Original Link: http://www.anandtech.com/show/2111
BigFoot Networks Killer NIC: Killer Marketing or Killer Product?by Gary Key on October 31, 2006 2:00 AM EST
- Posted in
Here we are today reviewing one of the more controversial personal computer products to be released in the last couple of years. If you thought the AGEIA PhysX product generated controversy about whether it was a viable product or not then you should read the comments around the Internet about the Bigfoot Networks Killer NIC. The phrase Snake Oil is one of the most commonly used online terms to describe the Killer NIC. While this did not surprise us given the aggressive marketing of the product, we think it is a bit unfair if the commenter has never actually used the product.
Hopefully our review today will prove or dispel this phrase. Of course our article commentary on the product just might inflame the masses into a further war of words on the subject. First, however, we need to revisit the first line of this paragraph. We really cannot call the article you are going to read today a review; it is more like an evaluation of a very controversial yet interesting product.
Yes, we will present data such as frame rates and ping times in several of the latest games available today. Along with this empirical data will also be NIC specific results, but our test results should only be used as part of an overall evaluation of the product. When it comes right down to it, the actual experience we will convey of using this product on a daily basis should be the crux of your purchasing decision.
At this time you might be thinking we drank some Snake Oil but let us explain our comments. Our test results cannot be accurately replicated by our readers, other review sites, or even ourselves in a very controlled environment. Our test results are accurate based upon the criteria we utilized at the time of testing but they cannot be consistently replicated. In some ways our testing was a grand experiment that provided more information about our network capability and broadband provider than the Killer NIC. Wow, maybe there was something in that glass besides water now that we think about it....
In actuality, figuring out how to properly test this beast of a card was somewhat perplexing at first, reached a frustrating crescendo, and even continues to be puzzling to some degree today. The problem lies in developing a set of benchmarks that will be consistent, repeatable, and fair. Those words consistent and repeatable sound so simple when using our standard benchmark suite on a daily basis, but they still haunt us to this day when testing this card.
We centered our efforts on reducing the almost infinite set of variables in trying to test this card and its competition. Almost infinite set of variables you ask? For starters, the two most important variables to address are our network connection and the performance of the servers we connected with during testing. We tried several different methods to address (tame) this variability with each ending in a dead end, endless nights of desperation, or just creating additional complexity that did not provide any new details. In the end, there just was no way to accurately control these two critical variables that were completely out of our control. Besides our connection and the server, any testing that involves the Internet is subject to countless potential influences on performance, and with games you also throw in other players which vary in number and location.
While not perfect and open for debate, we feel like the benchmarks we utilized will provide a general indication of the card's performance while our experiences with the Killer NIC will provide the other piece of the puzzle. After all, it's not every day that a network card is launched specifically targeting the online gaming audience with promises of giving you the competitive edge you need in the heat of battle. The marketing blitz continues with statements like, "Killer frees up your computer's processor to focus on the game giving you those extra Frames Per Second and lower Pings when you need them most. Finally you can focus not just on winning - but dominating." We figure it takes a big set of brass ones to make such claims... or maybe, just maybe, the product actually works as advertised.
Let's find out if aggressive marketing or engineering genius defines the capability of this product.
Technology behind the Killer NIC
We will not be spending several pages and displaying numerous charts in an attempt to explain in absolute detail how the networking architecture and technology operates. Instead we will provide a high level technology overview in our explanations, which will hopefully provide the basic information needed to show why there are advantages in offloading the data packet processing from the CPU to a dedicated processing unit. Other technologies such as RDMA and Onloading are available but in the interest of space and keeping our readers awake we will not detail these options.
The basic technology the Killer NIC utilizes has been in the corporate server market for a few years. One of the most prevalent technologies utilized and the one our Killer NIC is based upon is the TCP/IP Offload Engine (TOE). TOE technology (okay that phrase deserves a laugh) is basically designed to offload all tasks associated with protocol processing from the main system processor and move it to the TOE network interface cards (TNIC). TOE technology also consists of software extensions to existing TCP/IP stacks within the operating system that enable the use of these dedicated hardware data planes for packet processing.
The process required to place packets of data inside TCP/IP packets can consume a significant amount CPU cycles dependent upon the size of the packet and amount of traffic. These dedicated cards have proven very effective in relieving TCP/IP packet processing from the CPU resulting in greater system performance from the server. The process allows the system's CPU to recover lost cycles so that applications that are CPU bound are now unaffected by TCP/IP processing. This technology is very beneficial in a corporate server or datacenter environment where there is a heavy volume of traffic that usually consists of large blocks of data being transferred, but does it really belong on your desktop where the actual CPU overhead is generally minimal? Before we address this question we need to take a further look at how the typical NIC operates.
