Benchmarks
These PC laptops from Cyberpower were powered by an Intel i7 2630QM 2 GHz quad core processor, 8 GB DDR3 RAM, a Seagate ST9500420AS 500 GB HDD, and an Nvidia GeForce GTX 460M running a 1080p display. Windows 7 Ultimate 64 bit is the OS of choice. The sole difference between them is that one was connected to the network using the Killer Wireless-N 1102 NIC and the other was connected using an Intel Centrino Ultimate-N 6300 AGN NIC. Both were connected via mini PCI-Express.

We ran two synthetic benchmarks provided by Bigfoot Networks, plus a file transfer test. We also played some games online and at two LAN parties to test the network gaming performance.

For most of our benchmarking, both laptops were connected wirelessly to an ASUS RT-N56U dual-band Gigabit Wireless-N router. Both were reporting a 300 Mbps connection to router via the 5 GHz network. They were positioned approximately 10 feet from the router and had direct line of sight to it. All other wireless devices, including our smartphones, were turned off in order to reduce interference.

As a wired host for our tests, we used one of our standard high-end gaming desktop test rigs. It was connected directly to the ASUS RT-N56U router via Cat5e Ethernet cabling. Windows 7 reported a connection speed of 1.0 Gbps.

The first of the tests was GANE, BFN’s Gaming and Network Efficiency test.

Directly from BFN:

• GANE measures and compares the latency between two networked PCs. It does this by sending 100 byte packets over the local network on a round trip every 50 ms. We selected 100 bytes as the packet size because this is the typical size of a game network packet.
• We selected 50ms as the interval for sending these packets because this is a typical interval time for most game traffic. These are important settings, because, as you’ll discover in your evaluation, many network modules are not optimized for this sort of game network traffic.

First, we executed the GANE test with nothing running in the background. We ran it several times.

At its worst performance, the Killer laptop was 5.2% faster. At its best, though it was 83.7% faster.

Raw data!

You can see here that in the worst case, the min run, the jitter was higher. That means that the measure of latency for the Killer Wireless-N was more inconsistent than the Intel NIC. However, in the best case, it was 50% more consistent. Having low latency sporadically doesn’t mean anything. Such can actually be more frustrating than a higher, yet consistent latency. having consistently low latency is what really matters, and the Killer NIC is king in that kingdom.

I conducted the same test again, but this time as the YouTube test! To simulate some multitasking, I watch a few videos during test, changing resolutions to 720p from 360p. I watched pieces of four videos. The Killer performance is spectacular, and really shows potential here. There are a lot of MMO gamers who play while doing other things — why should their ping suck because they want to listen to streaming audio or video while farming?

The Killer Wireless-N had a 4.733 ms mean ping, 21.04 ms average of the worst 10%, while the Intel card had a 6.854 ms mean ping, 19.78 ms average of the worst 10%. Thus, the Killer Wireless-N was 44.8% faster, with 41.4% less jitter.

To really understand the value of such lower jitter on at least a theoretical level, imagine a platform game like Mario or Sonic the Hedgehog. Imagine that you’re trying to time a jump onto moving platforms, with lava or spikes beneath. If they’re moving really quickly or really slowly, you can eventually time it right. If they’re moving really quickly then suddenly slow or stop, then move slowly, then stop, then move really fast, then stop, then slow, and so on, you’ll get frustrated by the inconsistency and probably throw the controller through the television.

The next test is the synthetic Netperf test. This test simply rams data down the pipe as fast as it can; it’s a throughput test. The numbers presented below are measured in Mbps, each figure being one run.

TCP is the protocol used for transfers which must be 100% guaranteed to arrive, such as file transfers. UDP, on the other hand, is connection-less. That is, a packet is sent and forgotten about. This is useful in applications where timeliness is more important than complete data, such as gaming and streaming media.

netperf:
- TCP, 20 secs
K: 161.00 157.22 160.11 164.64
I: 138.79 129.26 137.25 129.87
- TCP, 120 secs
K: 162.28
I: 128.62
- UDP, 20 secs
K: 39.33 38.82 38.74 38.50
I: 3.29 3.81 3.32 3.81
- UDP, 120 secs
K: 39.41
I: 3.69

It’s clear to see that the Killer NIC tops the Intel by 36 Mbps in the TCP tests. However, the astounding factor here is the UDP performance: the Killer Wireless-N is more than 10 times more performant than the Intel card.

The file transfer test focuses on upload throughput. The numbers are listed in Mbps, with three indicative runs shown.

FTT write
Killer
98.01 97.45 97.64
Intel
106.45 109.93 107.38

Here, we can see that the Intel card is still king of straight-up file transfers. This has always been the case in such comparisons. It’s important to note that with each release of the Killer NIC drivers or hardware revision, the gap continues to close.