DX9 vs. DX10 - CRYSIS - Did DX 10 Deliver?

DX9 vs. DX10

During our testing we noticed two main differences between runningCrysis in DX9 and DX10. It must be understood that we only noticedthese differences when were comparing static screenshots in Photoshopwhile staring at them intently and flipping back and forth betweenimages to look for difference. These were not differences we easilyspotted while actually playing the game. The first difference is thatwhen running in DX9 the HDR seems to have less color range compared toDX10. For example in DX10 we saw a better contrast of light and darkcolors compared to the DX9 screenshots. The other difference is thatDX9 is much more inefficient as far as system memory goes, it gobblesup much more RAM than DX10 uses. We will explore these things below.

Below is a gallery of images. The first row is all from the NVIDIAGeForce 8800 GTX at 1600x1200 with the highest playable settings used here. While the second row is all from the ATI Radeon HD 3870 at 1280x1024 with the highest playable settings used here.

Image Quality

NVIDIA Screenshots

ATI Screenshots

In the first image of each row the difference between DX9 and DX10 HDRis very pronounced in the clouds next to the sun. You notice how inDX10 the clouds closer to the sun are brighter and the light seems tospread out over the clouds further away better then in DX9. Thoughthese screenshots are the exception to rule; with these you are able tovery easily make out the differences in HDR.

With the second and third screenshot of each row it is much harder totell the difference, even when you know what you are looking for. Inthe second screenshot a very careful observer may notice that the truckin DX10 looks darker (because of the more realistic lighting) then inDX9. Though that is just the tip of iceberg there is also very subtledifference in the lighting of the trailer in the left of the picture,the fence, the building, and even the road.

Looking at the third screenshot it does seem like they look to be thesame and even someone that knows what to look for may have a hard timeseeming it, but there is a difference. The fourth and fifth are thefull size screenshot used in the third. Go ahead and open up those twoin separate tabs and switch back and forth between them. Notice how inDX9 the color of the rocks, the road, and the plants seem grayer. Alsonotice how the DX9 sky seems monotone compared with the DX10 sky.Finally, notice the more realistic shadows of the crown of rocks at thetop of the hill in DX10.

So what does this all mean? It means DX10 implements HDR in Crysisbetter. Does this mean that we should play in DX10 because of thebetter image quality? No. The difference between the image quality isjust so small that even when comparing the two side by side it is stillhard to tell the difference most of the time. It just comes down to:Technically there is a difference but you will never notice it inreal-world gameplay.

System Memory

While it does seem to be nit picking and won’t really mean much to mostthat are running 2GB of RAM or more there was some behavior in RAMusage that we thought worthy of noting. To put it simply, DX9 ate upmuch more system RAM while the game was running compared to runningCrysis in DX10. To measure the amount of system memory we used Crysis’sDevMode readouts which you can see in the screenshots above, where itsays MEM=???? MB. This readout indicates total system memory usage ofCrysis.

As we see in the NVIDIA screenshots in DX10 there is about 1GB ofsystem memory being used but when we look at that same screenshot inDX9 there is about 1.5GB of system memory used, and sometimes it waseven higher. That means DX9 is so inefficient that it uses about a halfa GB of RAM or more with the exact same image quality settings.

Looking at the ATI screenshots from the 3870 there is only about 150MBdifference. The reason for the smaller difference here is that it isproportional to the in-game settings. The NVIDIA and 3870 cards bothhad system memory usage as high as 1.6 GB of RAM in DX9 mode. While inDX10 they were only around 1GB. So the higher the in-game settings inCrysis the more inefficient DX9 is.

While this is very interesting that DX10 seems to be more efficientlyat using system memory in the end it still means nothing as far as thegameplay experience goes. We didn’t notice the higher RAM usage eatingaway at gameplay performance, for example DX9 did produce slightlyfaster framerates than DX10 on every video card tested here today.

Gameplay Summary

In looking over both NVIDIA’s older 8800 GTX video card and the newer8800 GTS 512MB we see that there was no improvement in the in-gamesettings in DX9 mode versus DX10. Although playing in DX9 does providean increased average framerate it is unperceivable in real gameplay.From all of our gameplay testing in DX9 and DX10 our conclusion is thatwith an NVIDIA GPU based video card there is no reason not to play inDX10. DX9 does not improve framerates enough to allow you to set higherin-game settings or make the game have a better feel. DX10 also has theadvantage of reduced system memory usage, which could be more importanton machines with lower amounts of physical RAM.

