Testbed Configuration and Methods
We are going to investigate the performance of EVGA Killer Xeno Pro using the following universal testbed:
- Intel Core i7-975 Extreme Edition processor (3.33 GHz, 6.4 GT/s QPI);
- Scythe SCKTN-3000 Katana 3 CPU cooler;
- Gigabyte GA-EX58-Extreme mainboard (Intel X58 Express chipset);
- Corsair XMS3-12800C9 (3 x 2 GB, 1333 MHz, 9-9-9-24, 2T);
- Samsung Spinpoint F1 HDD (1 TB, 32 MB buffer, SATA II);
- Ultra X4 850 W Modular power supply;
- Dell 3007WFP monitor (30", 2560x1600 @ 60 Hz max display resolution);
- Microsoft Windows 7 Ultimate 64-bit;
- EVGA Killer Xeno Pro 188.8.131.52-19 driver;
- Nvidia GeForce 196.21 WHQL for Nvidia GeForce;
- ATI Catalyst 10.2 for ATI Radeon HD.
In order to check out the effects of EVGA Killer Xeno Pro on the gaming performance we chose ATI Radeon HD 5850 and Nvidia GeForce GTX 285 graphics cards. The graphics card drivers were configured in the following standard way:
- Smoothvision HD: Anti-Aliasing: Use application settings/Box Filter
- Catalyst A.I.: Standard
- Mipmap Detail Level: High Quality
- Wait for vertical refresh: Always Off
- AAMode: Quality
- Other settings: default
- Texture filtering – Quality: High quality
- Texture filtering – Trilinear optimization: Off
- Texture filtering – Anisotropic sample optimization: Off
- Vertical sync: Force off
- Antialiasing - Gamma correction: On
- Antialiasing - Transparency: Multisampling
- Set PhysX GPU acceleration: Enabled
- Ambient Occlusion: Off
- Other settings: default
We used the following games:
- Battlefield: Bad Company 2
- Team Fortress 2
- World of Warcraft: Wraith of the Lich King
Unfortunately, this review does not include the highly popular Eve Online MMORPG which cannot report its response time in digital format. We selected the highest graphics quality settings in each game and also turned on 4x MSAA. Some online gamers lower the level of detail to achieve as high a frame rate as possible, but we don’t think there is any point in doing so. After all, you cannot get higher than your monitor’s refresh time which is 60, 75 or 120 Hz, whereas your visual enjoyment would suffer. We performed our tests at many popular display resolutions: 1280x1024, 1680x1050, 1920x1200 and 2560x1600. The bottom speed data are published, too. We measured the frame rate with the Fraps 3.0.2 tool manually by testing each game three times and averaging the results.
Our Internet connection was provided by Starman. It is a CATV provider that uses cable television infrastructure for the last mile. Our Internet package offers an uplink speed of 6 Mbps and a downlink speed of 350 Kbps. Using PowerBoost technology, the connection has a peak uplink bandwidth of 18 Mbps for data amounts of 10 megabytes and a peak downlink bandwidth of 1 Mbps for data amounts of 2 megabytes. These parameters are not very high for today, but the connection is stable and has low latencies, making it possible to enjoy many online games together with voice communication.
Here is how we went about testing Bigfoot Networks’ technologies: each game was tested twice – with an EVGA Killer Xeno Pro and with a mainboard-integrated Gigabit Ethernet controller Realtek 8111D. This controller is not as widespread as similar controllers from Marvell. It is linked to a PCIe 1.1 bus, supports Jumbo Frames up to 9000 bytes, and can offload the CPU from checksum computations for both TCP and UDP. Each game was tested for 15 minutes and we recorded the response time each minute. Additionally we performed the same tests with uTorrent version 2.0 (build 18488) downloading 3 large files in the background. We did not limit the bandwidth allotted to the torrent client in order to check out the performance of the Killer Xeno Pro in automatic mode.
The test situations were as diverse as possible to make them closer to real gaming scenarios. It is next to impossible to repeat the same test sequence exactly because the other players are beyond the tester’s control, especially in MMORPGs. You can’t play always in the same way even in shooters on your local network. The only way to exactly compare two network adapters would be to use them in two identical testbeds, but we have only one test configuration with top-end components. Anyway, we tried to make all test runs as similar as possible, but we must confess there is some inherent inaccuracy in our testing method.