Rage3D 3DMark06 First Look Rage3D 3DMark06 First Look3D graphics benchmarking continues to be the central component in the analysis of video graphics performance. Today’s video cards (GPUs) have become increasingly more complex and with it the software that utilizes it. This has also made the task of properly assessing the overall potential of GPUs more difficult. It is essential to use a wide variety of applications in order to obtain as much information about the GPU as possible. One of these applications of course is synthetic benchmarks.
3DMark has been the de facto tool in synthetic benchmarking for the past seven years and has been used by practically all on-line and paper publications. More importantly it has been used extensively by the entire graphics loving community. More than 30 million benchmark results have been submitted to Futuremark’s Online ResultBrowser database, which is a remarkable amount! ORB allows for a great deal of comparison between users and their many different and/or similar system configurations, providing a great deal of value on where you system performance may be.
It has been just a little over a year since 3DMark05 was released but a lot has changed since then. Video graphics processors capable of Shader Model 3.0 have been released by both parties allowing it to become widespread. SM 3.0 is the minimum standard in graphics capability moving forward. Pixel shaders and more pixel shaders is all we’ve been hearing and it is fast becoming a reality with games such as F.E.A.R. HDR continues to be preached as the next big breakthrough in graphics realism with next generation game adoption just around the corner.
With all these changes 3DMark05 has rapidly aged and its use to assess future graphics performance has diminished. Scores have almost reached 20,000! Futuremark looked to rectify this and has released 3DMark06 which incorporates many future trends mentioned. They have built upon the 3DMark05 platform for 3DMark06, and nearly nearly every aspect of it has been improved.
When you first run 3DMark06 you might do a double check and wonder if you actually ran 3DMark05! 3DMark06 re-uses quite a bit of content from 3DMark05, improving upon three tests and only adding one complete new one. You would notice a distinct difference in image quality though if you ran the two side by side. Much work has gone into improving the artwork by adding more detail.
Most of the effort involved in 3DMark06 has been upgrading the 3D engine to support the two key features 3DMark05 lacked, Shader Model (SM) 3.0 and High Dynamic Range (HDR) rendering.
3DMark06 uses these features in ShaderModel 3.0:
Much emphasis was placed on dynamic shadows in 3DMark05 and this has continued with 3DMark06. Futuremark has introduced a new type of depth shadow map called Cascaded Shadow Maps (CSM). The implementation is a new approach to achieve angle independent quality on dynamic shadows for all objects on screen. The exact specification of its implementation is pretty technical but like 3DMark05 not all hardware fully supports this method in hardware. For ATI cards the only card capable of hardware rendering it is the X1800 and X1600 series which supports depth textures via a DF24 depth map format.
| Feature | 3DMark05 | 3DMark06 |
| Shader model | 2.0 | 2.0 and 3.0 |
| HDR Rendering | No | Yes |
| Dynamic Shadows | 2 LiSPSM Shadow Maps | 5 Uniform Shadow Maps |
| Smooth Shadows | Yes (4 sample) | Yes (16 sample and 4 sample) |
| Graphics Test | 3 (All SM2.0) | 4 (Two SM2.0 and Two HDR/SM3.0) |
| CPU Tests | 2 (Software vertex shaders) | 2 (Complex AI and physics) |
| SM3.0 Feature Tests | No | Yes |
| Number of tests | 16 | 19 |
| Level of Detail | High | Very high |
| Texture Size | High | Very High |
| Materials | Simple | Complex |
| Shader Complexity | High (Max 96 instructions) | Very High (Max 512 Instructions) |
| Amount of lights | High | Very High |
| Sub Scores | 1 (CPU Score) | 3 (SM2.0, HDR/SM3.0, and CPU Score) |
| VRAM Usage | ~128MB | ~256MB |
| Playable Game | No | Yes |
| Target GPU | 2nd Generation SM2.0 or better | 2nd Generation SM3.0 or better |
| Targer CPU | Single Core | Dual/Multi Core and Single Core |
The scoring method in 3DMark has also changed and now takes into consideration CPU results. Futuremark’s rationale for this is:
This was done due to the fact that there are more and more games using complex AI algorithms and complex physics calculations. The graphics tests have been optimized to be as GPU bound as possible, hence they aren’t affected much at all by the CPU. In order to create a very game-like scenario, all graphics tests and the CPU tests are used for the final 3DMark score, and thus correlate very well to what performance users will see in future games.
3DMark06 has significantly grown in file size, more than twice the size of 3DMark05. The major contributors to this increase are:
3DMark06 brings back Return to Proxycon with improved artwork, more shadow casting objects, more lights, and using the new shadow-technique but still only using SM2.0. The graphics test definitely looks better with more detailed textures and lighting. Graphic Test 1 is representative of FPS type games which take place indoors and also reflects multiplayer action due to all the characters in the scene.
