by Yaroslav Lyssenko
12/07/2010 | 05:47 AM
Many great civilizations of the past thrived, among other things, due to complex systems of manufacturing and production regulations. By the 19th century human society had grown so sophisticated that there emerged an even greater need for uniform measurement and standardization as a consequence of the Industrial Revolution. Initially being part of the contract between supplier and purchaser, those norms and standards were later adopted across all countries and industries. The development of the transistor and the personal computer at large would not have been possible without strictly adhering to commonly accepted standards.
As the number of transistors in a single electronic chip was growing up and the whole computers with all their components were getting more and more complex, it became very hard to judge the raw horse power of a given system based solely on its CPU megahertz numbers. A standard measuring tool had to be devised, so that average users as well as journalists and manufacturers could compare different models and configurations available on the market with each other.
It has taken Futuremark Corporation quite a while to build solid reputation as the developer of an industry etalon, but they have been successful in achieving that goal. Almost twelve years since its initial release, we now welcome 3DMark, the Eleventh!
Before we jump to describing the latest installment in the series, let us take some time to pay tribute to Futuremark’s earlier efforts that deserve a separate page in the history of computer graphics. The first computer benchmarking tool with the 3DMark logo was developed by Futuremark Corporation (formerly MadOnion.com and initially just Futuremark) in 1998.
3DMark99 was based on an MAX-FX engine developed by Remedy Entertainment. The benchmark stressed the hardware but moderately, focusing on software-based fixed transformation and lighting as well as multitexturing. Journalists and average users were slow to adopt the novelty but that was just the beginning. After a few updates and intermediate versions the industry finally accepted 3DMark2001 as a standard measuring tool.
Not only did the new version expect to be run on premium-class hardware components, but it also made an effort to promote brand-new technologies such as Pixel Shader 1.4 to render water and Vertex Shaders 1.1 for realistic vegetation.
The arrival of 3DMark03 in February 2003 made it clear that Futuremark was more than determined to spearhead visual innovations on the PC. That version proved to be a hard trial even for the then-best graphics cards with full DirectX 9.0 support. The test scenes were most impressing, especially on the right hardware, and made one wonder whether the next generation of DirectX 9.0 based video games was going to look just as good.
In many respects 3DMark06 turned out to be the last triumph of the DirectX 9.0 API. Its extensive use of HDR rendering and shadow mapping resulted in truly remarkable visuals and funny benchmarking results, with even the latest top-end graphics cards begging for mercy. Physics calculations were used to determine the CPU’s computing power, emphasizing the important role of the CPU in a gaming platform’s performance.
The rapid development of Microsoft’s DirectX API and the launch of the Windows Vista operating system forced Futuremark to release the Vantage version. Although its overall support for DirectX 10 is rather low, 3DMark Vantage still manages to impress even now, two year after its release.
So, 3DMark has already had a glorious history for a measuring tool, but as with any other software product, there is always a new and improved version just around the corner.
The Vantage name has come and gone but the good tradition of developing new benchmarks from scratch remains. 3DMark11 runs on Futuremark’s in-house DirectX 11 rendering engine which makes use of multithreading, tessellation, lightning and post processing.
The color scheme has changed significantly since 3DMark Vantage, just as the overall interface. The most remarkable achievement definitely worth mentioning is the size of the distribution. In our times it is very hard to impress someone with a 10GB video game, but this cutting-edge benchmarking tool asks for only 300 megabytes of disk storage.
There are a total of six tests in 3DMark11 that vary in their requirements:
Graphics Test 1 – No tessellation. Instead, this test benchmarks the GPU’s capability of processing advanced lighting with several light sources each of which casts shadows.
Graphics Test 2 – Medium tessellation for geometry and a medium-difficulty lighting model (a few light sources with shadows).
Graphics Test 3 – There is medium tessellation again for rendering the pillars, statues and some of the vegetation, the sun being the single shadow-casting light source.
Graphics Test 4 – Heavy tessellation is used to add complexity and details to the scene. Additional light sources make this scene extra hard to render.
Physics test – This test benchmarks the CPU performance by simulating rigid body physics with a large number of objects.
Combined Test – This test combines the physics test and graphics test 4 in a single scene. While the CPU is busy handling the rigid-body calculations, the graphics subsystem has to deal with volumetric lighting, tessellation, post-processing and soft-body physics simulation using DirectCompute.
Since PhysX is currently considered to be an Nvidia-only physics API, the programmers at Futuremark decided to stick to unrestricted technologies and employed Bullet Physics for physics simulation in the 3DMark11 tests. This is a professional open-source library for collision detection, rigid body and soft body dynamics.
The new benchmark’s main feature, besides the Results menu, is the demo mode which was more than missed by the community in the previous 3DMark incarnation. The demo mode combines the scenes from the Deep Sea and High Temple tests with a soundtrack by Pedro Macedo Camacho.
The minimum hardware requirements published by the developer are surprisingly low. A compatible DirectX 11 graphics card accompanied by a 1.8GHz dual-core CPU and only 1 gigabyte of system memory aren’t really the baseline one would expect from a high-stress benchmark.
Entry preset (E)
The Entry preset is designed to test entry-level hardware with a very low load on the graphics card. As you can see, most of the options are just turned off. The benchmark runs at 1024x600 and is clearly aimed at nongaming notebooks and netbooks.
Performance preset (P)
If you are a proud owner of a mainstream system, the Performance preset is designed to stress your hardware. With high-definition content spreading across the globe, it comes as no surprise that the resolution is set at 1280x720 or 720p, which is a popular television format.
