Having tested three original GeForce GTX 780 Ti graphics cards from Zotac, MSI and Inno3D, we couldn’t help checking them out in SLI configurations. We were not able to combine all of them into a single subsystem though, just because we only had two of them at one time. Testing a 3-way SLI configuration (with a 3-monitor setup) is still on our plans, but today we want to share with you the results of our testing a SLI configuration built out of two GTX 780 Ti cards in comparison with five opponents, two of which are dual-GPU solutions.
Testbed and Methods
Here is the list of components we use in our testbed.
- Mainboard: Intel Siler DX79SR (Intel X79 Express, LGA 2011, BIOS 0590 dated 17.07.2013)
- CPU: Intel Core i7-3970X Extreme Edition 3.5/4.0 GHz (Sandy Bridge-E, C2, 1.1 V, 6x256KB L2 cache, 15MB L3 cache)
- CPU cooler: Phanteks PH-TC14PЕ (2xCorsair AF140 fans, 900 RPM)
- Thermal grease: ARCTIC MX-4
- Graphics cards:
- AMD Radeon HD 7990 (2x3GB, 1100/6300 MHz)
- Nvidia GeForce GTX 690 (2x2GB, 1006-1111/7208 MHz)
- Nvidia GeForce GTX Titan (6GB, 837-876/6008 MHz and 1006-1045/7200 MHz)
- Inno3D iChill GeForce GTX 780 Ti HerculeZ X3 Ultra (3GB, 1006-1072/7200 MHz)
- Nvidia GeForce GTX 780 Ti (3GB, 1006-1058/7200 MHz)
- ASUS ROG MARS 760 (2x2GB, 1006-1072/6008 MHz)
- AMD Radeon R9 290X (4GB, 1000/5000 MHz)
- System memory: DDR3 4x8GB G.SKILL TridentX F3-2133C9Q-32GTX (XMP: 2133 MHz, 9-11-11-31, 1.6 volts)
- System disk: SSD 256GB Crucial m4 (SATA 6 Gbit/s, CT256M4SSD2, BIOS v0009)
- Games/software disk: Western Digital VelociRaptor (SATA-2, 300 GB, 10000 RPM, 16 MB cache, NCQ) in a Scythe Quiet Drive 3.5" enclosure
- Backup disk: Samsung Ecogreen F4 HD204UI (SATA-2, 2 TB, 5400 RPM, 32 MB cache, NCQ)
- Sound card: Auzen X-Fi HomeTheater HD
- Computer case: Antec Twelve Hundred (front panel: three Noiseblocker NB-Multiframe S-Series MF12-S2 fans at 1020 RPM; back panel: two Noiseblocker NB-BlackSilentPRO PL-1 fans at 1020 RPM; top panel: one preinstalled 200mm fan at 400 RPM)
- Control & monitoring panel: Zalman ZM-MFC3
- Power supply: Corsair AX1200i (1200 W), 120mm fan
- Monitor: 27" (DVI-I, 2560x1440, 60 Hz)
Our SLI configuration was built out of a reference GeForce GTX 780 Ti (in the mainboard's top PCIe slot) and an original GTX 780 Ti from Inno3D:
The clock rates of the reference 780 Ti are increased to those of the Inno3D. Two dual-GPU cards from AMD and Nvidia are included into this testing for the comparison’s sake:
Their GPUs are overclocked by about 100 MHz. The memory clock rate of the Radeon HD 7990 is increased to the maximum of 6300 MHz. The memory frequency of the GeForce GTX 690 is set at 7200 MHz, the same as with the GTX 780 Ti configuration.
Besides the dual-GPU solutions, we also include an Nvidia GeForce GTX Titan 6GB (at 1006-1045/7200 MHz) and an AMD Radeon R9 290X 4GB (at its default clock rates and 70% fan speed to avoid GPU throttling).
So, we’ve got quite a comprehensive selection of graphics solutions. We only miss a CrossFireX tandem based on Radeon R9 290X cards but their availability is still low, so we have to postpone such testing for now.
Besides the abovementioned cards, we also have a dual-processor ASUS ROG MARS 760 2x2GB working at its default clock rates:
This model is no competitor to the GeForce GTX 780 Ti and we have already compared it with the other cards, yet it will help make our picture more complete.
In order to lower the dependence of the graphics cards’ performance on the overall platform speed, we overclocked our 32nm six-core CPU to 4.8 GHz by setting its frequency multiplier at x48 and enabling Load-Line Calibration. The CPU voltage was increased to 1.38 volts in the mainboard’s BIOS:
Hyper-Threading was turned on. We used 32 GB of system memory at 2.133 GHz with timings of 9-11-11-20_CR1 and voltage of 1.6125 volts.
