Testbed and Methods
First, let me say a few words about the hardware that we used during this test session. All graphics cards were benchmarked in a closed system case with the following configuration:
- Mainboard:ASUS P6T Deluxe (Intel X58 Express), LGA 1366, BIOS 1701;
- CPU: Intel Core i7-920, 2.67 GHz, 1.25 V, L2 4 x 256 KB, L3 8 MB (Bloomfield, C0);
- CPU cooler: Thermalright IFX-14 (two Thermalright TR-FDB fans at 1320 RPM);
- Thermal interface: Tuniq TX-3;
- System memory: DDR3 PC3-12800 3 x 2GB OCZ Platinum Low-Voltage Triple Channel (Spec: 1600 MHz / 7-7-7-24 / 1.65 V);
- Disk subsystem: Western Digital VelociRaptor (300GB, SATA-II, 10000 RPM, 16MB cache, NCQ) inside Scythe Quiet Drive 3.5” HDD silencer and cooler;
- Backup HDD: Western Digital Caviar Green WD10EADS (SATA-II, 1000 GB, 5400 RPM, 32 MB, NCQ);
- Optical drive: Samsung SH-S183L DVD-burner;
- System case: Antec Twelve Hundred (front panel: two Noiseblocker NB-Multiframe S-Series MF12-S1 fans at 900 RPM and Scythe Gentle Typhoon fan at 900 RPM; back panel: two Scythe SlipStream 120 fans at 900 RPM; top panel: standard 200 mm fan at 400 RPM);
- Control and monitoring panel: Zalman ZM-MFC2
- Power supply: Zalman ZM1000-HP 1000 W (with a default 140 mm fan).
To minimize the platform influence on the performance of the tested graphics cards I overclocked our 45 nm quad-core CPU with the multiplier set at 21x and “Load-Line Calibration” enabled to 4.1GHz. The processor Vcore was increased to 1.3825V in the mainboard BIOS:
The system memory worked at 1.56 GHz frequency with 7-7-7-14_1T timings and 1.64V voltage:
All other parameters available in the mainboard BIOS and connected with CPU or memory overclocking remained unchanged (set to Auto).
We are going to compare the performance of our Radeon HD 5850 and HD 5870 against that of two other graphics solutions on ATI chips: Diamond Radeon HD 4890 1 GB XOC (working at the reference frequencies for HD 4890) and HIS Radeon HD 4870 X2 2 x 1 GB:
All graphics accelerators except Radeon HD 5850 were tested in nominal modes. Radeon HD 5850 was tested at its default frequencies of 725 / 4000 MHz and at 850 / 4800 MHz that are the frequencies of Radeon HD 5870. This way we could check out the performance difference between the GPUs of these two graphics accelerators. Moreover, Radeon HD 5850 was also tested at maximum overclocked frequencies of 1010 / 4960 MHz in order to see if this graphics card will be able to outperform more expensive competition due to overclocking. We do understand clearly that other graphics cards can also be overclocked and hence run faster, but it wasn’t our goal to compare the performance of all solutions during overclocking today. That is why only our key hero got this “special treatment”.
Now let’s move on to software and benchmarking tools that we used. All tests were performed under the new Windows 7 Ultimate RTM x64 operating system with the following drivers:
- Intel Chipset Drivers 184.108.40.2060 WHQL for the mainboard chipset;
- DirectX End-User Runtimes from August 2009;
- Catalyst 9.11 beta 4 graphics card drivers for ATI based graphics solutions;
- GeForce/ION Driver 191.07 WHQL graphics card drivers for Nvidia based graphics solutions.
The graphics cards were tested in two resolutions: 1680x1050 and1920x1200. We believe that it doesn’t make sense to test these graphics accelerators in 1280x1024, because the platform performance will be a limiting factor in most cases. Our monitor didn’t support resolutions above 1920x1200, but it is a minor issue, because very few gamers use higher screen resolutions anyway and the tested graphics cards cannot provide sufficient performance to ensure comfortable gaming experience in 2560x1600 resolution.
The tests were performed in two image quality modes: “High Quality” without any image quality enhancements and “HQ+ AF16x+AA4/8x” with enabled 16x anisotropic filtering and 4x full screen anti-aliasing (or 8x FSAA if the average framerate was high enough for comfortable gaming experience). We enabled anisotropic filtering and full-screen anti-aliasing from the game settings or configuration files. If the corresponding options were missing, we changed these settings in the Control Panel. Vertical sync was always off in driver control panels.
