In-Depth Look: Futuremark 3D Mark 06

[PAGETITLE=Introduction] The time has come for a new version of 3D Mark, and we have everything you need to know. To date, there have been 7 versions of the popular benchmarking tool, including this new one. With each new version, the tests got much tougher on your system, but anyone with a new high-end system have been able to run them without too much of a sweat. One nice thing about 3D Mark is that all the tests are ahead of the times, so we can use them for quite some time. Heck, there are still some who still use '01 and '03 religiously. A new version may catch some by surprise, because some of us have yet to even touch the 10K mark in '03, and it only get's much more difficult in '06. This is not some small update though, but a completely revamped product. The tests are all hardcore as we have come to expect, and if you run a GPU with less than 256MB of memory, then expect to weep after the test is over. This article hopes to give you an in-depth look at the new version, with all the information you can want. We have put quite a bit of testing into '06 on both Intel and AMD platforms, Single Core and Dual Core, to give a good idea of what you need for a great score. In addition to scores from '06, we have re-tested the systems with all versions from 2001 - 2005 to see how the scores scale to the latest version. From the screen shot above, you can see that the main screen hasn't changed much at all, although it is slightly longer. One thing that may catch your eye is the fact that the default resolution is now 1280*1024, rather than 1024*768. This only makes sense, since if you have a high end computer, chances are you are running any game above that resolution. You may also notice that the score layout is a tad different, but we will get into that in our benchmarking page. Let's get more into what exactly is new.. [PAGEBREAK] [PAGETITLE=What's new in 06?] Whenever I see a new version of the program released, I admit I get excited. Most times, the tests are so bleeding edge, that you can't experience the same graphics with any current game. That's for the better though, because who wants to play the latest game at 8 frames per second? Without a doubt, the tests are enjoyable to watch, and if you want to jump right into the screen shots, they can be viewed on page six. One thing you will notice immediately, is that the tests look familiar. All three tests from '05 make a triumphant return in '06, but make no mistake, they are not that similar. In addition to those though, a new test entitled "Deep Freeze" has been added, which will be used to show off HDR's capabilities of your card. The reason Futuremark releases new versions of their program, is because that the graphic technologies are constantly evolving. As with all previous versions, this new one puts bleeding edge GPU features to use. We will see some Shader Model 3.0 being used in some of the tests, and as already mentioned, HDR will play a large role also. Every aspect of these new tests have been "kicked up a notch". The level of detail has been made larger and the same goes for texture sizes. The Shader complexity is quintupled, lights increased and VRAM usage increased. Whereas a 128MB graphics card may have sufficed in '05, you will want a 256MB in '06 to obtain a decent score. It's apparent that Dual Core processors are going to be the choice for most people in the near future. We are already seeing huge benefits of the processors, and games are currently in production that are developed to specifically take advantage of the tech. It's no surprise then, to see Dual Core / Multi-Core support in '06, and trust me, it does make a difference. When you run the benchmark, you will notice that an AGEIA PhysX logo is featured. At first, I believed this was because the card would be supported in the tests, but that's not the case. For the CPU tests, Futuremark implemented the PhysX physics libraries, which is why they included the logo. They have no intentions of adding support for the card, especially since there is no expected release date. [PAGEBREAK] [PAGETITLE=Look at the tests] You can click on any of the screen shots for a larger version, but be warned, they are large. Return to Proxycon This is the first test borrowed from the previous version, but as with the others, everything has been improved. The actual engine has been improved in various ways, to better allow all the new tech. While the test adds new shadow techniques and lights, it will still only utilize SM 2.0. This test tries to recreate an FPS game, with large rooms and tight corridors. There is a lot of action going on with bullets flying and metal cargo being hit. Most of the materials in this test can have light reflected off of it, and when the camera zooms out, it can really slow down. The test makes use of directional lights as you can see in the screen shot; it's quite impressive. Firefly Forest Firefly Forest was certainly the 'prettiest' test from 05, and it makes a return in the new version. Again, the