The standard NIC available today usually processes TCP/IP operations in software that can create a substantial system overhead depending upon the network traffic on the host machine. Typically the areas that create increased system overhead are data copies along with protocol and interrupt processing. When a NIC receives a typical data packet, a series of interactions with the CPU begins which will handle the data and route it to the appropriate application. The CPU is first notified there is a data packet waiting and generally the processor will read the packet header and determine the contents of the data payload. It then requests the data payload and after verifying it, delivers it to the waiting application.
These data packets are buffered or queued on the host system. Depending upon the size and volume of the packets this constant fetching of information can create additional delays due to memory latencies and/or poor buffer management. The majority of standard desktop NICs also incorporate hardware checksum support and additional software enhancements to help eliminate transmit-data copies. This is advantageous when combined with packet prioritization techniques to control and enhance outbound traffic with intelligent queuing algorithms.
However, these same NICs cannot eliminate the receive-data copy routines that consume the majority of processor cycles in this process. A TNIC performs protocol processing on its dedicated processor before placing the data on the host system. TNICs will generally use zero-copy algorithms to place the packet data directly into the application buffers or memory. This routine bypasses the normal process of handshakes between the processor, NIC, memory, and application resulting in greatly reduced system overhead depending upon the packet size.
Most corporate or data center networks deal with large data payloads that typically are 8 Kbit/sec up to 64 Kbit/sec in nature (though we fully understand this can vary greatly). Our example will involve a 32 Kbit/sec application packet receipt that usually results in thirty or more interrupt-generating events between the host system and a typical NIC. Each of these multiple events are required to buffer the information, generate the data into Ethernet packets, process the incoming acknowledgements, and send the data to the waiting application. This process basically reverses itself if a reply is generated by the application and returned to the sender. This entire process can create significant protocol-processing overhead, memory latencies, and interrupt delays on the host system. We need to reiterate that our comments about "significant" system overhead are geared towards a corporate server or datacenter environment and not the typical desktop.
Depending upon the application and network traffic a TNIC can greatly reduce the network transaction load on the host system by changing the transaction process from one event per Ethernet packet to one event per application network I/O. The 32 Kbit/sec application packet process now becomes a single data-path offload process that moves all data packet processing to the TNIC. This eliminates the thirty or so interrupts along with the majority of system overhead required to process this single packet. In a data center or corporate server environment with large content delivery requirements to multiple users the savings in system overhead due to network transactions can have a significant impact. In some instances replacing a standard NIC in the server with a TNIC almost has the same effect as adding another CPU. That's an impressive savings in cost and power requirements, but once again is this technology needed on the desktop?
BigFoot Networks believes it is and we will see what they have to say about it and their technology next.
Killer NIC Technology
Hopefully you are still with us after the previous segment as writing it was better than taking a dose of Lunesta. In all seriousness, the technology of offloading network transactions to a dedicated processor has proven to be very beneficial in the corporate server environment. The typical TNIC is designed to handle data payloads that are larger than 8KB and in certain instances will have reduced performance (lower throughput and higher latencies) with smaller and more frequent data payloads in the 1KB to 4KB range. This range is what most messaging traffic, web services, and real-time data applications such as games utilize at this time. TNICs are generally optimized for TCP (transmission control protocol) packets where the vast majority of games today utilize UDP (user datagram protocol) packets for data transmission.
The differences between the two protocols are numerous but we will hit the highlights. TCP has a standard header length of twenty bytes versus eight for UDP. The normal TCP header will contain metric information such as sequence and acknowledgement numbers along with a requirement for a checksum number. UDP packets do not include metrics and the checksum information is optional. In other words, UDP does not provide the reliability, security, or ordering (queue) guarantees that TCP can deliver. The datagrams in the UDP packet may arrive out order or not at all and your system or application may never notice. Unlike TCP, UDP provides no guarantees for delivery or proper queuing, so why use it? The answer is simple: UDP is faster and far more efficient for time sensitive applications such as gaming, and you don't need every data packet to game properly. (I.e., if you miss one packet that says player X is at coordinates (10,10,15) but you get the next packet that shows X at (12,11,15), the missing packet will not seriously impact the overall experience unless of course you missed a shot or took one.) With this simple premise in mind BigFoot Networks decided to take TOE technology and design a TNIC that focused on UDP protocols and latency reduction.
The main technology focus of BigFoot Networks is centered on their LLR technology. LLR (Lag and Latency Reduction) technology that implements a 1-packet 1-interrupt model to eliminate the entire queuing and buffering operations standard NICs do during the packet receipt and transmission process. When in game mode, the Killer NIC will also completely bypass the Windows networking stack which contributes to further latency or lag reductions depending upon the application. When BigFoot Networks discusses ping improvements in games they are not talking about reducing ping through your network or at the server. This is completely out of their control and although their marketing information is not clear about it the reduction in ping comes on the host machine. These reductions come from bypassing the Windows Network Stack while in Game mode. Depending upon the application and packet size there is generally a 1~3ms delay due to system buffering and another 3~10ms delay in the queuing and processing of data packets in the current Windows Network Stack.
