Unigine Heaven 4.0 Benchmark Tool Walk Through
Reviewed by: Guest_Jim_*
Reviewed on: March 12, 2013
Introduction & Simple Results
The Unigine Heaven Benchmark has been out since 2009, but over the past three and a half years, it has been updated to reach its current version; 4.0. At the time it was released it was the first benchmarking tool to utilize DirectX 11 and still does prove itself to be a good demonstration of some of the API-libraries features, such as tessellation. The benchmark also serves well as a stress-test for GPUs, as it pushes the polygon count quite high.
The setting of Heaven is a "magical steampunk world" with brass, wood, gears, and a zeppelin decorating a city environment. Also at the center of the city is a statue of a dragon that looks ready to leap to life and defend the town if needed. Unlike some other benchmarking tools, Heaven allows the user to freely roam the world, instead of staying to a set path. This is not surprising though, because Unigine is also using the benchmark to demonstrate the power of its graphics engine. The benchmark consists of a fly-through of the city in a total of 26 scenes, that also show time changing from night to day. When free-roaming, you have the ability to control the time of day.
The benchmark comes in three different paid editions, similar to the 3DMark benchmark, with each successive edition boasting the features of the previous:
- Presets for easy testing and comparison
- Custom settings
- GPU monitoring (version 4.0 added temperature and clock monitoring)
- Interactive mode
- Personal use allowed under the license
- Benchmark looping
- Command line automation
- CSV formatted reporting support
- Software rendering mode
- Per-frame deep analysis
- Commercial use allowed under the license
- Technical support
If there is one critique I can make about the Heaven benchmark, it is the heavy use of tessellation. By heavy, I really mean excessive, because the difference between elements with the tessellation on and off can be staggering. The most extreme examples of this are the dragon statue and the stairs. Without tessellation, the stairs are actually simple ramps without any step-geometry. With tessellation enabled, they become actual stairs. The dragon on the other hand shifts from something that you could also see as being a pet, with its relatively smooth body, into a viciously spiked nightmare.
To keep this article from being too short, the next page is going to contain graphs of the frames-per-second the benchmark recorded in its tests. Also, my system specs are listed there.
Detailed Results & Conclusion
The tests were run in succession and using the latest GeForce 314.07 drivers.
- Processor: AMD Phenom II 720 BE, 4-core, @3.40GHz (17.0x200)
- Cooling: Corsair H50 with push-pull Akasa 120mm fans
- Motherboard: ASUS M4A785TD-V EVO 785G
- Memory: Corsair Dominator 2x2 GB and G.Skill Ripjaws 2x4 GB (12 GB total) @ 1600MHZ 9-10-9-28
- GPU: EVGA GTX 570 1280 MB (797/1594/1950)
- Hard Drive: 1 x Western Digital Caviar Black 750 GB SATA
- Power Supply: Corsair 750TX
- OS: Windows 7- Home Premium 64-bit
Of course, those are just comments about the visual. As a benchmark, this software is actually quite good. Running three tests on the basic setting the scores differ by no more than two, literally increasing one point, each time I ran the test.
- API: DirectX 9
- Quality: medium
- Resolution: 1280x720, windowed mode
- Anti-aliasing: x2
- VSync: disabled
- Stereo: disabled
- Tessellation: disabled
The extreme test was similarly consistent with scores varying by no more than one point.
- API: DirectX 11
- Quality: ultra
- Resolution: 1600x900, windowed mode
- Anti-aliasing: x8
- VSync: disabled
- Stereo: disabled
- Tessellation: Extreme
By looking at the minimum/maximum/average scoring, we can analyze how much variation in FPS each preset delivers based on the system specifications.
Unsurprisingly, the minimum FPS across the Basic tests is fairly consistent, though the maximum FPS varies a bit more. Altogether, I am not surprised by the variance of the maximum FPS, because with the GPU pushing out so many frames, something that could affect its performance by just 1% would be more apparent than if fewer frames were being pushed. What is more important though, is that the minimum and average FPS numbers are consistent, as variances there would likely indicate a problem with the GPU or the benchmark.
With the Extreme tests, we find the numbers to be more consistent than the Basic tests, with the minimum FPS values differing by no more than 0.1 FPS. The maximum FPS values, though linearly closer, are actually again off by only about 1%. Actually, the Basic maximum FPS values differed from their average by only 0.9%, while the Extreme maximum FPS values differed from their average by only 1.4%. This consistency between the different settings is actually somewhat impressive, though perhaps it is not surprising, as the Basic and Extreme benchmarks are the same fundamental test, but with different effect settings and resolution.
All in all, this is a pretty solid benchmark. The only issue I could have with it is how much it feels like a tech demo for tessellation with so much geometry relying on that one feature. Of course, tessellation can really push a GPU, so it makes sense to employ that feature to test a graphics card. The consistency is a definite plus for the software also, as it keeps systematic error to a minimum in the scores you
brag about post for others to see.