EVGA Killer Xeno Pro: The Impact of Network Offloading
by Derek Wilson on July 3, 2009 4:20 AM EST- Posted in
- Networking
Experience Testing
Because we couldn't perform as many useful repeatable tests as we wanted, we have done quite a bit of just plain gaming. We played with the hardware and without the hardware. We tested EVE Online and Team Fortress 2. Bigfoot reports that Team Fortress 2 sees some of the highest benefit from their technology, and we included EVE in order to gauge impact on network games / MMOs that were not singled out by Bigfoot. We played around with WoW for a while, but we don't have a high enough character to do anything where latency could really matter (large parties playing end-game content). These tests were done the way we normally game: with nothing running in the background and no downloading going on.
In playing on our Core i7 965 system with an NVIDIA GeForce GTX 280 and 6GB of RAM, we spent a couple hours with each game. Half of our time was with onboard networking and the other half with the Killer Xeno Pro. Both games were run at their highest quality settings and resolution on our 30" panel.
In EVE we ran some missions and got into a little PvP action. While we made more isk (EVE's in-game currency) playing with the Killer Xeno Pro, this was just the result of the missions we were handed. Neither PvE nor PvP situations felt any different with the onboard NIC versus the Killer Xeno Pro. Action was just as smooth and the UI was just as responsive no matter what was going on. We felt the same sort of loading hiccups when changing areas with both networking solutions as well: the Killer Xeno Pro just didn't deliver any tangible benefit in EVE Online.
Our Team Fortress 2 testing consisted of lots of different games played on both the on-board NIC and the Killer Xeno Pro.
We do need to preface this by acknowledging the fact that none of us are really twitch shooter experts. Sure, we all played and loved Counter Strike and CS:S, Unreal Tournament in all its incarnations, and many other FPS games, but we aren't the kind of people who run moderate resolutions with 16-bit color and most of the options turned as low as possible in order to get every single possible advantage. We are also not professional gamers; but we do love to game.
That being said, we really didn't notice any difference in our gaming experience with or without the Killer Xeno Pro. I tend to like sniping in games, and typically even non-twitch gamers can tell if they're being screwed out of kills by network issues. I didn't experience this sort of frustration with either solution. Game play was smooth and not jerky or problematic even in larger fire fights when there were no other issues at play. When playing both with and without the Killer Xeno Pro, we experienced some issues when on servers with issues.
It is just a fact that the most important factor is going to be finding a game where you and all the other players have a low latency connection to the server. The slight difference of a minimally reduced client side latency is not going to have a higher impact than any sort of other network issues.
In other words (and to sum up), when you have a bad connection, the Killer Xeno Pro is not going to fix it; when you have a good connection, the Killer Xeno Pro is not going to make the experience any better.
Facebook
Twitter
RSS
121 Comments
View All Comments
navilor - Sunday, July 5, 2009 - link
Thank you for your insight. I value your input.World of Warcraft uses TCP. I couldn't believe it as I thought it would use UDP. I ran Wireshark on my network just to make sure.
The CPU doesn't have to generate an interrupt for any packets at all when packets are processed. I believe that it would be similar interrupt coalescing but without the assumed latency increase.
has407 - Sunday, July 5, 2009 - link
Interesting.... Maybe the WoW dev's need some remedial instruction in network programming :) Even if they use TCP, they could easily segregate and prioritize time-sensitive TCP traffic. Hnmmm... so is the Killer NIC a $120 compensator for bad app design?As to whether "The CPU doesn't have to generate an interrupt for any packets at all when packets are processed."... Ummm... yes it does, in some form or another--even if it isn't a hardware interrupt--as that is how it eventually gets the app's attention (e.g., via DPC/IRP whether the result of a hardware interrupt or polling). You get one or the other: coalescing interrupts (hardware or software) and increased latency, or more interrupts and decreased latency.
I can believe that the Killer NIC driver tries to split the line between latency and interrupt overhead. I can also believe it reduces latency bit, but I find it hard to believe it reduces by a significant amount--unless of course you're doing torrents and other stuff while you're gaming--in which case the appropriate answer is: Don't do That.
Again, I assert that a properly tuned system with a decent and lower-cost NIC would fare as well. But I'd like to see some properly engineered tests to confirm that.
mindless1 - Saturday, July 4, 2009 - link
You overlook something significant. "Offloading" isn't necessarily a good thing, if the processor it's offloaded to is no faster, let alone slower, than the main CPU. I'm not suggesting it is or isn't, but the core ideal that offloading is a positive thing is quite misleading.DerekWilson - Sunday, July 5, 2009 - link
This is a good point -- if you offload it you would want to offload it to something that could do stuff faster.the issue here, though, isn't as much that the 400MHz PPC core actually be able to do the work faster than something like a Core i7 3GHz ... the major bottleneck in network processing on windows is the operating system and the software network stack ...
bypassing the OS, even though it doesn't seem to deliver a better experience from what we can tell, really does seem to be faster when using the Killer Xeno Pro ... but again the major issue is not client side processing but the rest of the network when we are talking about gaming on a standard desktop ...
swaaye - Saturday, July 4, 2009 - link
Well it's nice to hear that you THINK that you are seeing a benefit.The problem is that it's apparently rather unprovable through testing and that means that it likely is placebo effect.
I'd like to hear a network / NIC engineer chime in on these cards anonymously.
navilor - Sunday, July 5, 2009 - link
Have a friend randomly switch someone's connection from their KillerNIC to an onboard NIC while they are out of the room. Ask them if they see a difference. Repeat several times.DerekWilson - Sunday, July 5, 2009 - link
I'd want to try and make it double blind and do it with multiple people ...It'd be a great idea for a LAN party or tournament (maybe not during competition though as people would surely cry foul even if it didn't benefit anyone).
haplo602 - Saturday, July 4, 2009 - link
I see this hoax gets the trashing it deserves ... folks just google windows xp tcp tuning or similar for vista and you will find advices that put this 120$ nonsene right into the recycling bin where it belongs ...swaaye - Friday, July 3, 2009 - link
Note that even Marvell and Realtek integrated NICs have substantial hardware offload these days. I just don't see the justification at all for one of these cards....RU482 - Friday, July 3, 2009 - link
Where I work, we use a PC to serve up hard drive images to as many as 96 computers at a time (using Symantec Ghost). With a consumer grade Realtek 8169 ethernet card, we achieve around 600MB/min (that's the metric that is reported) throughput. With a card like the one in this article, could we expect to increase throughput rates...slightly or dramatically?thoughts?