I saw a BigFoot KillerNic (killernic.com) in the Microcenter the other day, and wondered how good it really was, so I bought it figuring if it sucked I could return it. It did, and I did.
I put the KillerNic up against the real world experiences I could simulate in my humble home, and compared it against my personal Favorite TOE accelerator the Alacritech’s 1000×1.
TCP/IP Offload Engines or TOE’s allow much of the CPU cycles used by network traffic to be handled by the NIC card rather than by the CPU. This allows your CPU to do the things it is good at like Physics, or Serving Web Pages.
It should be said that one of the biggest claims of the Killer Nic is something I would not pay any money for because it is irrelevant, Reduced Ping Times. Pings are a convenient indicator of the network between you and what ever you are Pinging. But lowering your ping time at the PC is kind of silly. Think of it like driving in the fog. If you had a device that could detect someone flashing their brights at you could reduce the amount of time that it took to bright back at them. But that wouldn’t mean there was less fog.
What a TOE can do is offload ACK’s and other TCI/IP functions to the Nic card. The Killer NIC is a Full Off load TOE which has some good points and some bad points. If your environment has intermittent connection establishment and termination and is prone to dropped packets, a full off-load solution would be best. For Gaming this is unlikely the case.
Alacritech’s 1000×1, automatically off-loads the data transmission/reception information or data path to the NIC while the host TCP/IP stack retains the responsibility for connection establishment and termination and error handling. This means that games and software won’t get any unexpected responses from the network stack because it is still Windows Network Stack at the core.
So the test.
Throughput on a Typical Internet Connection:
I started with some “Simple” tests. With my network impairment device hooked between my Linux NAS I attempted to move 98 Gigs of Data in a variety of sized files to a freshly formatted drive with the Impairment profile set to Docsis 2.0 8 megabit Cable modem.
This profile mirrors packet loss, Judder, packet re-ordering and other impairments.
Using the Atheros Nic on the Motherboard I averaged 6.4 Megabits per second.
The Killer Nic Scored slightly higher with 6.6 Mbps.
The Alacritech Smoked the other two averaging 7.4 Mbps.
Streaming Video Quality On a Multicast
This test uses VLC to multicast a file. I streamed a file encoded for ATT’s UVerse Product an HD h.264 file From one PC to the other over a Null Nic Cable to Limit the packet loss.
The Atheros Nic Lost .01% of packets
The Killer Nic Lost .005% of packets
The Alacritech Lost .004% of packets
I then did the test through the network impairment device adding only judder.
The Atheros Nic Lost .15% of packets
The Killer Nic Lost .07% of packets
The Alacritech Lost .01% of packets
The Alacritech destroyed the KillerNic in this test.
Setting up a fair test for this was rather hard. I wanted to pic something which would be fair, but also a little less mainstream so that it was less likely that KillerNic was doing anything to modify the results.
I set up an Unreal Tournament 2004 Match with the server on the Lan and the impairment set to the Same 8mbps Docsis 2 settings, and configured the game so that it was being played against my two Laptops. Making sure that I came in at the same spawn point I just let the Bots Play and monitored frame rates for 6 hours to make sure the results were based on a long enough run.
|Atheros||27 FPS||36 FPS||48 FPS|
|Killer||29 FPS||39 FPS||56 FPS|
|Alacritech||29 FPS||47 FPS||57 FPS|
You can see the Alacritech was amazing, boosting the the Average FPS by 33%.
The Killer definitely helped and by about the amount that people in forums and the box said about 10-15%.
All in all the KillerNic doesn’t suck and it is cheaper than the Alacritech, but Either is overkill for gaming for the extra money you’d definitely win with a better graphics card or CPU. The Only rationale that makes sense is you could take the card with you for future upgrades.
The Alacritech is Expensive, well worth it if you are doing server things. The $900 price tag can often double the throughput of a server saving you the price of the NIC just in the cost of the windows license, let alone the SQL or Media Services license.
The KillerNic didn’t appear to break any of my server applications but I didn’t get to do a lot of testing, so it is hard to tell. The Killer NIC offers one last feature which is of interest… It runs Linux on its card so you can do some interesting things with locking down a server you are shipping to someone else to use. Like write your own application to monitor bandwidth usage, for billing based on delivery, or connect to a VPN on the NIC card rather than configuring a router.