When we sat down to start testing the Killer NIC, we had grand plans for glorious graphs that would illustrate what—if any—impact the card had on in-game frame rates and ping times. Unfortunately, getting reliable data proved problematic. Online multiplayer games have a high degree of variability, which does wonders for replay value, but also makes gathering consistent data difficult. To get consistent data, you need to play the same games in the same way multiple times with multiple configurations. Despite repeated efforts, most of the games we tried would only produce consistent frame rates for two or three out of five test runs. The results that didn't match tended to be all over the map, leaving us with little confidence in the results overall.
Gathering reasonable ping data was even more difficult. The ping data had to come from inside of a game in order to be relevant. When trying to capture frame rate data, we had FRAPS doing most of the work, but we had no such tool for recording in-game ping times. Collecting ping data required manually recording ping times displayed in the games themselves. This job would be easy if all games had ping or latency counters in their HUD. Most don't, and some that do—I'm looking at you, Counter-Strike: Source—display different ping data in the HUD than they do in the scoreboard. We wanted to pull ping data every 10 to 15 seconds, but having to bring up a scoreboard that often was so disruptive to gameplay that we scuttled the idea.
So instead of presenting all sorts of objective measures of the Killer NIC's impact on game performance, I'm going to talk about my subjective perceptions: how gaming with the Killer NIC feels different than gaming with alternatives from Marvell and Nvidia. That is, after all, the $250 question about the Killer NIC.
I should note up front that it was very difficult to discern differences in game performance, lag, and overall responsiveness between the nForce and Yukon network controllers. Both use the standard Windows network stack and neither are highly optimized to reduce latency, so that isn't an unexpected result. The Killer NIC is something of a different story.
I'm probably more familiar with this game than any other recent multiplayer title, so it's a good place to start. Battlefield 2 is full of frantic gameplay, particularly on servers packed with 64 players, so there's no shortage of action at any given time. With so much going on, lag is easy to spot, and it often gets you killed.
Here, the Killer NIC felt right at home. Lag wasn't completely eliminated, particularly on 64-player servers, but episodes of stuttering and jerkiness were definitely reduced. In particular, the Killer NIC seemed to deal with artillery strikes and multiple explosions (thanks, grenade spammers) better than the other networking solutions. That made the game feel more fluid overall, if only because there were fewer interruptions to its normal level of responsiveness. The difference wasn't night and day, though. Battlefield 2 is perfectly playable with the nForce and Yukon GigE controllers; the Killer NIC just suffered from fewer lag episodes, and those that did strike felt less severe.
Lag has probably been blamed for more deaths and poor performances in Counter-Strike than any other game, making this title ripe for the Killer NIC. I used to be a regular CS player back in the day—well, it was up until around Beta 5, so make that way back in the day—so I was in somewhat unfamiliar territory with the latest Source incarnation. There seems to be a lot more rushing these days, and in those rushes, the Killer NIC showed the most promise.
Perhaps because it's largely confined to less expansive environments where less is going on in one's immediate vicinity, there wasn't as much lag in Counter-Strike as in Battlefield 2. Still, when charging guns blazing into a large group with grenades and furniture flying left and right, lag would occasionally rear its ugly head. And it's there that the Killer NIC suffered from less jerkiness and stuttering than the nForce or Yukon network controllers. Not much less, but enough that I noticed.
There was another effect here that was more subtle than a reduction in the frequency and severity of stuttering or jerky gameplay. At times when playing on the Killer NIC, typically in large crowds or with lots of action on the screen, the controls felt just a smidge more connected and responsive, even when there was no obvious lag. This reminded me a little of the difference in control responsiveness between early Quake and Unreal titles. Quake's controls always felt just that little bit tighter.
I haven't played a lot of F.E.A.R. multiplayer, and judging by the number of servers online, neither have most folks. Still, I was curious to see how the Killer NIC fared, and somewhat surprised when it seemed to have little impact on what very little lag I experienced in the game. Of all the games I played, F.E.A.R. suffered from the least amount of stuttering and lag-induced annoyances, likely in part because game servers tend to have fewer players overall. With less lag to combat, the Killer NIC's effects were muted at best. If anything, the NIC contributed a slight veneer of overall smoothness, but one that was difficult to detect.
Massively multiplayer online role-playing games are quite a bit different from first-person shooters, so I fired up Guild Wars to see whether the Killer might be able to flex its muscles in Old Ascalon. The admittedly feeble character I created for testing doesn't have access to much of the Guild Wars world, but between roaming the countryside and wandering crowded city areas, I started noticing a pattern in the Killer NIC's behavior.
In open country, with little going on around me, there wasn't much lag for the Killer NIC to correct. Here, I didn't feel any real difference between Bigfoot's network controller and those offered by Marvell and Nvidia. However, in crowded city areas with loads of characters moving about, lag was more prevalent, and the Killer NIC suffered from less hitching and stuttering. It didn't eliminate every instance of lag, but it smoothed more of the bumps than the nForce and Yukon network controllers.
Multitasking with Battlefield 2
A key component of Bigfoot's Lag and Latency Reduction tech is a packet prioritization scheme that gives game packets preference over all else. In theory, the Killer NIC should allow you to play games while transferring files or downloading via BitTorrent with no impact on actual gameplay. Nvidia does packet prioritization, too, but unfortunately it isn't yet supported in the company's Vista x64 drivers.
To put packet prioritization to the test, we first played a few rounds of Battlefield 2 on our test system with a network file transfer in progress. The Nvidia and Marvell NICs faltered heavily here, experiencing intermittent but severe lag that—hand to my heart—actually got me killed a few times. The Killer NIC reduced both the severity and frequency of those lag episodes, but gameplay was still noticeably more jerky than without the file transfer running in the background. It was easily playable, but annoying enough to be unpleasant.
Next, we moved to BitTorrent, returning to Battlefield 2 with µTorrent downloading a couple of Linux ISOs in the background. Here we experienced an incredible amount of stuttering with the Marvell and Nvidia network controllers, and pings were between four and five times higher than they were without the BitTorrent tracker running. Switching to the Killer NIC didn't improve matters much, either; there was less stuttering than with the other NICs, but still enough to render the game unplayable.
µTorrent wasn't designed with the Killer NIC in mind, but Bigfoot's BitTorrent FNApp was, so we fired it up next. This dropped in-game ping times by about half compared to what we saw with µTorrent, and we'd consider the result playable. However, lag was definitely more frequent and severe than without the BitTorrent FNApp running, suggesting that it takes more than just client-side packet prioritization to combat the effects of background BitTorrent downloads.