DX9 vs DX10: World in Conflict

[Stormcrow]

It seems like DX10 is under full steam, with Bioshock already out and lagging system and World in Conflict next on the agenda. For those of us lucky enough to have a video card with a GPU powerful enough to render the new Direct3D monster, the WIC demo has included with it perhaps one of the most comprehensive built in benchmarking sessions seen in a demo or even a full scale game, allowing complete free range control over every aspect of the graphical goodness.
Of course, being of the graphically inclined sort, I just had to get my hands on it and start running some benchies with my G80. The results are both, shocking and disappointing, especially considering the resolution and setup but physics have never looked so good. But the good news is that WIC allows you to switch between DX9 and DX10 rendering modes, a truly usefully tool for getting a real in game look at the performance of these new gen cards in DX10 as compared to DX9 which they dominate.

The demo itself runs about a minute and a half, but completely and utterly covers all bases for one looking to stress a system. If you want a DX10 benchmark that will take your system to it's max, this is it. Complex shading, details, extreme physics from flying debris and dust, and some of the most realistic lighting and shadows to date. Not to mention a spectacular nuclear explosion that pull even the best system to it's knees.

The Guinea Pig
Athlon X2 3800+ at 2.0Ghz
2x1GB DDR2 533 Kingston modules in Dual Channel
Evga GeForce 8800GTS 640mb stock
160GB Western Digital 7200RPM 8mb cache HDD
Windows Vista Premium 32-bit

Drivers:
Nvidia Forceware 163.44 Beta
AMD Dual Core Optimizer

Native monitor resolution of 1026x768

Since the resolution is set so low, you'll seriously need to consider that when running a display at 1600x1200.

The level of detail in the graphics options is astounding, you can choose to keep grass but disable bushes, change the time tracks and craters stay on the ground, even set the texture resolution of the water details. Captured below on max.

For the benchmarks, I selected everything at the highest for Max, and used ingame level adjustments when setting the quality lower. Note that only High and Very High use DX10 rendering standard, Medium and lower and the game defaults to a less intensive DX9 unless you check the box otherwise. Those with DX9 hardware can manually setup their graphics to match the high levels without DX10 Rendering.

DX10 Benchmarks:

Maxed:
Average-7
Minimum-3
Maximum-15

Very High:
Average-7
Minimum-4
Maximum-21

Very High No AA/AF:
Average-8
Minimum-5
Maximum-22

Medium:
Average-12
Minimum-5
Maximum-35

Low:
Average-25
Minimum-9
Maximum-75

Very Low:
Average-40
Minimum-23
Maximum-95

Antialiasing and Antriscopic Filtering seem to have little impact on performance, but are greatly noticeable in game. The minimum drops always occur when the nuke goes off, regardless of your system when the bomb drops so does your FPS. The highest was always when the B-52's come in, right before they deliver their payload which causes the second lowest drop in FPS due to the intense physics calculations.

DX9 Benchmarks:

Maxed:
Average-10
Minimum-5
Maximum-27

Very High:
Average-10
Minimum-5
Maximum-27

Very High No AA/AF:
Average-10
Minimum-5
Maximum-23

Medium:
Average-19
Minimum-9
Maximum-52

Low:
Average-32
Minimum-7
Maximum-77

Very Low:
Average-50
Minimum-25
Maximum-109

The highs show some wonky scores, which leads me to believe that perhaps the demo doesn't render beyond 4x AA/AF for DX9. Either way, the scores are noticeably higher, especially in the Lower end segment. The differences between DX10 and DX9 are hard to capture through images, but in game you will notice how flying debris behaves, lighting and shadows, the smoke and dust clouds, and texturing in general. Needless to say, one needs to experience the demo itself to draw conclusions on differences from a graphical standpoint.

Since the benchmark is extremely intensive and not what you'll be seeing ingame very often, I decided to run a quick check in the tutorial to see how it plays. Settings to Very High on DX10 mode, during the part you can call in infinite tactical aid.

Doing nothing: 37-40
Panning camera, moving camera: 32-35
Commanding units across the field: 31-32
Laser Guided missile exploding a building: 26-28
Building exploding, zoomed in close: 16-23
3 Heavy Artillery barrages at once: 21-24

The X2 3800+ is no slouch, but with an aging architecture and stock speed of 2.0GHz is not the hallmark of multicore gaming either. Quakindude ran the same benchmarks as well with the same graphics card but with a nicely overclocked E6600, and the results come out even more interesting.