The testbed ran Microsoft Windows 7 Ultimate x64 SP1 with all critical updates installed. We used the following drivers:
- Intel Chipset Drivers – 126.96.36.1996 WHQL dated 21.09.2013
- DirectX End-User Runtimes, dated 30 November 2010
- AMD Catalyst 13.12 WHQL (188.8.131.52) dated 18.12.2013
- Nvidia GeForce 331.93 Beta from 27.11.2013 for the GeForce GTX series and GeForce 334.67 Beta from 27.01.2014 for the GeForce GTX Titan
We benchmarked the graphics cards’ performance at two display resolutions: 1920x1080 and 2560x1440 pixels. There were two visual quality modes: “Quality+AF16x” means the default texturing quality in the drivers + 16x anisotropic filtering whereas “Quality+ AF16x+MSAA 4x(8x)” means 16x anisotropic filtering and 4x or 8x antialiasing. In some games we use antialiasing algorithms other than MSAA as indicated below and in the diagrams. We enabled anisotropic filtering and full-screen antialiasing from the game’s menu. If the corresponding options were missing, we changed these settings in the Control Panels of the Catalyst and GeForce drivers. We also disabled Vsync there. There were no other changes in the driver settings.
The graphics cards were tested in two benchmarks and 13 games updated to the latest versions.
- 3DMark (2013) (DirectX 9/11) – version 184.108.40.206: Cloud Gate, Fire Strike and Fire Strike Extreme scenes.
- Unigine Valley Bench (DirectX 11) – version 1.0: maximum visual quality settings, 16x AF and/or 4x MSAA, 1920х1080.
- Total War: SHOGUN 2 – Fall of the Samurai (DirectX 11) version 1.1.0: integrated benchmark (the Sekigahara battle) with maximum visual quality settings and 8x MSAA.
- Sniper Elite V2 Benchmark (DirectX 11) – version 1.05: we used Adrenaline Sniper Elite V2 Benchmark Tool v220.127.116.11 BETA with maximum graphics quality settings (“Ultra” profile), Advanced Shadows: HIGH, Ambient Occlusion: ON, Stereo 3D: OFF, Supersampling: OFF, two sequential runs of the test.
- Sleeping Dogs (DirectX 11) – version 1.5: we used Adrenaline Sleeping Dogs Benchmark Tool v18.104.22.168 with maximum image quality settings, Hi-Res Textures pack installed, FPS Limiter and V-Sync disabled, two consecutive runs of the built-in benchmark with quality antialiasing at Normal and Extreme levels.
- Hitman: Absolution (DirectX 11) version 1.0.447.0: built-in test with Ultra settings, enabled tessellation, FXAA and global lighting.
- Crysis 3 (DirectX 11) – version 22.214.171.1240: maximum visual quality settings, Motion Blur – Medium, lens flares – on, FXAA and MSAA 4x, two consecutive runs of a scripted scene from the beginning of the “Swamp” mission (110 seconds long).
- Tomb Raider (2013) (DirectX 11) – version 1.1.748.0: we used Adrenaline Benchmark Tool, all image quality settings set to “Ultra”, V-Sync disabled, FXAA and 2x SSAA antialiasing enabled, TessFX technology activated, two consecutive runs of the in-game benchmark.
- BioShock Infinite (DirectX 11) – version 126.96.36.19918: we used Adrenaline Action Benchmark Tool with “Ultra” and “Ultra+DOF” quality settings, two consecutive runs of the in-game benchmark.
- Metro: Last Light (DirectX 11) version 188.8.131.52: we used the built-in benchmark for two consecutive runs of the D6 scene. All image quality and tessellation settings were at “Very High”, Advanced PhysX technology enabled, with and without SSAA antialiasing.
- GRID 2 (DirectX 11) – version 184.108.40.20679: we used the built-in benchmark, the visual quality settings were all at their maximums, the tests were run with and without MSAA 8x antialiasing with eight cars on the Chicago track.
- Company of Heroes 2 (DirectX 11) – version 220.127.116.1111: two consecutive runs of the integrated benchmark at maximum image quality and physics effects settings.
- Total War: Rome II (DirectX 11) version 1.8.0 build 8891.481024: Extreme quality, V-Sync disabled, SSAA enabled, two consecutive runs of the integrated benchmark.
- Batman: Arkham Origins (DirectX 11) version 1.0 update 8: Ultra visual quality, V-Sync disabled, all the effects enabled, all DX11 Enhanced features enabled, Hardware Accelerated PhysX = Normal, two consecutive runs of the in-game benchmark.
- Battlefield 4 (DirectX 11) – version 1.4: Ultra settings, two successive runs of a scripted scene from the beginning of the “Tashgar” mission (110 seconds long).
We publish the bottom frame rate for games that report it. Each test was run twice, the final result being the best of the two if they differed by less than 1%. If we had a larger difference, we reran the test at least once again to get repeatable results.