Extreme preset (X)
Even if your hardware scored 10,000+ in 3DMark Vantage, the new version’s Extreme preset is designed to kill it on the spot. The Full HD resolution of 1920x1080 (1080p) is accompanied by every major eye-candy feature available in the DirectX 11 environment.
Don’t be fooled by the minimum system requirements because the Entry preset has almost none of the DirectX 11 eye-candy features. As soon as you enable the Performance or Extreme preset, things tend to become slideshow-like.
So, there are a lot of settings to be played with and a lot of hardware for us to test!
We are going to check out the hardware demands of Futuremark 3DMark11 on the following testbed:
We used the following versions of ATI Catalyst and Nvidia GeForce drivers:
The graphics cards’ drivers were configured to ensure maximum texture filtering quality and minimize the effect of software optimizations used by default. Transparent texture antialiasing was enabled. As a result, ATI Catalyst and Nvidia GeForce graphics card drivers were set up in the following way:
We used the Extreme preset and benchmarked 14 graphics cards:
Since we are interested in highest-quality visuals whether it you can achieve them with your driver or in-game settings, we will benchmark our graphics cards in 3DMark 11 using the Extreme preset. We run our tests at the following standard resolutions: 1600x900, 1920x1080 and 2560x1600.
There are no scores of 10,000+ points in the Extreme preset as we used to see in 3DMark Vantage. The dual-chip Radeon HD 5970 once again demonstrates that when it comes to raw computing power, it is still a formidable opponent, scoring almost 2400 points. Second place goes to the latest flagship from Nvidia: the GeForce GTX 580 gets a respectable score of 1947 points.
With only 1500 points to boast, the Radeon HD 5870 just misses the podium, being slightly inferior to the GeForce GTX 480 and outperforming the GeForce GTX 470 card.
The latest benchmark from Futuremark destroys the confidence in the hearts of hardware enthusiasts worldwide. Even though we lowered the resolution to 1600x900, we can see that the current generation of premium-class graphics cards have a really hard time. The 20fps barrier is unreachable even for the monstrous Radeon HD 5970 while the others can only yield some 10 frames per second.
The Radeon HD 5970 leads the way with 19 fps in the second test, but the tessellation algorithms implemented here help the GeForce series gain some ground. The Radeon HD 5870 struggles at the back at first but manages to outperform both the GeForce GTX 480 and the GeForce GTX 470 at higher resolutions. Well, this is hardly an achievement considering that the frame rates are as low as 4 fps.
Yet another test brings us unattractively low numbers. The GeForce GTX 580 is indeed the fastest single-chip graphics cards on the planet, but even it has problems catching up with the Radeon HD 5970. The more affordable solutions from AMD and Nvidia are chasing each other, but the frame rates are very low.
If you thought the previous tests were hard, think again. We’ve got only 7.9 fps from the Radeon HD 5970! Alas, the others can’t perform any better. The Radeon HD 5870 is only as good as the GeForce GTX 470, but the conclusion is still that 3DMark11’s Extreme preset is a frightening combination of settings for contemporary graphics cards.
There is one last point of interest for us to discuss. The Combined test provides an estimate of how good your system would be at handling high CPU and GPU loads simultaneously. The result is provided in 3DMark points and is clearly correlated to the overall scores displayed in the beginning of the benchmarking part of our review.
The numbers produced by this test confirm the trends we’ve already discovered. The Radeon HD 5970 is indeed an undisputed leader with 2500 points. Nvidia’s updated Fermi architecture feels quite at home with heavy loads simulated by 3DMark11 as the GeForce GTX 580 leaves behind every single-GPU graphics card on the market. The 1-year-old design represented by the Radeon HD 5870 has nothing to be ashamed of as it manages to squeeze in between the GeForce GTX 470 and the GeForce GTX 480 in the race.
In the Performance Mainstream segment of the market things are not that different. Clearly AMD's latest expansion with the Barts architecture came at the right time. The Radeon HD 6870 and the Radeon HD 6850 both manage to outperform not only the GeForce 460 1GB but also the GeForce GTX 460 768MB. It must be noted that the Radeon HD 6870 enjoys an impressive 35% lead over its opponents in this category.
It is no wonder that the Radeon HD 5770 has no competition in the Mainstream category. The GeForce GTS 450 can only achieve an honorable draw with the Radeon HD 5750 at only 690 3DMark11 points. The entry-level Radeon HD 5670 has no problems outperforming the GeForce GT 430 which is the slowest card in our Extreme preset benchmarking. Although its score of 336 points looks ridiculously small, we can remember the times when entry-level graphics cards couldn’t pass an earlier version of 3DMark at all. So, the little Fermi is doing quite fine for its price.
Futuremark Corporation has had a long history of designing, developing and delivering benchmarking tools. Overclockers and computer enthusiasts have been comparing their exploits with each other for almost a decade now, using 3DMark as the single frame of reference.
It might have been expected that the new 3DMark11 suite would be a killer of modern graphics cards. Even if you are a proud owner of a Radeon HD 5970 or a GeForce GTX 580, you won’t be able to say that there is no point in upgrading or overclocking you system after you run this benchmark. It will take some time for game developers implement the same amount of features and technologies as are abundant in the 3DMark11 engine, so it will serve as the hardest test of graphics performance for the near future.
It is a shame that for all the technical innovations and beauty, the benchmark has somehow lost its appeal as a demo. A demo mode is implemented again but we don’t think that someone will want to turn it on and enjoy it all day long in a loop. We guess the Mother Nature game test from 3DMark03 remains unsurpassed in terms of sheer aesthetics.
This concludes our review of the 3DMark11 suite. Once again Futuremark Corporation has managed to come up with a highly sophisticated piece of software which will take a few generations of graphics cards to get to high enough frame rates.