E6600 at 3.37GHZ
2x1GB Corsair Dominator DDR2 1066
Evga GeForce 8800GTS 640mb

DX10 Benchmarks

1024x768 - 1920x1200

Maxed:
Average- 21 - 15
Minimum- 12 - 8
Maximum- 38 - 30

Very High:
Average- 24 - 18
Minimum- 14 - 9
Maximum- 50 - 36

Very High No AA/AF:
Average- 31 - 22
Minimum- 13 - 10
Maximum- 82 - 43

Medium: Defaults to DX9 settings
Average- 52 - 44
Minimum- 25 - 25
Maximum- 126 - 76

Low:
Average- 82 - 80
Minimum- 27 - 11
Maximum- 178 - 186

Very Low:
Average- 116 - 103
Minimum- 22 - 20
Maximum- 228 - 234



DX9 Benchmarks

1024x768 - 1920x1200

Maxed:
Average- 35 - 24
Minimum- 17 - 14
Maximum- 81 - 48

Very High:
Average- 36 - 25
Minimum- 15 - 14
Maximum- 84 - 54

Very High No AA/AF:
Average- 36 - 30
Minimum- 15 - 15
Maximum- 84 - 64

Medium:
Average- 59 - 42
Minimum- 30 - 22
Maximum- 139 - 74

Low:

QD didn't run this one.


Very Low:
Average- 126 - 112
Minimum- 72 - 39
Maximum- 248 - 248


This test basically sums up my concerns about the X2 3800+ bottlenecking in higher performance, as even maxed at 1920x1600 with his setup still outperforms the same test at 1028x768 with mine. This means several things; as stated above even with two cores a slower clocked processor is going to bottleneck current multi-threaded games, a good Core2 is currently king of the gaming industry, and Nvidia's Quantum Effects technology still has a ways to go to completely take over the job of a Physics Processing Unit from the CPU.

So with all the game just about maxed for settings, with 4x AA/AF at 1026x768 with a modest setup built around a great GPU, we're looking at an average of 30 FPS. Since even dips to the teens hardly affected performance, it seems those of us with 8800s/2900s at resolutions around 1280x1026 will enjoy all the eyecandy. Those with higher resolutions up to 1920x1600 can enjoy most of the graphical goodness provided the processor and GPU are up to the task. As WIC is obviously very CPU demanding, even with a high end GPU a sluggish processor will force you to tone back on the settings to keep playable FPS. We're coming to the point now where even with two cores a slower clocked processor will handicap a game and GPU's full potential. As evident with the overclocked E6600, those with single cores or slower speeds may be reduced to running at lower graphical settings even if sporting a G80 or R600.

For those of you still chugging along with a 7900 or X1950, I'm betting you'll still have most of the settings available to you in DX9 mode, and probably won't even miss the added eyecandy since DX10 seems mostly to render physics and lighting effects more intensively. 8600/2600 users will get away with running WIC on High in DX10 mode, limited to lower resolutions and provided their CPU is strong enough. Medium settings might be best for those of us with slacking processors or weaker cards such as the 7600GT. All in all WIC demands a great deal from the system it's running on, but also adapts very easily to performance with lower end systems.

Since the demo is free and allows such control over what options to test, I highly recommend it to those of you wanting to give DX10 a whirl or needing to see whether to upgrade that DX9 card or hold off. The download takes about 4 hours, but if you can spare the time and want a better deal of control then just running through Bioshock with Fraps on, go for it. I'm actually hoping to see the demo benchmarks run for future testing of new video cards. At least until we get a Crysis and Alan Wake benchmarker.

Some Highlights of the tests, taken during Maxed benching in DX10 mode.
I'm giving links since they're 800x600 (max Photobucket upload)

http://img.photobucket.com/albums/v4.../Shoreline.jpg
http://img.photobucket.com/albums/v4...IC/n00koff.jpg
http://img.photobucket.com/albums/v4...JustNukeIt.jpg
http://img.photobucket.com/albums/v4...s/WIC/Tank.jpg
http://img.photobucket.com/albums/v4...IC/Bombers.jpg
http://img.photobucket.com/albums/v4...C/Bombrain.jpg
http://img.photobucket.com/albums/v4.../GPUKiller.jpg
http://img.photobucket.com/albums/v4...IC/Enddemo.jpg

[simple_inhibition]

holy crap that game looks bad@ss! looks like ill be jumping onto DX10 when i get home after all! thanks for the info and screenshots

[insomnia]

yah the game is pretty cool - I had it on med setting with a 1280x1024 rez and FPS was about avg 30 but when they used the big arty like a really big airstrike or a nuke my FPS went down to 12 - nothing lower then 10 I think. that was online.. I didn't look at my FPS in the tutorial.

