Use & Benchmarks
I performed benchmarking and monitoring on Windows 7 64-bit after installing the Killer 2100 into my standard testing rig with an ASUS M3A32-MVP motherboard, Athlon X2 6000+, 8 GB of DDR2 RAM, three SATA hard drives, a Kingston V-Series 128 GB SDD, and a Foxconn-made nVidia 8800 GTX inside a Cooler Master Cosmos S case.
I played Left 4 Dead 2 and a few other games, and noticed that my framerates were slightly higher. When combined with other activities in the background, such as streaming music or transferring a file, the 2100′s performance was consistent, whereas the onboard NIC’s performance would drop in and out.
I conducted six tests: two games, two file transfer tests, and three synthetic tests Bigfoot likes to use to show the differences between onboard NICs and Killer NICs.
Where stated, the onboard NIC in use is a Marvell Yukon 88E8056 PCI-Express Gigabit Ethernet Controller. I’ve also included results for the games and the file transfer test from testing with a Killer Xeno Pro, so that folks can see the progress which Bigfoot Networks has made in a year’s time.
Simply put, the purpose of this card is to improve performance by reducing lag. Its network traffic offloading affects two key factors of gaming: latency and frame rate. Reducing the load on the CPU because of I/O and network calculations increases the amount of time the CPU can devote to drawing frames.
The games I used were Team Fortress 2, a game which Bigfoot Networks uses in its own testing, and Unreal Tournament 3, a game the performance with which I am quite familiar on my computer. I played several maps with the onboard, switched to the Killer NICs, and switched back and forth once more. I played on the same server using the same settings, same character type, and same general playing style. I used Fraps to count framerate for UT3 and used the game’s own network traffic overlay to watch the ping. I used the network overlay built into TF2 to watch both statistics.
|TF2||Onboard NIC||Killer 2100||Killer Xeno Pro||2100 vs Onboard|
|FPS||58 fps||90 fps||80 fps||55% more|
|Ping||40 ms||40 ms||40 ms||0%|
|UT3||Onboard NIC||Killer 2100||Killer Xeno Pro||2100 vs Onboard|
|FPS||30 fps||35 fps||33 fps||16% more|
|Ping||80 ms||40 ms||75 ms||50% less|
Both games saw a considerable framerate improvement. UT3 saw a great ping improvement. My rig handles TF2 without a problem normally, but usually chugs a little on UT3, frequently dropping into the 20 FPS region. I saw this far less often while using the Killer NIC.
File Transfer Tests
The file transfer benchmarks were conducted copying a 1.33 GB file from my QNAP TS-809 Pro with four hard drives in RAID5 through my Dell gigabit switch. The file was read from the NAS via SMB (and then again via NFS just for the 2100 and onboard) to a RAMDisk on my rig. It was then written back to the NAS.
|SMB File Transfer||Onboard NIC||Killer 2100||Killer Xeno Pro||2100 vs Onboard|
|Read||88 MB/s||75 MB/s||11.3 MB/s||15% slower|
|Write||85 MB/s||73 MB/s||11.5 MB/s||15% slower|
It’s on this test where one can see the most improvement between the Killer Xeno Pro and the Killer 2100. Bigfoot Networks acknowledged the problem with the Xeno Pro’s TCP throughput and fixed it in the Killer 2100. It’s a combination of a software and hardware fix, reports VP Marketing John Drewry, so Killer Xeno Pro users should be able to upgrade to the latest firmware, as well, and see an improvement, as well.
|NFS File Transfer||Onboard NIC||Killer 2100||2100 vs Onboard|
|Read||10.2 MB/s||14.2 MB/s||1.39x faster|
|Write||9.9 MB/s||5.8 MB/s||41% slower|
I used NFS for a test because NFS uses UDP for transport. The Killer 2100 is designed to accelerate UDP, so I hoped there would be at least some speed gain. The gain of 39% was less than for which I’d hoped, but it’s still significant.
From an enterprise perspective, if Bigfoot Networks could design a version of the Killer NIC for NFS clients and servers and likely really sell it as a network performance enhancing tool. It could be through this that Linux gamers might see a trickle-down: NFS is a primarily Unix/Linux thing, so a driver would be necessary to run the Killer NIC in Linux at the enterprise level and the enthusiast level.
I used ProNFS, a simple NFS drive mapping tool for Windows. I would have used the built-in Services for Unix if my Windows 7 was the Professional edition.
BFN GaNE Test
The synthetic Gaming Network Efficiency (GaNE) test simulates a network game by doing two things simultaneously: run a system-intensive benchmark and send UDP packets through the onboard NIC and the Killer NIC simulaneously. The packets are sent in such a way which mimics how games would actually send packets during real online play.
My rig was configured to send packets to a listening computer, a recently-reviewed Lenovo ThinkPad Edge 13", while running the Resident Evil 5 benchmark.
This screenshot shows the graph on the ThinkPad near the end of the test.
Here’s a full graph showing the ping times throughout the test.
The GaNE program itself spits out a nice little alert message with the results following the conclusion of the test.
For the 1% of our users not using a graphical browser, here’s the results in beautiful tabular format.
|GaNE Test||Onboard||Killer 2100||Difference|
|Mean Ping||14.867 ms||1.692 ms||8.79x faster|
|Standard Deviation||14.324 ms||9.651 ms|
|Average of worst 10%||46.298 ms||12.148 ms||3.811x faster|
|Summary||Killer 2100 was 8.8x faster with 1.5x less jitter.|
BFN Netperf Test
The Netperf test tests both TCP and UDP throughput, simply outputting as fast as possible to a remote host.
|Netperf||Onboard NIC||Killer 2100||2100 vs Onboard|
|TCP||939.28 Mbps||879.815 Mbps||6.3% slower|
|UDP||43.3275 Mbps||848.885 Mbps||19.6x faster|
From this, it’s obvious that the TCP performance of the Killer 2100 isn’t quite as high as the onboard, but that the UDP performance is mind-bogglingly higher. This shows just how effective the Killer NIC could be if pushed to its limits. It’s fairly safe to assume that TCP performance will improve as new drivers are released.
BFN File Transfer Test
BFN’s file transfer test is a simply copy via SMB from another computer using a 2,095,804 KB file. This test took 121 seconds for the onboard and 120 seconds for the Killer 2100, about 17 MB/s. I’m not quite sure how this test was so much slower than my similar testing, but the results are close enough that this test shows only parity.
Colin Dean has been a writer for ThinkComputers since 2006.