What makes LLR work is the NPU (network processing unit) on the card. This processor powers both the Windows Network Stack bypass engine and the 1-packet 1-interrupt model. In short, this NPU gets the normal network transactions out of the graphics path in games. This can result in improvements in FPS (frames per second) and reduced lag. With a standard network card, before nearly every graphics frame is drawn, there is first a check to the server to see if a new data packet has arrived or if one needs to be sent. Checking the server for new data packets can use up processor clock cycles whether or not data is there. Instead of a multitude of interrupts as we discussed earlier, the Killer NIC will receive or send those data packets in a single instruction. The Killer NIC has the further ability of interrupting the game directly when new data arrives. The Killer NIC is designed around reducing latencies and not throughput.
While LLR technology is impressive to some degree we have to temper any enthusiasm with the fact that most games are designed very differently in their handling of network tasks. Some games do not check for new network data on every graphics frame so any FPS improvements will be minimal at best or completely nonexistent the majority of time. Several older games do not use UDP packets so performance could suffer as the Network Stack bypass model is not used and the card must act as standard NIC. The one thing that we have learned during testing is that many games do not report accurate latency (ping rates) so any improvements are not as measurable but at times can be felt do to smoother game play. The basic TOE and TNIC technology still applies to this card and has been proven over the past few years in the corporate server environment. Converting this technology to the desktop with the added spin of improving gaming is certainly an admirable feat but how well does it work? We will answer that question in a few pages but first let's take a look at the obligatory marketing information.
Killer NIC Marketing Materials
Instead of rewriting the BigFoot Networks marketing materials we will just present it as is in abbreviated form. There are five features that are prominently featured on the Killer NIC website and the descriptions listed below are a quick recap of these items. We have already implied the marketing utilized by BigFoot Networks is very aggressive, and while understandable to those with a marketing background (no offense to the suits) we believe the website information and advertisements set the expectations of this card too high. A little more education about the background of the technology utilized and how it really works would have been better than in your face statements about gaming domination and free T-Shirt giveaways.
In fact, the description of MaxFPS technology along with the associated graphs in the reviewers guide we received does more to explain how the card works and why than the entire website at this time. Our suggestion to BigFoot Networks is to openly provide the technical information and remember that not everyone is a thirteen year old (not that there is anything wrong with them) gamer with $279 in hand and wanting to pimp out their rig because a Killer NIC advertisement with a limping sword makes them feel inferior in some way. All this is academic anyway; if the hardware does not work then who really cares about the marketing, right? So here's the marketing spiel:
MaxFPS - MaxFPS will increase the frames per second (FPS) in most gaming systems. It does this by reducing the CPU utilization due to networking, and speeding up the main game loops of the game. For gaming systems that have older graphics cards, the additional performance in CPU, cache, and main system memory will improve the efficiency of the card. This allows it to run at more FPS or at higher resolutions and settings. For gaming systems that have new graphics cards, the FPS performance is usually limited by the performance of the main gaming loop or the CPU's ability to get data to the card. MaxFPS will improve the speed of the main gaming loop and reduce the CPU utilization and main system memory thrashing thus improving FPS. Depending upon the situation MaxFPS will use its direct transfer of data to host application memory. Latency is further reduced by eliminating the time to write in and out of the system memory.
UltimatePing - UltimatePing will reduce the effective UDP-internal ping time (ping using UDP sockets). A UDP-internal ping is very different than a standard ICMP ping such as one might get when calling "ping" from the command line. UDP-internal ping runs over a UDP socket and includes such overhead as UDP checksum, buffer copy, real data transfer, and stack efficiency. UltimatePing is based on two technologies. First it ensures that neither your operating system nor your network card is introducing unnecessary latency into your game by ensuring interrupts from the NIC and OS are efficient. Secondarily, UltimatePing uses a side effect of MaxFPS to improve the response time in the main game loop or threads.
PingThrottle - PingThrottle is a user controlled setting that literally adds latency to any outbound network traffic, effectively increasing ping a gamer has to the server. The max latency you can add is 20ms and the latency is real as it is processed by the NPU. (The basic premise is that this will allow you to placate anyone complaining about LPBs (Low Ping B...s), although the real application of the technology can be somewhat less desirable depending on your viewpoint.)
GameFirst - GameFirst basically prioritizes your inbound and outbound network traffic so that your gaming packets are delivered or received first. This feature is very useful if you are running other applications that are uploading or downloading data while gaming.
