We could not have been happier with the performance of Unreal Tournament III! The game ran exceptionally well on every video card we tested it on. Some might say that it is the natural result of a game engine that has been in development for 3 years, but we think Epic deserves more credit than that. They've created a game with more detail packed into every cubic inch of gamespace than many games have in entire levels, and they've done so in such a way that a sub-$200 video card can play at 1920x1200, with the highest available in-game settings. That is simply unprecedented. Of course, ATI also deserves huge credit for bringing such a capable video card to market at such an excellent price with the Radeon HD 3850.

Anti-Aliasing performance was disappointing, presenting us with a huge performance penalty for even just 2X MSAA. This is likely due to the fact that multisampling floating-point surfaces is a very memory and GPU intensive task, and of course, UT3 uses some seriously high-resolution floating-point post-processing. But that very same post-processing also serves to reduce the need for AA, by obscuring some hard edges. Of course, it's not a perfect solution, and it sometimes has the opposite effect. But it works well enough to be effective on current hardware.

After testing, we were only left wishing that we could scale the packaged graphics even higher for our high-end video cards, instead of merely fiddling with the AA setting. We did see a real improvement in the gameplay experience with the most expensive video card in this evaluation, but not such an improvement to justify the very high price of entry into that product segment. If indeed there is a DX10 patch in the works this could possibly help justify the use of enthusiast-class video cards such as the 8800 GTX in this game.

The FPS smoothing feature was interesting, and further proves how important it is to look at gameplay performance experienced from video cards rather than raw average framerates in comparison. Just because a certain video card may produce overall higher framerates (the 8800 GTS/3850/3870 in this evaluation) doesn't mean they are inherently faster. With FPS smoothing enabled the GTX resulted in what looks like slower framerates, but those framerates were smoother. It is now time, more than ever, to concentrate on real-world gameplay experiences delivered by video cards rather than looking at average framerate bar charts to compare performance. Overall, we think that the Radeon HD 3850 is the best value for playing Unreal Tournament III, other games notwithstanding. If you are a UT nut (and we know you're out there), you will simply not find a better bang for your buck than the ATI Radeon HD 3850.

The UT3 PhysX mod pack is a big win for AGEIA and their hopes of getting PhysX hardware in as many PCs as possible. The two levels that AGEIA released aren't the most amazing things we have ever seen out of the mod community and the performance hit our test system took even with the PhysX card installed definitely raised an eyebrow. Overall though, the additions to real-world game play were interesting enough to warrant a call to arms for creative level designers. The idea of releasing a tool kit that will allow gamers to easily create their own content that utilizes the technology is also a brilliant idea and will hopefully get some of the more diligent AGEIA fans off their butts making some even more inventive and original content based on UT3.

- DirectX 10: Fixes texture flickering in bushes, clouds, trees
- DirectX 10: Performance improvements
- DirectX 9: Fixes Anti-Aliasing corruption seen on the ATI Radeon 2900 Series
- DirectX 9: Anti-Aliasing performance enhancements
- DirectX 9: Crossfire performance improvements
- DirectX 9: Fixes random graphics corruption during game play

