3DMark06 can be observed an evolution of 3Dmark05 which upgrades, improves and otherwise refines weaknesses of the latter. In fact technically speaking from a programmer's point of view, 3DMark05 was fundamentally based around Shader Model 2.0 (SM2.0) within DirectX 9 (first release). The reasoning behind this is actually very simple and may be explained as follows. At the time of development then launch, SM2.0 was the method of choice for implementing 'special effects' within game engines and there was generally very little penetration of SM3.0 (some may recall the SM3.0 FarCry patch Nvidia used to demonstrate their Geforce 6 series GPU's). In addition, SM3.0 compliant hardware was scarce with only the Nvidia 6 series being compatible, sadly at that time ATI didn't offer SM3.0 support in their X800 series GPU's. These two issues combined forced Futuremark to hold back from taking the giant leap from 3DMark03 (which in some cases still used SM1.0 and in others SM2.0) and as a result 3DMark05 became fundamentally SM2.0 and in Canyon Flight (Demo/Game test 3) SM3.0 albeit to a limited extent not taking into account advanced features of DirectX9. 3DMark06 still uses SM2.0 as the minimum supported however as can be predominantly visually noticed, there is great emphasis on SM3.0 and as such this is the key Shader Model. As each 3DMark release aims to accomplish, 3DMark06 equally contains increases in the complexity of geometry, dynamic soft shadows, texture detail and size, the number of lights, and Shader complexity now increased to 495 instructions (up from 96 in 3DMark05). All these improvements contribute in making 3DMark06 resemble to a degree strictly possible current and foreseeable future technological trends in the PC gaming industry. Of course, no matter what Futuremark does, there will always be those keen to stress 3DMark as purely synthetic in gauging system 3D performance. It is nevertheless now an industry standard (and used for marketing of products), and perhaps more importantly, continues to be a viable indication tool in classifying the level of capability between systems. Its features will never answer the questions 'How many FPS will my system give me at 1280x1024 x8AA x8AF in Quake 4 or F.E.A.R' or precisely measure 'When will my system become obsolete for playing the latest games' although the crude estimation of system 3D performance it renders, continuously acts as a best effort storyteller. Below is a rundown on 'What's intrinsically new' in 3DMark06.
Game tests renamed
Although not innately technical but still worth noting, what was previously always known as 'Game tests' has now been replaced under the name 'Graphics tests'. The motives for this are twofold. 3DMark06 stresses the graphics cards GPU more than ever before and is above all about the emphasis on 3D.
New default benchmark resolution
XGA (1024x768) is to be no longer Futuremark has decided and the default benchmark and demo resolution is now SXGA (1280x1024). Is this a bad thing? In light of how the IT industry has been evolving over the years, the answer should read 'no'. The performance of graphics cards, processor speeds, and display capabilities of monitors have all increased to the level that begs one to ask why people stick to XGA. Yes, launching 3DMark06 on for instance an Nvidia 6600GT at 1280x1024 will cause the latter to run and hide rather quickly but the same would occur for any card with only 16GB/sec of local memory bandwidth and similar prowess. Higher resolutions are a good thing for gaming and rest assured soon XGA will become what SVGA (800x600) is today. Higher resolutions are in fact necessary in order to make 3D scenes look more refined. For owners of 15 inch LCD's however, this is an extra nail in the coffin leaving no option but to stay on the 1024x768 bandwagon.
Dual core support within CPU tests
With the advent of Intel's Pentium D and AMD's X2 processors, dual (multi) core personal computing is well and truly the new paradigm of the future. 3DMark06 takes on this fate by supporting multithreading in its two revised CPU tests which perform three distinct operations: 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. Because of how the CPU tests score now matters in the final 3DMark score (see below), any type of parallel thread processing capability in one's processor(s) (i.e. logical: Hyper Threading, physical: two or more physical cores) has lesser or greater potential in influencing the latter. As games developers slowly but surely find stronger pills to cure their thread synchronization programming nightmares (and migraines), it makes perfect sense why Futuremark chose to implement this measurement enhancement. Sadly the standard Game Tests (both SM2.0 and HDR/SM3.0) are still not dual (multi) processor core aware hence for now remain a strict single threaded affair.