FNA - FNA stands for Flexible Network Architecture. It is in effect the infinite flexibility of the Killer NIC. FNapps are designed to allow a user to run an application with a minimal or reduced impact on the main system's CPU, memory subsystem, or hard disk. FNapps can be anything from simple packet monitoring utilities like firewalls to full blown VoIP programs or file sharing programs like BitTorrent. FNapps can be designed to utilize the dedicated USB port as well as the Gigabit port. FNapps are designed with the included open-source Linux compiler and source code. The card includes a Linux console, 64MB of RAM, and an embedded Linux build.
The single most touted feature on the Killer NIC is the MaxFPS system that includes the LLR technology. Please note these comments as we will find out shortly if they work or not. The second most hyped feature is FNA. We believe this feature can really set this card apart from others and has the potential to be the most useful feature. Unfortunately, there are no FNapps available to test at this time. Considering the time that has passed since the card was introduced and the press material describing how easy it is to make an FNapp we would have expected something to test by now. FNapps are still MIA at this time, though we understand there might be some released over the course of the next few weeks.
UltimatePing actually worked in a few of our benchmarks. PingThrottle worked as advertised and by using it continuously during a heated Quake 4 match we were able to get kicked off the server and brandished cheaters for life. (It was fun while it lasted.) GameFirst worked but the outgoing packet prioritization worked about the same as our NVIDIA nForce 590 NIC and D-Link Gamer Lounge DGL-4100 router. The inbound capability is what makes the difference although we typically would not be caught downloading or uploading files while gaming online.
Killer NIC Card
The Killer NIC ships with an installation disc that includes several Linux programs, a few decals, product documentation package, and a full retail copy of F.E.A.R. on DVD.
|Click to enlarge|
The Killer NIC is a very interesting card and since the Halloween season is upon us we have to say the thought of removing the "K" heatsink and using it as part of our Blade costume came to mind several different times. The PCB is black in color and contains a significant amount of components that you would never expect to see on a typical NIC. There are also several red LEDs on the PCB whose blinking movements can be adjusted by the control panel. Overall, the card's design theme follows its naming convention.
The card is equipped with a Freescale MPC8347E communications processor, a Xilinx Spartan FPGA, an integrated Broadcom Gigabit PHY, and 64MB of RAM for the embedded Linux build that is accessed by the FNapp Console. The Freescale MPC8347E communications processor is designed for general communications requirements from Ethernet routers to Internet network appliances. The Xilinx Spartan FPGA is a programmable gate array that houses Bigfoot Networks' proprietary and patent pending algorithms that also utilize the 64MB of onboard RAM.
The card utilizes a standard RJ-45 port and includes a USB 2.0 port. The USB 2.0 port is not visible to Windows and is designed to be used by an FNapp. The thought process is that an FNapp can read or write data to a flash drive or external HD. This would be very advantageous for writing data directly to an external hard drive via a BitTorrent type FNapp as one example.
The amount of integrated components along with its basic design features is one of the reasons why this card costs $279.99. This is a price tag that generally puts this card out of reach for most users who are typically satisfied with their NIC's current performance.
The card was very easy to set up. You need to disable your current NIC for optimum performance and compatibility. This is an action that we highly advise after a couple of aborted installs with the original driver set. The card requires a single PCI 2.2 slot and is only compatible with 32-bit Windows XP at this time. After installing the card, a quick reboot, and hardware recognition by XP, you install the driver set off the CD although we highly suggest downloading the latest release first. Another reboot is required and then you are set to use the card for total world domination. Well, at least this is probably your thought process if you read through the marketing material and other documentation online.
Additional 64-bit XP and Linux support is expected within the next six to eight weeks. However, it is still uncertain as to what features will be enabled for Linux and what if any performance benefits will be gained. Windows Vista support should be available by the time the OS ships and it will be interesting to see if there are any improvements considering Microsoft has completely rewritten the network stack code for efficiency and latency reduction. At this point in time there are no plans for a PCI Express version of the Killer NIC. According BigFoot Networks the PCI interface was utilized to improve transactional latencies, programming familiarity, and the fact that most systems have the required amount of PCI slots. We disagree with that last point as the majority of performance oriented boards are now PCI slot starved, and we would expect PCI-E to be a future consideration depending on the success of the current design.
