again the issue when panning with the feeling of acceleration for the crosshair movement. With vsync disabled the performance is just around 35 fps per eye and the GPU load is 99% for the first and just 20% for the second GPU which makes it better, but for more comfort a higher fps is needed with vsync disabled. And some performance optimizations would help as well in getting better experience.
12. Kane & Lynch 2: Dog Days
The 1080p 24Hz 3D mode works, with vsync disabled the game it is quite comfortable to be played. Over 50 fps average with around 55% load on both GPUs.
13. Lara Croft and the Guardian of Light
The 1080p 24Hz 3D mode works just fine, the game is played better with vsync disabled, although you can still play with it enabled at 24Hz. This title is not very demanding, so there are no issues with the performance of the SLI setup.
14. Metro 2033
Quite a demanding game as you should know, the 1080p 24Hz 3D mode works, vsync is disabled by default, no in-game option for controlling it. With DX11 mode, very high details, no AA gets you just 20 fps average (per eye), around 53% load on the GPUs, not very playable in that situation, so had to lower the details to get around 40 fps average which makes it playable.
15. Need for Speed: SHIFT
This is another game that I could not make the 1080p 24Hz 3D mode work, although there is an option for selecting the right resolution and refresh rate int he game options.However the game runs at 60Hz no matter that you’ve set it at 24Hz from the in-game options, after restarting it to apply the change it is reverting to 1080p 60Hz automatically. Running it at 720p 60Hz 3D mode and with 8xAA gives out an interesting warning window that the AA will be downgraded to 4xAA to ensure good performance in the game (recorded it in the video above), did not have trouble with other titles where AA was set to 8x or even 16x like the message in NFS. The average fps in this situation is around 37 with both GPUs loaded to around 53%.
The 1080p 24Hz 3D mode does not work here, when switching to 1080p resolution from within the game it goes to 50Hz, no matter if the desktop is set to 24Hz 3D mode or not. 720p mode works just fine, but again at 50Hz only. At 720p 50Hz, 4xAA the game runs with an average of about 40 fps and both GPUs are loaded to just around 30%.
The 1080p 24Hz 3D mode works, and at 24Hz vsynced the game is playable, but looking around creates a kind of strange feeling, like there is some kind of inertia when you are moving your crosshair. Disabling vsync from the in-game menu makes it much more comfortable to play at 1080p 24Hz 3D mode. Around 60% load with vsync off for both GPUs and fps of over 100, with vsync enabled the load on both GPUs gets to just about 10%.
18. StarCraft 2: Wings of Liberty
In that quite popular title the 1080p 24Hz 3D mode works just fine and with vsync enabled the scrolling of the map is not so good, but with vsync disabled is it much better and the game is quite playable. Over 60 fps average with both GPUs loaded to around 50%.
19. Transformers: War for Cybertron
Here the 1080p 24Hz 3D mode works as well and the the game is playable with vsync enabled, when vsync is disabled it rises just to 30 fps, because the game is capped at 30 max fps.
The 1080p 24Hz 3D mode works, vsync is off by default and there is no option to enable it. The game works very well with over 140 fps average framerate.
So out of 20 tested games, just a single one is unplayable due to some weird behavior, two games are completely playable, but have some issues when using AA, some others have minor issues. The 1080p 24Hz 3D mode did not work in only three games out of 20 and the reason for that is something within the games that forcing higher refresh rate. If you want to play at 1080p 24Hz 3D mode you should disable the vsync and be sure to have a high average fps in order to make the experience better. And finally some performance optimizations would also help a lot, especially with more demanding games and for multi-GPU configurations.
The new Nvidia Fermi-based GeForce GTX 470 and GTX 480 GPUs were officially announced a few days ago, but we still don’t have them on the market. Currently one a selected group of reviewers are being able to play with video cards based on the new GPUs, run benchmarks and even try them with 3D Vision, something that I sure most of you are quite interested in. Unfortunately I still haven’t had the chance to personally try a GeForce GTX 470 or GTX 480 video card with 3D Vision, so I can’t tell you much either and this leaves us with only the official results coming from Nvidia. The chart above shows a comparison between a GTX 285 (the former top model) and the GTX 480 (the new top model) running at 1920×1080 resolution with High Detail levels and even with some AA. The results coming from Nvidia are measured in frames per second (the scale on the left) across multiple new and popular game titles like Resident Evil 5, Battlefield Bad Company 2, Left4Dead 2 and Batman: Arkham Asylum. And based on these results (and some more perhaps) Nvidia is claiming that the new GTX 480 is up to 2x faster than the GTX 285, and that on Full HD resolution, with maximum detail levels and even with some Anti-Aliasing active. So if we trust these results you can say that the new Fermi-based video cards (the GTX 470 should be a good choice too, especially if you are on tighter budget) are the perfect choice for gaming in stereoscopic 3D mode with 3D Vision, so now we just have to wait a bit more for them to be available on the market…
With the introduction of the Fermi GPU architecture comes another interesting new feature that will interest some of the 3D Vision owners. Of course I’m talking about the 3D Vision Surround technology that adds support for stereoscopic 3D gaming on triple monitor setups. That however requires you to have two GeForce GTX 470/480 GPUs in SLI configuration for optimum performance, but it should also work with two GeForce GTX 200-series video cards again in SLI configuration. The reason for needing more GPUs is because 3D Vision Surround requires up to 746 million pixels per second of rendering horsepower and that is one hell of a burden for a single GPU to cope up with and still provide good enough framerate for comfortable playback. You should have in mind that in the currently available release 197 of the video drivers there is not yet support for 3D Vision Surround and you’ll be able to play in stereo 3D only on a single monitor. 3D Vision Surround will be enabled in the next release 256 of the drivers, that is due for release in April 2010, so a few more days of wait… but with no cards on the market yet that should not be an issue, right? And it seems that with the next official driver for the Fermi-based cards we are also going to have support for games using DirectX 11, so we’ll be able to play those in stereo 3D mode without having to revert to DX10 or DX9 anymore… ;)