3. Test Machine Setup and Configuration

We selected a fairly current motherboard and a recent Intel quad-core CPU as the foundation for our system and, because the Killer NIC targets gamers, we also selected an 8800 GT graphics card as well. Because we built these systems purely for testing, we used a $79 bench rack for the two builds: the HSPC Tech Station computer workbench.

A Tale of Two Systems

We built two systems to test side by side: both identical, except that one includes a Killer NIC and the other uses a built-in GbE Ethernet interface on the motherboard. This approach is critical to our testing for reasons that may not be intuitively or immediately obvious, but all relate to how networks behave and change over time. Simply put, we attached two machines to the same Internet access chain (LAN, broadband router, ISP connection and Internet link) at the same time so that when both machines were run at the same time they would encounter the same Internet access conditions (latency, throughput, upload/download speeds, broadband congestion and so forth).

Indeed we could have used a single system, and conducted our whole sequence of tests twice, once with the Killer NIC installed and the other time using only the motherboard’s built-in GbE interface. But because conditions on the Internet vary so widely and wildly over time, we really didn’t think that would give us the ability to compare results between the two machines. Only by running both at the same time, in the same environment, can we factor out all of the variables related to Internet conditions at any given moment. We can also safely assume that the differences between the two sets of values may be attributed to the single hardware difference between the two machines—namely, the presence or absence of the Killer NIC itself.

Base Configuration

The other configuration details for the test systems appear in Table 1, and several photos follow thereafter.

Table 1: Quad-core systems

The Difference Lies Within the Network Interface

One again, the only difference between our two test systems is the presence or absence of the Bigfoot Networks M1 Killer NIC card. Our photos show the unit with the card installed; the other unit looks exactly the same, except that there’s no PCI Ethernet card installed, and we plugged our network cable into the left-hand RJ-45 connector instead. According to the Gigabyte specifications, this is a Realtek 8111B networking chipset, which supports 10/100/1000 Mb speeds. It also shows up on our GbE switch as a gigabit interface (a NetGear FVS1024G VPN switch/router), so it’s seen on the network as a gigabit device as well (of course, the same is true for the machine with the M1 Killer NIC installed).

The M1 Killer NIC, by contrast, uses the Broadcom BCM5461A1KPFG networking chipset, which offers roughly the same feature set. Based on that parity, which is summarized in both vendors’ claims to fully support 802.3, 802.3u and 802.3ab; we can’t but believe that Bigfoot Networks put significant engineering resources into writing the Windows XP and Vista drivers for the Killer NIC cards, and in optimizing the TCP/IP implementation that runs on those cards’ PowerPC processors.

From the front, the HSPC bench case presents the motherboard from the rear, and shows the business ends of both the 8800 GT and the M1 Killer NIC, as well as the port block at the left.

From the rear, the 24-pin ATX connector and Corsair Dominator RAM are visible on the top deck, and you get a good view of the ODIN PSU sticker and the SSD driver on the bottom deck.

Our goal in creating our two test rigs was to put together a system in which it would make sense for a serious gamer to include a $200-plus network interface card like the M1 Killer NIC. At a ballpark cost of around $2,800, not including the Killer NIC itself, we think our test system surely qualifies. And simply dropping the 32 GB SSD in favor of a pair of fast, capable Seagate Barracuda 7200.10 500 GB drives not only ups storage capacity considerably, it also drops the price by $650!