Our platform design represents a configuration that we feel is appropriate for testing the Killer NIC. It is a blend of components we felt like would be in a gaming system whose owner would potentially look at the Killer NIC and more importantly could afford the Killer NIC. Our motherboard choice was dictated by our testing requirements for comparing the best overall on-board NIC offering against what is being billed as the best NIC for gaming, period.
|Standard Test Bed
Performance Test Configuration
|Processor:||Intel Core 2 Duo E6300
(1.86GHz, 2MB Unified Cache)
|RAM:||2x1GB GeIL PC2-6400 800MHz Plus (GX22GB6400PDC)
DDR2-800 3-4-3-9 timings, 2.20V
(Micron Memory Chips)
|Hard Drive:||Seagate 320GB 7200RPM SATA2 16MB Buffer|
|System Platform Drivers:||NVIDIA 9.53|
|Video Cards:||1 x MSI X1950XTX|
|Video Drivers:||MSI/ATI Catalyst 6.10|
|CPU Cooling:||Scythe Infinity|
|Power Supply:||OCZ GameXstream 700W|
|Optical Drive:||Sony 18X AW-Q170A-B2|
|Case:||Cooler Master CM Stacker 830|
|Sound Card:||Bluegears b-Enspirer|
|Motherboards:||Asus P5N32-SLI Premium (NVIDIA nForce 590SLI)|
|Operating System:||Windows XP Professional SP2|
We are using an Intel Core 2 Duo E6300 as it offers an excellent blend of price and performance at this time. Our processor choice is representative of what a typical mid to upper range gamer would utilize in their system currently and allows us to concentrate additional funds on a high-end GPU. Our high-end GPU choice is the MSI X1950XTX that addresses our system performance needs while ensuring our standard 1280x1024 resolution choice will not be completely GPU bound in testing. A 2GB memory configuration was chosen as most enthusiasts are currently purchasing this amount of memory.
All other components in our test configuration are typical in a current gaming system. We intentionally chose the Asus P5N32-SLI Premium for testing as it utilizes the new MCP-55 found in the NVIDIA nForce 590SLI chipsets for both Intel and AMD platforms. The networking capability found in the nForce 590SLI is currently the best available for on-board solutions from both an overall performance and features viewpoint. These features include packet prioritization, teaming, and TCP/IP acceleration. Additional information about these features can be located here.
As we alluded to on the front page, designing a set of testing methodologies that would be consistent and repeatable for the Killer NIC is nearly impossible. It still is to be quite honest. However, we feel like the test criteria we devised is fair for both BigFoot Networks and our readers. There is a phrase that we had to abide by during testing and it makes sense when reviewing our results: Sometimes close enough is good enough.
How do you test a card when the two most critical variables in the online gaming experience, our network connection and the game's server performance, are totally out of our control? Besides these two variables it is impossible to guarantee the movements, actions, experience level, character class, and quantity of players on the servers we connected to would be the same or at the very least close enough for each of our tests. Our testing methodology is by no means perfect we found it to be good enough so this is what we decided to do.
We set up two identical systems for testing as our main focus is to determine if the $279.99 Killer NIC provides a better overall online gaming experience in a wide variety of games than the onboard and essentially free NIC on the Asus P5N32-SLI Premium. Each system had the exact same software image with the only differences being the driver load for the Killer NIC and the nForce 590SLI. Each system was connected to a D-Link Gamer Lounge router with all GameFuel options disabled. We have both Cable and DSL service but decided to strictly use the DSL connection to further reduce the variables. Our Cable connection speed varies greatly depending upon the time of day so this option was not viable. We have enough variables to contend with and adding yet another one was not in our best interest.
We utilized FRAPS to capture our frame rates in each game and the ping time measurements came from the games internal ping rate value where applicable. The ping rates were captured by a third party every sixty seconds and then averaged over the length of game play. We played each game for an average of fifteen minutes online for a total of ten separate sessions at different periods during the day or night. We typically logged on to servers that were fully populated when possible or were close to the maximum amount of players for the selected map.
We utilized each test system at the same time, each player played as the same character class when possible, and connected to the same game server simultaneously. Each player stayed together, followed the same path, and performed the same actions as much as possible during testing. If one player was killed during action then we started the process over until we reached our fifteen minute mark except in Counter Strike: Source and Quake 4 where the time limits ranged anywhere from seven to twenty minutes. Our World of WarCraft play time was 30 minutes in each section due to our play locations and character travel requirements.
Our testing routine was fairly methodical in nature for the FPS and MMO games. We will not mention names but one particular tester took a significant amount of time to learn how to stay out of harms way in most games. It was so bad at one point that we thought about calling Dick Cheney to give him a lesson on how to use a shotgun. Overall, we spent about 300 hours online to bring you around a 100 hours worth of results.
In our RTS testing we had to change our testing methods slightly. Each player played the same map and as the same class each time in a single versus single player game. We scripted out a series of events for each player to follow and probably had a 98% success rate in following the script. We also tried random two on two player matches on the same map and server to see if there would be any noticeable differences in our results. There were not any real differences so we will present our one on one player results. Our outside test party hosted the game utilizing the same DSL service and had a similar system setup on their end. Ping rates could not be accurately measured in our RTS games so they will not be presented, although they were typically very low at game startup due to our test routine.
