The first test we decided to use was the Game Network Efficiency (GANE) Test, which is in place to simulate real world network performance for online PC games. This test is used to create a gaming load on a Windows PC as well as transmission of gaming network traffic over a local network. This test is also designed to test two network cards at once making direct comparisons between both at the same time.
The setup is a little complex, but ill describe it in a step by step manner.
Firstly you connect the host (listen/server) PC to a gigabit switch through the standard onboard gigabit ethernet connection. The Killer 2100 is then installed on the other test PC and it is connected to the gigabit switch via both the Killer Network gaming card and the onboard NIC.
On the server PC we then configure the properties of the onboard network card to have a static IP address. We then adjust the subnet mask to 255.255.255.0. The test PC is then configured in a similar manner, but we assign different static IP’s to both the onboard NIC and the Killer 2100. Again the subnet mask is set to 255.255.255.0.
On the main test PC we then go into advanced via the IPv3 properties window. We then uncheck the ‘Automatic Metric” box and set the metric value to “1”. Both cards need this applied as the NIC uses it to send network traffic. If we leave this setting at ‘automatic’, Windows does not alternate between the NICs and instead will send all the traffic through one card rendering the test useless.
The next stage is to run GANE on both the server/listen PC and the main test rig. GANE measures then compares latency between two network cards installed on any PC. This is handled by a procedure of sending 100 byte packets over the local network on a round trip, every 50ms. Kitguru has selected 100bytes as the packet size because this is a good real world representation of a standard network packet. We want to run this size specifically to test Bigfoots claims that many network cards are not optimised for this ‘game’ related data packet size.
While we set up the server PC we also need to run a game benchmark on the main PC and in this case we will use Resident Evil 5 in DX10 mode with the built in benchmark at maximum settings on our screen at 1920×1200 resolution. While we are running this, we are sending packets between both NICs on the main PC to the receiving (listening) PC. This mirrors a real life situation of playing a game online while transmitting data back and forward.
Adapter 1 is the Killer 2100 and Adapter 2 is the onboard Intel solution. The results above show that the Killer 2100 is 4.0 faster than the on board solution and delivers a result with almost 25 times less jitter. On a lesser solution this can actually be as high as 35 times faster with 120 times less jitter!
Our average UDP ping was 0.000287643 on the Killer 2100 and it was 0.003075006 on the Intel solution. Our average mean ping was 0.232525 on the Killer 2100 and 0.921490 on the Intel solution. The worst case scenario is more critical as it could cause lag online. The Intel solution is 5.287735 while it is 0.355063 on the Killer 2100 … massive differences. Bear in mind this is obviously over a very tight local network, but when heading online to game on a server thousands of miles away this will increase, exponentially.