All games were reinstalled fresh under the new operating system and updated with the latest patches available at the time of tests. So, the complete list of test applications includes three popular synthetic benchmarking suites and 17 games of various genres, including such new titles as Wolfenstein, Batman: Arkham Asylum and Resident Evil 5. Here is the complete list of tests used with the settings (all games listed in their release order):
- 3DMark 2006 (Direct3D 9/10) – build 1.1.0, default settings and 1920x1200+AF16x+AA8x;
- 3DMark Vantage (Direct3D 10) – v220.127.116.11, Performance and Extreme profiles (basic tests only);
- Unigine Heaven Demo (Direct3D 11) – version 1.0, maximum graphics quality settings including shadows, activated tessellation;
- World in Conflict (Direct3D 10) – version 18.104.22.168 (b34), “Very High” graphics quality profile, UI texture quality = Compressed; Water reflection size = 512, other settings – by default;
- Crysis (Direct3D 10) – game version 1.2.1, “Very High” settings profile, two runs of “Assault harbor” test from Crysis Benchmark Tool version 22.214.171.124;
- Unreal Tournament 3 (Direct3D 9) – version 2.1, highest graphics quality settings (level 5), Motion Blur and Hardware Physics enabled, a FlyBy of the “DM-ShangriLa” map (two consecutive cycles) using HardwareOC UT3 Bench v126.96.36.199;
- Lost Planet: Colonies (Direct3D 10) – version 1.0, Maximum Quality settings, DX10 HDR Rendering, integrated benchmark including two scenes, but the results are provided from the first scene only (ARENA 1);
- S.T.A.L.K.E.R.: Clear Sky (Direct3D 10.1) – game version 1.5.10, Improved Full DX10 Lighting profile plus 16x anisotropic filtering and other maximum graphics quality settings, my own s04 demo record (a triple run of the test) on the first gaming level;
- Far Cry 2 (Direct3D 10) – version 1.03, Ultra High settings profile, two runs of the Ranch Small test from Far Cry 2 Benchmark Tool (v188.8.131.52);
- Call of Duty 5: World at War (Direct3D 9) – version 1.6, graphics and textures are set at “Extra” level, Breach demo from the same-name level;
- Left 4 Dead (Direct3D 9) – version 184.108.40.206 b3939, maximum quality, new d6 demo (two runs) on “Lighthouse” map in “Survival” game mode;
- Cryostasis: Sleep of Reason (Direct3D 10.1) – version 1.0.2, shader model 4.0, high image quality settings, “Caustic” enabled, tested in own “d3” demo at “11. Fear” level;
- Warhammer 40,000: Dawn of War II (Direct3D 10.1) – version 220.127.116.1134, image quality settings set to Ultra level in the game menu, two runs of built-in benchmark;
- BattleForge (Direct3D 11) – version 1.1, maximum image quality settings, shadows enabled, SSAO technology disabled, two runs of the built-in benchmark;
- Stormrise (Direct3D 10.1) – version 18.104.22.168, maximum effects and shadows quality, Ambient Occlusion disabled, two runs of the “$mn_sp05” mission demo scene;
- Tom Clancy’s H.A.W.X. (Direct3D 9) – version 1.03, maximum graphics quality settings; HDR, DOF and Ambient Occlusion enabled, two runs of the built-in benchmark;
- Call of Juarez: Bound in Blood (Direct3D 10.1) – version 22.214.171.124, maximum graphics quality settings, Shadow map size = 1024, 100-second demo in the beginning of “Miners Massacre” level;
- Wolfenstein (OpenGL 2.0) – version 1.2, maximum graphics quality settings, own d1 demo recording on Facility level;
- Batman: Arkham Asylum (Direct3D 9) – version 1.1, maximum level of detail, physics enabled, two runs of the built-in benchmark;
- Resident Evil 5 (Direct3D 10.1) – variable benchmark with maximum graphics quality settings without motion blur, we took AVG values from the third scene for further analysis, because it was the most resource hungry.
Here I’d like to add that if the game allowed recording the minimal fps readings, they were also added to the charts. We ran each game test or benchmark twice and took the best result for the diagrams, but only if the difference between them didn’t exceed 1%. If it did exceed 1%, we ran the tests at least one more time to achieve repeatability of results.
Now let’s talk about the actual benchmarks.