We also will present results with our D-Link Gamer Lounge DGL-4100 router with GameFuel technology implemented. This router basically promises to accomplish several of the Killer NIC features on the router. This includes providing a platform that reduces latency and boosts network efficiency and performance while intelligently managing and automatically prioritizing network traffic. Sounds familiar right? We are including results from our D-Link DGE-550T PCI Gigabit network adapter to show the results from using a standard PCI NIC in your system with and without our D-Link DGL-4100 optimized.
Since we did not have four identical systems our test plan for the Gaming Router and PCI NIC consisted of powering down our test systems, inserting our D-Link PCI NIC card into each system, removing the Killer NIC and disabling the nForce 590SLI NIC in the BIOS, swapping out the hard drives for a D-Link NIC image, rebooting, and then setting up the D-Link router for optimized operations with our games. We then tested each machine in the same manner listed earlier. The difference being that we ran our tests with GameFuel on and then ran the tests again with GameFuel off. This meant there was usually a ninety minute difference in test results between our Killer NIC / nForce 590SLI combination and our D-Link Router / D-Link NIC results.
A lot can happen in ninety minutes when playing online but fortunately our variables were not off the map after each individual test session. The greatest variability we noticed in the results actually came from the time differences between running our D-Link NIC with GameFuel enabled and disabled. Please be aware of this when reviewing the numbers. However, considering our results are averaged from ten different sessions the overall variability is fairly minimal between test sessions. We set the Killer NIC to Game mode for all of our gaming benchmarks.
There were so many different combinations that we could have tried but in the end we decided to utilize this combination of components to present our results. We could have utilized a standalone PCI-E NIC, used a different video card or processor, and tested strictly on a LAN to remove most variables. We actually did all of that and the results were different but the percentage differences in scores between each solution were always about the same except when using a single processor in World of WarCraft. We will explain that issue shortly. It is time to take a look at the benchmarks and see how well this Killer NIC performs in a high-end gaming system.
As usual, gaming performance was tested with a variety of current games. We ran benchmarks at our standard 1280x1024 resolution. Given the number of users that run 19 inch LCDs these days, 1280x1024 represents one of the most commonly used resolutions at this time.
This benchmark is probably the most intense one we ran as it occurred on the Daqing Oilfield map with 64 players battling it out in a chaotic frenzy. This was also the most difficult map we had to experience from a viewpoint of keeping our players together and trying to accomplish the same tasks. If you saw two players constantly missing the vehicle rides and spending a lot of time in the trees sniping then it was us. We ran Battlefield 2 using the highest quality graphics settings available in the video settings. The game itself is best experienced with average in-game frame rates of 40 and up.
In our first game we actually find our D-Link combination performing slightly better than the other solutions from a benchmark viewpoint. During actual game play we noticed the frame rates and general smoothness of the system was best on the NVIDIA nForce 590SLI and followed closely by the Killer NIC. Several times during our testing with the D-Link DGL-4100 we would see several pauses while vehicles passed or several characters were in close proximity of each other. This behavior was worse on the D-Link DGE-550T when GameFuel was disabled on the router. In our first game we do not see $279.99 worth of game play improvement and the Killer NIC certainly did not allow us to dominate.
We played both the Asylum and Deadwood 16-player maps. We decided to report the Asylum results as this is our favorite map currently. The Deadwood map generated similar results so we are not hiding any information. F.E.A.R. is a very graphics intensive game and we switch all graphic settings to maximum (except for soft shadows, which are disabled). An average frame rate for F.E.A.R. can dip into the teens at times which is not good for a first person shooter. We still found the game playable around 35fps or above in multiplayer.
This is the money title for the Killer NIC. BigFoot Networks ships a copy of F.E.A.R. with every card so it should be expected to offer better frame rates and ping rates. We see a 7% increase in frame rates over the NVIDIA nForce 590SLI NIC and a 10% increase over the D-Link combination. In actual game play we found the Killer NIC to offer the most fluid experience and our ability to quickly transverse the map actually felt smoother than the other two solutions. The ping rates were actually higher than the D-Link router and nForce 590SLI but we did not find ping rate making a huge difference in this game. The question is did the Killer NIC improve our game play or skills? It really did not but it could for those who are significantly better players at this game than we are currently. In this case we will chalk a point up for the Killer NIC for improving frame rates.
This game should be a significant test for the Killer NIC as it requires extremely good hand to eye coordination along with lightning fast reflexes. This is just the game that would truly benefit from both improved ping and frame rates. We played on the Campgrounds Redux map with a total of 16 players. This is our favorite map from Quake III Arena and we were glad to see it return. We set our video settings to high and then fired up the game.