Click images to enlarge.
Full/extensive use of SM3.0 & DirectX9C
3DMark06 utilises all features present within SM3.0 with the only exception being the vFace Register, an option that programmatically enables for instance performing changing lightning conditions as through amending the facing position of a triangle. In order to create HDR, three vital aspects of DirectX9C are used, namely 16 bit floating textures, blending and filtering. Floating point precision is a big must in creating realistic real-time shadows in that it allows rising the smoothness and refinement of lightning/shadow effects through allowing a greater spectrum freedom to developers. In example, the more objects (and their associated needs i.e. light source) a given scene has, the more beneficial this becomes. For hardware that doesn't support 16bit floating point filtering (SM2.0 people pay attention) 3DMark06 will just use its Shader emulation instead.
New tests
3DMark06 now includes 19 separate tests adding three more from that available in 3DMark05, all of which added are SM3.0. As some may have heard already, the parallel nature of modern GPU designs allows them to be used for tasks other than pure graphics rendering. With this in mind the new 'Shader Particles Test' tests the physics of particles using a simple gravity and resistance model. In practise within games, such calculations can be used to control the dispersion of raindrops or a wind effect on flags, clothes, leafs and other similar scene objects. 'Perlin Noise', the second of the new tests is actually very different as it's a method for texturing that's recently gained interest for its bandwidth saving characteristics. The basic principle is as follows: combine a noise function together with an interpolation function to reduce the need for loading textures thus saving precious bandwidth. Instead one employs PS3.0 (Pixel Shaders) to cause amendments to the given texture(s), essentially a process that's been dubbed as adding 'noise' hence the name 'Perlin Noise'. The test in essence a Pixel Shader test consists of a total of 48 texture lookups and 447 arithmetic instructions, resulting in a total of 495 instructions. As experienced readers will already contemplate, the minimum specification for PS3.0 allows Shaders with up to 512 instructions hence this test rides pretty close to the limit. Last but not least, the third new addition is Game Test 4 or as Futuremark prefers to refer to it HDR/SM3.0 Game Test 2, which tests HDR performance, a technology 2005 saw occurring in games (and patches) and one that 2006 will witness even more.
New final score algorithm
3DMark06 introduces a new method to gauging the overall 3DMark score based on three separate ratings including what's inherently new, also a CPU score. The SM2.0 score is rendered from 'Return To Proxycon' (Graphics Test 1) and 'Firefly Forest' (Graphics Test 2), the HDR/SM3.0 from 'Canyon Flight' (Graphics Test 3) and 'Deep Freeze' (Graphics Test 4), and finally the CPU score from CPU Test 1 & 2. From this the combined graphics score from all run graphics tests (GS) is rendered differently depending on if the system's graphics hardware supports SM2.0 and below or SM3.0 and below. For SM3.0 compliant hardware GS = 0.5 x (SM2.0 score + HDR/SM3.0 score) while for SM2.0 GS = 0.75 x SM2.0 Score. Lastly for all keen mathematicians, the final 3DMark score = 2.5 x 1.0/ ((1.7/GS + 0.3/CPU Score)/2).
And a game too?
Many fans missed it in 3DMark03 and 05 and believe you me your groans (maybe other things too) have been heard by Futuremark and thus without much further wordiness, allow me to introduce a game but no name. Strange is it not? Therefore all one must be concerned with is a game as we are forced to refer to it that probably will yield endless fun and here's why. Set in red hilly dessert terrain, the same as used in the CPU Tests (named 'Red Valley'), one plays a spaceship that must shoot down as many speeders as talent allows all within only 5 minutes. Each destroyed speeder earns the player 30 points and the best part isn't the score but the fact it can submitted to the ORB (Online Results Browser) alongside the 3D benchmark data. A nice little finishing touch wouldn't you agree?