AMD’s DX10 GPUs on the other hand are a completely different story.With the cards used we experienced higher framerates in DX9 thatallowed us to play the game with better in-game quality settings. Theframerate difference was great enough between DX9 and DX10 that withthe single Radeon HD 3870, DX9 saved us from “Low” quality settings.And to put it simply, “Low” settings in Crysis pretty much suck.

DX9 vs. DX10 Image Quality

We did find IQ differences, but they were ever so slight. We discoveredthat running Crysis in DX10 provides better HDR contrast than in DX9.There seemed to be a better overall tone of light and dark colors whenwe compared still screen shot images. When playing the game it was nearimpossible to tell the difference between DX9 and DX10, and we evenknew the exact nuances we were trying to realize. Taking the “PepsiChallenge” in Crysis comparing IQ between DX9 and DX10 would likely endin a draw. Only upon very close examination of screenshots could wetruly identify the differences.

So where are the pretty DX10 Crysis graphics promised? They can befound in the “Very High” settings, but unfortunately those settings arenot playable with today’s highest-end GPUs! When running DX10 Crysisyou have the in-game settings of “Very High” that are available wherethese uber-settings are not available in DX9. Even with the mostexpensive $1200 Quad-SLI configuration we still weren’t able to scale Crysis to its highest DX10 settings and have an acceptable gameplay experience.

The Bottom Line

NVIDIA GPU based video cards have no real-world gameplay differencesbetween DX9 and DX10, so there is no reason not to run in DX10. AMD ATIGPU based video cards do suffer a performance penalty for using DX10.If you are running an NVIDA GPU based video card, continue to run inDX10, but if you are running an AMD ATI GPU based video card you mightwant to consider running in DX9 for the best Crysis experience.

We feel that HardOCP should not let AMD’s poor DX10 performance retardour testing. DX10 is the future whether we like it or not, and we wantto encourage game developers to continue to push the envelope andprovide a good gaming experience. We also want to encourage NVIDIA andAMD to continue to develop and mature their DX10 drivers in both singleand multi-GPU configurations for games.

Kyle’s Notes: The GPU companies have been evangelizing the “huge benefits” of DX10 graphics to us here at HardOCP literally for years now. The quote below is ours from mid-2006.

The DirectX 10 API is a move in the right directionwhen it comes to gaming. It looks as though it is easily going to allowfuture GPUs from ATI and NVIDIA to shine in the gaming arena. GivenATI’s and NVIDIA’s quick advancements in GPUs as of late, we think thatthe DirectX 10 gaming era will truly be one that reaches the next levelas it is shaping up to be exactly what the game content developers havebeen wanting for years now. And when the game content developers arehappy, that means us gamers are happy.

I don’t think many enthusiasts are “happy” with DX10. Some might besatisfied with it by now. I know I am personally, but I am not exactlyexcited about it either. The fact of the matter is that we have notseen any gaming titles greatly benefit from the technology that we canjustify beyond a screen shot. As Matt mentioned, DX10 is the futurewhether we like it or not and we do not feel as though backsliding onAPI in our video card evaluations is acceptable or wise. We intend tohold the GPU’s companies’ feet to the fire in regards to DX10. And doknow that we have been asked to not use DX10 in our testing since therelease of Windows Vista. We have not ignored DX10 titles or thebenefits sometimes associated with them. And while we will keep DX10 asour testing base, we are not going to simply assume it is the bestalternative for our readers. We have other DX9 vs. DX10 articles thatare game specific here, here, here, here and here.

All is not doom and gloom though. I think we have some Triple Atitles coming this year that will hopefully expose DX10 advantages. FarCry 2, STALKER:Clear Sky, Fallout 3, and Borderlands come to mind.

We thought that if we held out a while and let DX10 drivers matureand Crysis itself mature, that we might see better results. As itstands now Crysis is not the DX10 poster child that we were led tobelieve it would be. Maybe Crysis is an incredibly forward lookingtitle and not “badly coded” as rumored. I guess time will tell. I doknow this; I will play through the first third of the game again when Ican set global “Very High” DX10 settings with MSAA at 1600x1200 orbetter. That will be a 3D environment that will surely be impressive.

Hmm there alot more differences in Crysis dx10 compared to dx9 that where never mentioned.

The dx10 version has better depth of field, adds alot more motion blur, uses 3d textures and it also adds better graphic physics.

All these things can't really be captured well in screenshots, but they become very apparent while playing the game.