I’ve talked in a number of places over the past year about the “Killer NIC”, a product that was announced about a year ago and promised the world – at a price. Simply put, this is a network card for your PC, except it costs $250, which is probably 50 to 100 times more than a network card has any business costing. I’ve spent the last year slating this product whenever I heard about it, but I’ve now got some performance stats courtesy of the US PC Gamer, so lets talk.
The Killer NIC is a card for gamers it claims. It claims that it will reduce lag and latency in your games, and speed them up by moving the entire network stack out of your CPU and onto the network card. Normally this isn’t the case, a network card is more or less just an interface that passes messages between the network and your CPU, it doesn’t actually take the message apart or read it. In an era where graphics cards do graphics computation and physics cards handle physics computation, why shouldn’t network cards move network-related computation and put that onto its own processor architecture? There isn’t a reason, and this is where the Killer NIC seems to shine – so without further ado, lets get into the numbers.
PC Gamer used a “high-end” system, and did comparative testing across three games – Counter-Strike, Battlefield 2 and World of Warcraft. In CS, they found a 2ms drop in ping and an increase of 21fps. In BF2, a 1ms drop in ping, and a 13 fps increase. Finally, in WoW, they saw a 38ms drop in ping and a 25fps increase.
So now lets talk about this. Firstly, as far as I know, the only games where ping really matters are the “twitch” games, first person shooters where you absolutely need to know where the enemy is to be able to shoot him. But these are the games where the Killer NIC has least effect. In a game where ping is much less critical, WoW, it had a larger effect, but rather than say that this is prove that it works, if you actually think about it, what this could well be saying is that the Killer NIC can optimise a game’s netcode a bit, and for games where the developers have not placed an emphasis on squeezing the last few milliseconds from the ping value, that might have a serious impact in numerical terms, but in gaming terms? I’m not so sure that it makes any difference at all, particularly in a game like WoW where ping largely affects… well nothing really. OK, when your ping hits insane levels, WoW becomes a nightmare of stilted avatars skipping round the screen, and interacting with anything becomes a ponderous process of waiting for windows to open. There are times in high-end raids where maybe having a low ping is advantageous, but really there is little worth in dropping to a very low ping if you already have a low one. And the actual returns on a ping decreasing by 1 millisecond in a twitch game? They have to be negligible. Look at it this way, human reaction time for an average person is, at best, around 200 milliseconds. For a highly trained athlete, you could be talking as low as 150 milliseconds. So is a decrease of 1ms in the delay between the data being sent and received (which is essentially what ping measures) actually going to have that big an effect? Probably not. And anyone who claims it does really ought to apply for a job as a fighter pilot.
So having debunked the worth of the gains received in network performance, its time to move on to the increase in frames per second. Its pretty obvious that if you do all the message processing stuff outside of the CPU, then that means you have more free CPU to do number crunching in the game, so you will see a framerate increase. There is absolutely no arguing with that, and the numbers bear that up. To put the numbers in context, the increase is somewhere between a quarter and a third of the original value across the board, which shows a pretty scary amount of your CPU is being munched by the networking tasks. The problem I have here is again, it should make a negligible difference, and again it comes down to biology. The human eye perceives images at around 32 fps. Movies run at around 24 fps, and appear to run smoothly to us. So by that, an increase of 59 fps to 72 fps (BF2) should not be perceptible, and the same goes for all the other games tested. Now I know a lot of people out there will moan, and say that they can tell between 50 fps and 100 fps. There are too many of you to ignore, but from a pure numbers point of view, its not possible as far as I know. So yes, it gives a nice boost to your system’s performance, but is it actually going to make your gaming experience better? Probably not.
The Killer NIC also has some other features that are worth mentioning, if only to ridicule. Firstly, it does traffic shaping to prioritise gaming traffic. That’s a nice feature isn’t it. Because it means that your gaming packets get processed first all the way to the server right? Not exactly. What this is going to do is send out your gaming packets first from your machine, and intersperse other traffic dependent on the shaping system in use. But networks aren’t point to point connections, each packet has to go through a series of hops across different networks to get where it is going, and your shaping has no effect on how that packet is processed at every hop after that. So lets say, in the best case scenario, you are connected by a single wire to another machine, and you are downloading a file from that machine and playing a game with the user of that machine. Well sure, in that case, your gaming traffic is sent out first over and above requests for parts of the file you want. That’s great. But put it in a more common context, lets say you and your flatmate/family member/whoever have a machine at home each and share an internet connection. You are playing a game on the internet whilst the other person downloads a file. At your machine, your gaming packets are sent out first, but at the router for your house, they are being received and processed at the same time as the other person’s file packets, and the order that the Killer NIC is imposing on prioritisation isn’t being enforced in any way other than the way your packets arrive. That is going to be true at each hop the packets take, so realistically this kind of traffic shaping isn’t going to do any good at all being implemented at just the client level. Now if you could buy Killer Routers, and get a Killer ISP, maybe we’d be have a ball game, but as it stands, its just a daft idea.
Finally, lets wrap up by mentioning again the price of this device. Its $250. That’s about £125. That’s a lot of money for not a lot of perceptible returns. Depending on your system (and the PC Gamer guys didn’t publish their spec, so I can’t comment on that), there are possibly a number of more useful places that could be spent – more RAM, a better processor, a down-payment on a new graphics card. Don’t forget that the Killer NIC only has an onboard 400Mhz processor, and whilst that might have been optimised for the kind of computation being undertaken, for £125 you can probably drop an equivalent upgrade onto your system that will be useful for offline gaming too. That’s not to say I think that the concept of dedicated network controllers is a bad one, I don’t, and I think its absolutely the direction everyone will head eventually – don’t forget the scorn poured onto the first of the dedicated graphics cards. But at this price point? With these features? The Killer NIC is absolutely a product for people who don’t know better, are too cool for school and have way too much money.