• You can play most modern games with maximum quality settings in the most popular resolution only on the high-end GeForce 8800 GTX, not always at that - minimum frame rates in some cases drop too low.
• New games use up to 600-700 MB of video memory. It does not mean that all these resources must be in local memory of a graphics card. Games often give resource control (textures, etc) to API, especially as Direct3D 10 uses video memory virtualization. Nevertheless, there is an apparent tendency to increase requirements to video memory size on graphics cards. So 512 MB can be considered an optimal solution now. 256-320 MB - insufficient, 0.7-1 GB - not used in games so far. But this memory size makes sense for high-end solutions, because they can provide acceptable frame rates in higher resolutions.
• We can see an evident increase in draw calls. Games may now use 2000 calls per frame, although the most optimized projects do fine with 500-1000 calls per frame. The increase of draw calls affects the growing dependence of 3D applications on CPUs - more draw calls generate a heavier load on a CPU.
• The number of draw calls grows together with the amount of processed geometry. We are not surprised to see 300000-500000 polygons per frame anymore. Most advanced games may use up to million triangles per average frame and up to 2-3 millions in extreme cases.
• Performance of a computer with the GeForce 8600 was not limited by a CPU in all tested games, it depended on the graphics card only. On the other hand, the GeForce 8800 GTX was often idle in our conditions, up to one fourth or third of the time. Conclusion: such graphics card needs either a more powerful CPU or better optimizations in games (fewer draw calls in the first place.)
• The ROP load was always low - up to 10-20% in both cases. Of course, we should take into account the FPS difference. But the number and capacities of ROPs were not the main stumbling block in our tests. The G80 and G84 demonstrate similar results, except for one case (PT Boats), probably owing to high requirements to video memory size.
• Modern games still heavily load texture units, their minimum load is 30%, up to 75% maximum (in the G80!) Plans of some companies to change the ratio between the number of texture and shader units were apparently premature - even such an advanced game as Lost Planet: Extreme Condition uses TMUs so actively.
• The load on geometry units and input assembler (it fetches geometry and other data from memory to be used by other units) in games is quite low. It never limits rendering speed, although the load on input assembler is heavier in some projects than in other games.
• Unified processors in all new games act as the main bottleneck. Their peak load in almost all games reaches 70-90%, which speaks of their full utilization. So their performance determines rendering speed. Consequently, performance of these units plays the most important role in modern games. As the G84 does not have many such units, the GeForce 8600 cannot demonstrate better results. The load on shader units in the G80 is evidently lower. This GPU has a certain performance margin of unified ALUs.
• It's now confirmed that new applications have higher requirements to shader units. The games we tested show that importance of computing units in GPUs grows, and that it will continue to grow (it slowed down a little in multiplatform projects for apparent reasons). They will become even more important, if future games use not only vertex and pixel shaders, but also geometry shaders.

To begin with, we took the following system and ran the CPU and GPU benchmarks in 32 bit and 64 bit mode: Windows Vista Ultimate x64, Intel Core 2 Quad Q6600 @ 3GHz , 4GB system RAM, NVIDIA GeForce 8800 GTX 768MB, ForceWare 163.69 drivers.

We took two extremes to see what the results would churn up: 800 x 600, running in DX9 with all detail settings at Low, and 1680 x 1050, running DX10 with all detail settings on Very High. Each benchmark batch file was run 5 times, with all of the collated results averaged. The results are in the image to the right.

Before we all jump to any conclusions though, bare in mind that this is a demo and not the final retail game - we could genuinely see noticeable gains by using the 64 bit version but if we assume that the demo is a pretty close code version to the launch one, then not all of us are going to be so lucky. Please let us know what you find out: try running the CPU and GPU benchmarks across a range of settings in 32 and 64 bit, and then post your findings!

If you tweak the configuration files in CVarGroups by copying and pasting the "very high" settings (1st paragraph) IN PLACE of the "high" settings (last paragraph) the game will load the highest possible settings even though the drop-down menus display "high." The difference between "high" settings and the tweaked settings is immense: shadows are deeper, more realistic; the leaves have better reflective properties, better textures; the colours are better; and the level of detail is simply stunning.

With these settings I'm running the game between 15-25 FPS at 1440x900 and (wait for it) 8x AA, and it looks PERFECT. Best of all, this is in XP. So I'm happy. I have XP and I'm playing the game at settings higher than DX9 allows (strictly speaking). Give this a try if your rig can handle it.

The Crysis settings system works by only offering you resolutions that the computer can handle. What's frustrating is that none of the cards here today gave us the ability to run at our native 2560 x 1600 resolution with any detail settings. With that said though, given the Ultra was only cracking 50FPS at 1920 x 1200 we wouldn't be expecting big numbers at 2560 x 1600. Also, if their system was flawless it wouldn't let you choose past 800 x 600 for the 8600GTS and HD 2600 XT. While I don't doubt that there are people out there with an 8600GTS saying "Nah this game runs great at 1280 x 1024 and high settings", it's clear that these people have no idea what smooth performance looks like.

Look at the performance from the 8800 GTX and the 8800 GTS 640MB cards compared to one another, we found some interesting inflections. For instance, at 1024x768 without AA, the 8800 GTS 640MB system out performed the GTX system. Yes, I know the testing process wasn't exactly the same, so we have to make some broader generalizations, but it would appear that the quad-core CPU that Jeremy used was in fact a factor in overall performance as the Crytek developers indicated. Another interesting note that came from this weekend - NVIDIA acknowledged that SLI scaling on Crysis was some crippled for the time being. A new driver is going to be released that will help with it, but they are saying that Crytek has a couple of fixes of their own that need to be made for proper multi-GPU performance so again, we might want to wait for the final retail version of the software to really get into the multi-GPU capabilities of the engine.