The Firefly Forest test is back and looks fairly similar to its prior rendition. The new version employs improved artwork and the new shadow technique, but still only uses SM2.0. The firefly in the test has a friend this time around to add more excitement and additional graphical workload.
Canyon Flight is also back from 3DMark05 but it has been vastly revamped making the new test quite different. Canyon Flight is the first test to incorporate Shader Model 3.0 and also use HDR rendering. The test employs a new shadow technique and a completely new shadow filtering technique. A lot of the artwork has also been improved and you can definitely tell the difference. It’s like day and night!
The image difference between HDR and non HDR in these screenshots explains very easily why everyone is so excited about HDR. It adds so much realism to the scene. I can’t describe all the technical aspects to this scene so I’ll pull from Futuremark’s whitepaper on the subject:
This test gives an example of a large scale outdoor scene with HDR rendering, smooth shadows and complex SM3.0 shaders. The scene is very complex with large areas of water reflecting the high canyon walls. The HDR rendering is one of the key points of interest in this scene, proving the increasing importance of floating point rendering to achieve realistic scenes. The water in this scene not only features realistic looking HDR reflections and HDR refractions, but it also has a depth fog, making the sea monster swimming under the airship actually look deep down in the water. The surface of the water is distorted using 2 scrolling normal maps and four Gerstner wave functions. This scene also uses a complex heterogeneous fog to make the whole canyon appear humid. The air in this scene also uses the same atmospheric light scattering algorithm as in 3DMark05 making distant cliffs of the canyon really look far away. The sky uses a more complex atmospheric light scattering algorithm than in 3DMark05, with cloud blending.SM 3.0 / HDR Graphics Test 2: Deep Freeze
The only completely new test in 3DMark06 is Deep Freeze and it’s quite impressive. Making use of HDR, it shows a fairly realistic scene of a research base in desolate Antarctica. It truly does seem like an opening scene in movie where you expect a lone person to walk in from the horizon. This test is a showcase for using HDR effects in vast landscapes, and the pronounced effects of dynamic long soft shadows showcased in the transition from daytime to night-time.
 Deep Freeze |
 Deep Freeze |
 Deep Freeze |
 Deep Freeze |
Futuremark has created an all new CPU test for 3DMark06 that consist of a game scene in a mars like setting with a maze of canyons, and 87 vehicles of various types which try to get to the castle while being targeted by tanks.
The CPU test was designed with the trend towards multi-processor cores in mind. In this manner they split the test into three components: game logic, physics and path finding AI. The game logic, including the graphics engine operation, runs in a single main thread that also drives the other two tasks. The physics simulation runs in a single separate thread, and is synched with the main thread at each physics step. The path finding AI runs in a number of worker threads (the number of threads is scaled with available processors), and is synched with the main thread at set intervals, generally some multiple of the physics step interval.
The path finding algorithm used is D* Lite (http://www.cc.gatech.edu/fac/Sven.Koenig/) and the physics are being computed using the AGEIA PhysX library (http://www.ageia.com). There are two CPU tests (1 and 2) in 3DMark06. CPU test 1 has a larger physics load than CPU test 2 but a lower AI load.
Ratchet tested out 3DMark06 on a whole bunch of cards for me. Great guy! ;)
| 3DMark06 Overall Score |
||||||||||||||||||||||||||||||||||||||
|
||||||||||||||||||||||||||||||||||||||
The SLI cards are on top which is of no surprise. Going by 3DMarks the 6800GT SLI option looks to be a fairly good investment. I was fully expecting the X1800XT to be on top and am surprised that the 7800 edges it out. Hardware shadow acceleration on the 7800 accounts for a bit of this but the majority is due to the 7800 winning out in both SM2.0 and SM3.0 scores. The SM2.0 score is reasonable as the 7800 has eight more pixel shader units but I was expecting the SM3.0 score to favour the X1800. This tells us that there isn't extensive use of flow control / dynamic branching which the 1800 series is suited towards.
The midrange cards are all fairly close to each other with the X1600 flexing its pixel shading prowess and ability to hardware accelerate 3DMark's dynamic shadows. ATI's last generations high end doesn't do too poorly even though they are incapable of running the SM 3.0 tests.
The low-end is fairly predictable with the X1300 just a bit above baseline score but beats out both the X800GT and X700 Pro.
3DMark06 brings Futuremark’s synthetic benchmark up-to-date as a relative indicator of future performance with the application of SM 3.0 and HDR. It truly showcases how spectacular HDR will improve image quality in next generation games. Its use of CPU performance as part of the score is new and interesting and adds a new dimension to the benchmark. I don’t believe there has been extensive use of dynamic branching in the new shaders, but we haven’t seen any games use much of this capability yet either, making it difficult to assess its future importance.
Overall I feel 3DMark06 represents a good refresh the the franchise though, and expect it will continue its popularity until we see the next true major release.