Actually, Quake 4 limits true frame rate to 60 fps during game play whether or not VSync is enabled. However, various timedemos and FRAPS will report the frame rate your system is capable of providing without the cap. In this case, we believe due to the speed of the game that ping rate is an extremely important performance measurement. This is one game where the Killer NIC loses on both fronts and it is mainly attributable to the excellent network code already provided in the game engine that does not benefit from offloading technology. The current performance leads us to believe we can expect to see the same results in the eagerly awaited Enemy Territory: Quake Wars.
Counter Strike Source
This is another game that BigFoot Networks touts as a premier title that will show improvements in both frame and ping rates. We are playing on the Office map with 32 players and we varied our online session times to ensure we were playing on a full server. Our settings were once again set to high quality at a 1280x1024 resolution. We find the game is very playable at 50 fps or better.
We did see frame rate improvements between three to five percent in this game. However, at these frame rates we really do not think a difference of 5% is going to matter much. Like Quake 4, an excellent ping rate in this game will make a difference and in this case we just did not see a large enough variation to make a difference in our experience. The Killer NIC did provide a positive experience to some degree in CSS as game play seemed to be a little smoother after firing your weapon than our other solutions. It was not a fire, frame hitch, and then move event we experienced a few other times during testing, especially with the D-Link PCI NIC. The Killer NIC gets a point for improving frame rates as advertised. We have to take it away for failing to improve our ping rates in a game where UltimatePing (reducing ping rates on the host system) could make a difference in actual game play.
Call of Duty 2
This is another popular shooter where frame rate and ping rate are very important especially in close quarters fighting. We played on the Caen map with a 24 player limit. The amount of packet processing this game generates is significantly higher at times than Counter Strike: Source so we expected our Killer NIC to shine in this area. Our video resolution is set to 1280x1024 with all options but AA enabled. We find the game is generally playable with a 40 fps or higher rate in multiplayer.
These results surprised us as we see minor frame rate improvements but ping rates were up to 7% greater than our D-Link router or NVIDIA NIC. The frame rate improvement is only worth mentioning because this game is generally GPU bound so any benchmark differences between systems is noticeable. Were the frame rate differences during game play noticeable? No they were not but we can tell you the increased ping rates were noticeable. This was the first game we encountered where the actual game play was not smooth with the Killer NIC. We noticed choppy break points at certain sections, especially when going building to building in the center of the map. We tried this map at different times during the day and noticed this issue several times. It seemed to be worse when throwing a grenade as we blew our selves up twice during testing. We might be slow and have weak arms in real life but our game characters did not need this handicap.
Ghost Recon: Advanced Warfighter
While not the most popular online shooter this game has picked up steam lately with the 1.3 patch. We are playing on the Strongpoint map with a total of 12 players. Our video resolution is at 1280x1024 HQ and we found this game to be very playable at frame rates above 40 fps. The servers we connected with were generally showing ping rates around 45ms to 60ms at the time of connection.
We see a very slight frame rate improvement in this game but nothing that was noticeable during game play. It was difficult to tell any differences in game play between each product and as far as we could tell all systems played the game in a similar manner. If we had to make a choice we would choose the NVIDIA nForce 590SLI NIC as it seemed a tad bit smoother in certain areas.
Company of Heroes
Company of Heroes is one of our favorite RTS games around the office. We are playing the Lorraine map in one versus one player mode. Our other test party hosted the game and we found their system configuration to be nearly identical to ours. Our resolution was set to 1280x1024 with all options on high except for AA. This game is very playable with frame rates above 35 fps in multiplayer action.
Bigfoot Networks stated to us early on not to expect any measureable performance differences in the majority of RTS games. Sure enough we did not see any improvements although the Killer NIC did score a little higher than our D-Link solutions. Once again, we did not notice any real differences in game play between any of the products. We actually tried the Killer NIC in APP mode and were greeted with basically the same level of performance.
Rise of Legends
This game was released a few months back and like most real time strategy games is very CPU intensive but still offers a very visual experience. In order to experience the game properly you need a fast CPU, very good memory subsystem, and a decent GPU to play at the higher resolutions. This is one game that seems to have the perfect recipe for the Killer NIC. We are playing the Spire Forest map with Vinci versus Vinci. We generally found the game to be enjoyable with an average frame rate above 40 fps.
Although BigFoot Networks had stated most RTS games will not show any improvements we see a slight increase in frame rates with this game. We have to temper our results with the fact that there was not any noticeable difference in game play between the various products once again. In a blind performance test, any setup could have won for the most enjoyable game play experience.
