The reason why the KillerNIC can help to increase the frame rates during gaming and reduce lag and latency is because the card can do all of the network processing on its own. This is made possible through the use of a 400 MHz NPU (CPU) and 64MB DDR RAM. By decreasing you CPU's workload, your CPU can focus on processing game related tasks like geometry and environment processing. In games where the CPU is the bottleneck, this will give you additional FPS.
The KillerNIC can do all of the UDP (User Datagram Protocol) protocol handling calculations and processing on board, but when it comes to TCP (Transmission Control Protocol) traffic the KillerNIC will function like any other network card and it won't be able to reduce CPU load. This isn't a real problem because most games utilize the UDP protocol for sending out their multiplayer traffic, because UDP has less overhead. For example the protocol itself does not perform any consistency checks. Also in general, UDP packets get from point A to point B faster than compared to TCP. On the other hand the UDP protocol is not as reliable as TCP because UDP packets can arrive out of order or go missing without notice. UDP is mainly used by games, Voice over IP, IPTV, and TFTP applications. Basically every application where you need low latency and high bandwidth regardless of the reliability impact.
Now even though network processing doesn't take up a lot of CPU time it can make online gaming performance vary a bit because it interrupts the CPU every time a certain amount of data is in its buffer. One of the real issues with most network cards is that the network architecture is optimized for throughput and not latency.
Most network architectures today only process network packets when there are a certain number of packets in cue to be handled. In a bandwidth oriented environment like this even time critical packets, like game or VOIP packets can be held up by other network traffic because the card accumulates a certain amount of data before it's given CPU processing time. With the KillerNIC, network processing is done on-the-fly which means that your private network latency is reduced even without using any resources outside the KillerNIC board. The network architecture of the KillerNIC is in other words focused on reducing latency and CPU load.
In certain games the performance improvements can be massive if the game has an advanced net-engine that utilizes a decent amount of bandwidth. One of the most advanced net-engines is used by Counter-Strike:Source. The net traffic while playing a game like Counter-Strike:Source mostly consist of player positions and other game related information such as weapon used and stuff like that. This information is then interpreted by both the game client and server to provide an up-to-date list of events in game.
Besides being able to process the packets on-the-fly the KillerNIC NPU can also sort the packets so that game related info get top priority over other traffic. This feature has the potential to reduce hardware "lag" because game packets are automatically processed first by the KillerNIC even under massive network load.
The KillerNIC is a smart way to hardware accelerate networking and gain a smoother game play because it unlike other network cards has its own processor and RAM.
Having a 400 MHz network processor and 64 MB DDR RAM might be a bit overkill, but because the card runs a small Linux kernel you can actually code programs that can be run directly on the KillerNIC card without affecting you PC's performance. These small applications are called FNA applications. FNA stands for Flexible Network Architecture. These FNA applications can access all of the resources on the board including the USB 2.0 high speed port on the back of the card. This means that you basically have a small PC inside your PC if you hook up an external hard drive.