[[Dr. V]]

Only thing I have to say about the mini-review is that widescreen monitors will never run 1600x1200 natively. You may have been looking to say 1680x1050 - just a heads up.

Other than that really thorough review and you covered some good points.

I'm looking forward to this game just a bit more now!

Also, it seems like the insanely low FPS ratings at the higher qualities isn't the video card - because that video card can easily handle 1024x768 resolutions and get much better FPS than that - but it looks like the system is actually bottlenecking the card somewhat...I can see you have the dual core optimizer running and all that, but it seems like all these games are definitely meant to be run on Core 2 Duo-type CPUs (meaning as good as them or better). Not to mention quad core when running full out DX10 - there is no other way to get full 100% DX10 performance out of a DX10 game if you don't have quad core. But then again it's not necessary to have a similar gaming experience through dual core.

[gvblake22]

Originally Posted by ST!X

Only thing I have to say about the mini-review is that widescreen monitors will never run 1600x1200 natively. You may have been looking to say 1680x1050 - just a heads up.

Unless you have a 24" (or larger) screen that is capable of 1920x1200. But if we are talking about this sentance:

Originally Posted by Stormcrow

Since the resolution is set so low, you'll seriously need to consider that when running a widescreen display at 1600x1200.

Then I think it just needs some clarification. Widescreen monitors display with a 16:10 aspect ratio, not the traditional 4:3. The most common 16:10 resolutions are 1440x900, 1680x1050, and 1920x1200. The most common 4:3 resolutions are 1024x768 and 1600x1200. 1280x1024 is technically a 5:4 aspect ratio but is still lumped in with "standard" resolution monitors.

[[Dr. V]]

Within your video card settings you can always set a fixed custom aspect ratio of 4:3 and then set the res at 1600x1200 (or 1280x1024) and you'll see the image in the center of the screen...

I'm not sure if that's what you were referring to, but either way widescreen native resolutions do NOT include 1024x768, 1280x1024, or 1600x1200. They most commonly include (for 16:10) - 1280x800, 1440x900, 1680x1050, and 1920x1200. There are more but they aren't as common as the ones I listed. Because the graphics drivers often have customizing options, you can force the monitor to run at a different aspect ration and still conserve the quality.

[gvblake22]

ANYWAY...
Very cool writeup there Stormcrow! It's good to hear you can actually notice a difference between DX9 and DX10 and it isn't just a bunch of bad programming or somthing.
Thanks for the plethora of numbers. I know now for sure that my X1950 Pro just isn't going to cut it much longer if I want to play games at 1920x1200

[Yellowhello]

WOW, excellent write-up! Looks like a kickass game, but I don't think I could run DX10 on my 8600GT... But again, nice write-up!

[Quakindude]

What resolution did you use for all the testing? I'm assuming you went with 1026x768 since that is your monitors native resolution.

I've downloaded this and will do some tests myself to see if a faster CPU makes a big difference.

[Quakindude]

DX10 Benchmarks:


1024x768 - 1920x1200

Maxed:
Average- 21 - 15
Minimum- 12 - 8
Maximum- 38 - 30

Very High:
Average- 24 - 18
Minimum- 14 - 9
Maximum- 50 - 36

Very High No AA/AF:
Average- 31 - 22
Minimum- 13 - 10
Maximum- 82 - 43

Medium: Defaults to DX9 settings
Average- 52 - 44
Minimum- 25 - 25
Maximum- 126 - 76

Low:
Average- 82 - 80
Minimum- 27 - 11
Maximum- 178 - 186

Very Low:
Average- 116 - 103
Minimum- 22 - 20
Maximum- 228 - 234



So there's my benchmarks. The only time I restarted the game was when going to medium and below settings since the game starts using DX9 by default on those settings. Obviously, this game depends on a high end CPU just as much as it does a high-end graphics card. I would think I have less of a bottleneck than you do with your CPU.