Where do we come up with that "Ten Times Faster than a Regular NIC" claim?

Introducing the Game Networking Efficiency (GaNE) Test

GaNE was developed as a performance tool at Bigfoot Networks to answer a simple question – how long does it take a game data packet to get out of your PC and out to the network?  After much searching, we found that no such tool existed, so we wrote our own.

GaNE measures Game Data Latency.  The test computer contains two network cards, and the “listen” computer listens for the data that these cards transmit.  The test and server computers are connected via a Gigabit network consisting of Cat 5e or Cat 6 cables and a Gigabit switch.

After we set it up, we run GaNE to transmit a 100 byte packet every 50 milliseconds from each NIC installed in the TEST machine.  We use 100 bytes because that’s the average size of an online game packet.  We use 50 milliseconds because that’s the average amount of time between packets that many online games use. 

To run the test properly, we need to apply a gaming load to the TEST PC.  We usually run a gaming benchmark like the Resident Evil 5 benchmark to fully load the TEST PC.

By now, you’ve seen the results of these tests elsewhere, but if this is your first time, see the graph below.

The blue line represents the latency of a standard motherboard network controller during a game benchmark.  As you can see, it averages 6.04 milliseconds, but spikes up to 10.5  milliseconds at least once during the 90 second test.  In fact, it’s difficult to predict when a standard NIC will actually swing into high latency – repeated runs of this test will generate different values.

The red line represents the UDP latency of a Killer™ 2100.  As you can see, its average latency is 251 MICROseconds, with spikes no higher than 317 microseconds.

These test results are remarkable.  This shows that:

-    Latency in game networking conditions is an erratic, noisy affair, where the actual latency generated by your network card in no way resembles the “ping” number you see in your game.

-    Latency in standard NICs swings wildly compared to what we’ve previously understood.

-    The advantages provided by Killer™ 2100 are unbeatable.  In the above case, almost 24 times faster than the measured onboard NIC.

There’s a bigger effect.  We’ve measured that the first packet on the wire is the first packet at the server.  In-game, if you and your opponent pull the trigger at the same time, the Killer Gaming Network Card will get that trigger pull out to the network faster.  On a LAN, you win.  If the server is on the Internet and you both have the same latency to that server, you win.

And if you think 12 millisecond spikes aren’t bad, remember that 12 milliseconds is equal to one frame if you’re playing at 60 FPS.  And there are thousands of matches and raids won and lost every day because of one- or two-frame kills.

Let’s have some fun with this.  Imagine you and a friend are sitting in London at identical PCs, except you have a Killer™ 2100 installed.  You’re connected to the same gigabit Ethernet switch.  This gigabit Ethernet switch is connected to an Ethernet cable that connects to Oslo, Norway.  When you both click your mouse buttons, one packet goes from each PC to the big Ethernet cable.  Your packet leaves in about 250 microseconds.  Your friend’s packet leaves about 5.79 milliseconds later.  That means that your packet would already be in Oslo, 718 miles away, before your friend’s packet even left his computer.  Pack your mittens.*

Over the coming months, we’ll be giving the GaNE test software to more and more technical publications and websites so that they can show you the results for themselves. 

*assuming conservative propagation delay in Cat 5e cable of 4.8 nanoseconds per meter, the actual packet would be 751 miles down the cable after the 5.79 millisecond gap.