4. Online Gaming Results

We used two online environments to test the impact of the M1 Killer NIC on Internet access and responsiveness:

• Second Life: This MMORPG provides a demo that uses real-time game communications to expose visitors to how this virtual world looks, acts and feels. That also makes the demo a great tool for assessing Internet access, because it let us run both test rigs simultaneously and side-by-side in hands-off mode, where we can safely assume that both machines are dealing with the same client-to-server conditions, especially at the local network level, all the way through our broadband provider (Time Warner) to their Internet link. Of course, Second Life isn’t as graphically demanding as a good FPS, and explains our second choice for testing.
• Crysis: Because we could pick up two copies of this game, widely acknowledged to be among the most graphically demanding of today’s FPS offerings, at relatively low expense, we bought and installed them on our test machines. But that’s when things got interesting: because the Crysis demo (all 1.77 GB of it) plays itself locally, we had to play the game on both test machines at the same time in real time, so we could obtain actual frame rates and ping times using a live Internet connection. This called for two testers to work together in synch, literally right next to each other, to keep both active in the same areas of the game at the same time. This approach was effective but also very time-consuming, and explains why we drew the line at testing here.

We also looked at demos from World of Warcraft (WoW), Battlefield 2 and 2142 (Internet mode), Lord of the Rings Online and EverQuest 2, but all suffered from the same understandable but fatal defect from our point of view: all use streaming communications (not real-time request/reply communications) during their demos, and thus like Crysis, demand real-time game play to establish real performance metrics. That’s why you see our results include only the numbers we gleaned from Second Life and Crysis. If it helps further to understand why this field was so narrow, it took us over a day and half just to put results for these two environments together. All this said, here are the results.

Frame Rates

We used the same 24" monitor for all tests on both machines, with AA and AF turned off at a resolution of 1600x1200, and all detail settings turned up to their legal maximum values for both Crysis and Second Life. We then used FRAPS to capture frame rates for similar, easily repeated scenes in Crysis, and ran FRAPS against a set of identical scenes from the Second Life demo. The results of our testing are reported in Table 4, based on 30 runs for each game; there, values reported represent total frames per second, so higher values are better than lower ones.

FPS Reading for Crysis and Second Life

What we observe for the more demanding Crysis is about a 10% improvement in frame rates for the Killer NIC versus the Realtek GbE interface; for Second Life (abbreviated as 2L in the table), the difference is more like 5.6% (and collecting data is much easier in the latter, because nobody else is usually trying to kill you). The 10% difference in frame rates is pretty widely reported on forums and other coverage of the Killer NIC, so we’re comfortable in asserting that this is a typical difference that most gamers will experience for FPS games after installing this hardware. It’s a modest but tangible improvement that translates more into an improvement in fluidity and an ability to anticipate other players, than into a radical speed-up.

PING Times

Apocryphally, PING is sometimes decoded as an abbreviation for "Packet Internet Groper," but what it really does is to send a minimum-length IP/ICMP Echo request packet to a target system, which then responds with an ICMP Echo reply packet. Thus, PING provides a rough-and-ready measure of the minimum "round trip time" for a packet’s travel time from sender to receiver and then back again from receiver to sender. While most real game (and other) packets are longer than 64 bytes (the minimum legal size for an IP packet), 64-byte PING packets nevertheless provide a decent comparative measure for how fast or slow things are moving across the network between the sender and the target machine at any given moment. And of course, wide variation in PING times depending on time of day, broadband congestion and “Internet traffic conditions” also explain why we had to run two test machines at the same time to safely compare results between them.

Our testing concentrated on measuring PINGs sent at about the same moment (subject to human reaction time to "3 2 1 go!" announcements) from both test machines. Thus, the difference in PING times between the Killer NIC machine and the Realtek GbE interface machine reflects the difference in round trip time when using or not using a Killer NIC. Our results appear in Table 5, and are interesting. We used the Bigfoot Networks LagMeter (available through a third party source) to collect our PING data, and again accumulated 30 sets of data points to ensure reasonably representative values.

Table 5: PING Results for Crysis and Second Life (Min/Avg/Max msec)

What’s interesting about these results is a consistent speed-up in PING times between the Realtek GbE interface and the Killer NIC in a range from 8 (min) to 18 (avg) to 28 (max)milliseconds for Crysis, and from 5 (min) to 20 (avg) to 35 (max) for Second Life. This puts the average difference at about 18-20 milliseconds— which represents a substantial 23% to 25% improvement for a system with the M1 Killer NIC versus a built-in motherboard GbE interface. Though our other tests indicate this doesn’t translate into an equal percentage boost for frame rates (and reflects the size of the payloads in real gaming traffic as opposed to minimum packet size for PING traffic), it’s still a remarkable difference.

CPU Utilization

As we were running our game tests on both machines, we also kept an eye on CPU utilizations as reported in Windows Task Manager. In both of our test programs, we observed slight but measurable differences in CPU utilization between the machine with the M1 Killer NIC installed and the machine using the Realtek GbE interface on the Gigabyte motherboard. For Second Life, those differences were fairly small: they typically fell in a range of 1% to 4% (so that if, for example, the M1 machine reported 18% utilization we’d see somewhere between 19% and 22% on the other machine at the same time). For Crysis, those differences fell across a somewhat larger range from 1% to 6%, reflecting what we have to believe is more demanding network activity from that program.

Thus, there is some modest but still demonstrable proof for Bigfoot Networks’ claim that the Killer NIC can lighten the processing load on the PC’s CPU. Although we didn’t have the opportunity to test the hypothesis that a Killer NIC might benefit users of PCs with less powerful CPUs more than on our quad-core QX9650, we can’t help but speculate that in situations in which the CPU is more stressed under the processing load during Internet game play, the benefit of offloading is likely to be larger and more noticeable. In fact, this is a well-known effect on network servers with TOE cards, where differences in CPU utilization as large as 50% have been reported under heavy network loads with TOE cards in use (see Dell’s Boosting Data Transfer with TOE Technology or HP’s Using TCP/IP Offload Engine for some compelling evidence).

With a smaller number of TCP and UDP ports open at any given moment on a gaming PC and lower overall network traffic levels as well, the benefits of offloading are bound to be more limited on such a machine. But there’s no arguing that when handling its share of gaming and graphics processing, even a 3.0-GHz quad-core Intel processor benefits from how the Killer NIC offloads Layers 1 through 3 of the TCP/IP stack along with TCP checksums. We have to believe that this benefit will increase as the number of available processing cores and their processing speeds decrease. It might be interesting for owners of Killer NICs to post to our forums and share results with and without the card in place (give us your Second Life demo frame rates, PING times and CPU utilizations, as well as share frame rates from Crysis and other games of your choosing, please).