engine has been revamped and supports new lighting techniques and shadow effects. Like the Proxycon test, this one will only utilize SM 2.0 as well. In addition to the green fairy in the first version of the test, another pink one has been added. This test is loaded with foliage and lighting. The fairies fly all around the environment, and the respective colors reflect off the environment. The addition of light scattering and fog makes this test a treat for the eyes. Canyon Flight This is the third test borrowed from the 05 version, but has been immensely improved. This test takes full use of SM 3.0 and HDR, and is a huge leap from the first test. Everything looks amazing, from the ship, the pilot, the water and of course, the dragon. The comparison picture below can speak volumes. HDR plays a huge role in this test, but the only light source is the sun. You will see light reflected off of the dragon, the airship, the water and the canyon walls. The water is pixel shaded to give an ultra-realistic look, and it in my opinion is one of the most impressive features of the test. Deep Freeze Deep Freeze is a new test that utilizes SM 3.0 exclusively, in addition to HDR. It takes place in an Antarctic research base, and looks very much like a movie scene. Use of Dynamic Soft Shadows and Subsurface Scattering Effect is here, and helps make the test look extremely realistic. As with the Canyon Flight test, the only light source in this test is the sun. The sun rotates during the test, so shadows scattered all over, move realistically with it. This is an intense test and will run sluggish on some GPU's, but it's a gorgeous one to watch. Red Valley The CPU test takes place in a large canyon where a lot of action is taking place. There are 87 bots which are traversing through the canyon shooting at each other. The test is locked at 2FPS and 640*480 so that the game is CPU intensive rather than GPU intensive. This is a hardcore test for your CPU, because it uses three types of load: game logic, physics and path finding AI. It's obvious why a Dual Core processor will benefit here, and the program is coded to take advantage of it. [PAGEBREAK] [PAGETITLE=Benchmarking Setup] Now to the fun part... benchmarking! Because the older versions of 3D Mark are still commonly used, I have included benchmarks with each one of those, with the various system configurations. I did not have an ATI card on hand, so all of these results will be done on an eVGA 7800GT. The latest driver to be approved by Futuremark is the 81.98, so that's the one that will be used throughout all tests. We used two separate systems to benchmark with. The first is an AMD Single Core, and the other is an Intel Dual Core. As we will see in the results, Dual Core will effect your score quite a bit for the CPU tests. On each system, fresh installations of WinXP and Winx64 were used. All unnecessary services were stopped so that nothing would interfere with the final scores. Here are the systems being used for testing: AMD System CPU: 3200+ Venice @ 2.0GHz. Motherboard: DFI LanParty Ultra-D w/ 704-2BTA BIOS. Memory: Corsair 2GB PC4000 Platinum @ 3-4-4-8/2.8v. Video: eVGA 7800GT 256MB PCI-E. Sound: Chaintech AV-710. Storage: WD 160GB 7200 8MB. Odds and Ends: Titan Robela Water Cooling the CPU, 4x 120MM fans. Etcetera: Windows XP Professional SP2, Windows x64 Professional SP1, NVIDIA 81.98 Drivers. Intel System CPU: Pentium D 820 @ 2.8GHz. Motherboard: ASUS P5WD2 Premium w/ 0606 BIOS. Memory: OCZ 1GB Gold PC2-5400 XTC. Video: eVGA 7800GT 256MB PCI-E. Sound: Realtek HD Audio. Storage: WD 200GB 7200 8MB. Odds and Ends: D-Link WiFi PCI Card. Etcetera: Windows XP Professional SP2, Windows x64 Professional SP1, NVIDIA 81.98 Drivers. Overclocked Values AMD: 270 * 10, DDR540 @ 3-4-4-8: 2.7GHz. Intel: FSB 260 * 14, DDR2-695 @ 4-4-4-10: 3.64GHz. As mentioned earlier, the default test uses 1280*1024 as the resolution. I did not change this, even with the prior tests using 1024*768 as the default. You will also notice that the AMD system is using 2GB while the Intel uses one 1GB. The reason behind this is that I had a 2GB DDR2 kit die on me last week, so I was forced to stick with 1GB. I got in touch with Futuremark, and they noted that going from 1GB to 2GB in these tests will make a minute differences on very few systems. Before we get into the results, here are how the scores are calculated: SM2.0 Score = 120 x 0.5 x (SM2 GT1 fps + SM2 GT2 fps) HDR/SM3.0 Score = 100 x 0.5 x (SM3 GT1 fps + SM3 GT2 fps) CPU Score = 2500 x Sqrt (CPU1 fps x CPU2 fps) We'll define for clarity: GS for SM3.0 hardware = 0.5 x (SM2S + HDRSM3S) GS for SM2.0 hardware = 0.75 x SM2S. 3DMark Score = 2.5 x 1.0/ ((1.7/GS + 0.3/CPU Score)/2) i.e. 2.5 x weighted harmonic mean of GS and CPU Score. Now that you are confused, let's continue to the results. [PAGEBREAK] [PAGETITLE=Benchmarking Results] Before you ask me why I did benches on both x64 and XP, I can tell you...

In-Depth Look: Futuremark 3D Mark 06

Page 6 – Additional Test Screen shots

Rob Williams