World of WarCraft
WoW is the largest and probably most enjoyable MMO we have played and is a favorite amongst many of our staff members. This is another title that BigFoot Networks has explicitly stated will show performance improvements and is one that is mentioned in most of their marketing materials. We logged in at several different times that range from Wednesday morning to late night Saturdays to ensure we experienced a wide variety and number of players on the server. We started in the Dalaran Ruins and traveled to the Alterac Valley battlegrounds doing our best to battle the same creatures with each character.
We see about a 4% increase over our other products in frame rates. WoW is basically capped at 64 fps with dual core systems so it is difficult to determine if the Killer NIC would provide any further frame rate increases with our test platform. We have seen increases quoted in the 10% or plus range in this game so we utilized a 3.4GHz P4 in our test bed and noted a 9% increase in frame rates. The Killer NIC gets a point for increasing frame rates as advertised. Of course anything over 35 fps in this game is very playable so the increases while nice did not really improve our game play.
Ping rates did improve by up to 7% (host system improvements and server variables) and this is where we did notice a difference in game play during the busy test sessions on Friday and Saturday nights. When our area was full of players we did notice smoother game play at times with the Killer NIC. This did not happen when we were basically the only characters playing in certain sections. This game is designed to a certain degree for higher ping rates so there is some balancing that is already done by the server during combat. Overall, the Killer NIC performed almost as advertised in this benchmark.
In order to make sure we were not drinking the Snake Oil we did a blind performance test in WoW with a third party. The third party had no idea their on-board NIC had been replaced with the Killer NIC. The first reaction was a question asking if an additional one gigabyte of RAM had been added to the system or if the video card had been changed. This would normally make for very good publicity and we found it slightly amazing that someone would pick up on the differences in performance so quickly. However, this configuration was a dual core CPU based mid-range system and the user would not be one to spend $279.99 on a peripheral other than a new graphics card. If someone had offered them this card for $50 in order to improve their WoW experience they probably would have jumped at it instead of gasping when told the actual price.
Ethernet NIC Performance
Our current motherboard test suite includes LAN performance measurements and we will utilize this same test today. All of the boards listed utilize PCI or PCI Express based controllers with the only difference being the supplier of the core logic.
The Windows 2000 Driver Development Kit (DDK) includes a useful LAN testing utility called NTttcp. We used the NTttcp tool to test Ethernet throughput and the CPU utilization of the various Ethernet Controllers used on the Intel motherboards.
We set up one machine as the server; in this test, an Intel system with an Intel CSA Gigabit LAN connection. Intel CSA has a reputation for providing fast throughput and is a logical choice for our Gigabit LAN server.
On the server side, we used the following Command Line as suggested by the VIA whitepaper on LAN testing:
Ntttcpr -m 4,0,‹server IP› -a 4 -l 256000 -n 30000On the client side (the motherboard under test), we used the following Command Line:
Ntttcps -m 4,0,‹client IP› -a 4 -l 256000 -n 30000At the conclusion of the test, we captured the throughput and CPU utilization figures from the client screen.
No real surprises here as the Killer NIC has a Marvell PHY and utilizes the PCI interface for operation. The throughput rates and CPU utilization of the Killer NIC was very competitive with our other PCI based solutions. However, the NVIDIA nForce 590SLI had the best overall performance numbers in these tests. BigFoot Networks was adamant about the fact they concentrated on lower latency and not improving throughput on the card since most network connections while gaming will not saturate the PCI bus.
Ethernet Outbound Ping Performance
Our final benchmark takes a look at the ping performance of each of our products while playing F.E.A.R. and uploading a 400 MB file to a separate client on another network. Our D-Link DGL-4100 router and the NVIDIA nForce 590SLI include packet prioritization for outbound traffic while gaming. We decided to see how well these solutions stacked up with the Killer NIC that can also prioritize inbound traffic.
All three of our products that advertise packet prioritization are within 3% of each other in the numbers with our D-Link DGE-550T trailing by around 33%. Why a serious gamer would actually want to upload large files while playing online is beyond us but in case you want to then our three solutions provide the capability to do so without an issue. However, be aware that throughput suffered by an average of 44% for the upload file test with our three solutions when compared to the D-Link NIC during testing.
Our primary purpose in this review - or better yet product evaluation - was to prove whether or not this product works as advertised. We have to admit that it does to a certain degree as shown in our benchmarks. However, benchmarks do not tell the entire story with this card. We will try our best to explain our experiences with this card and provide some additional insight into our results now.
Our experience with the Killer NIC was very frustrating at times. We exprienced everything from driver incompatibilities to botched driver installs. We explained it to technical support best last week by stating that, "It was time to toss the card out the window and get on with life." A scornful statement but one that perfectly described our feelings at the time after the uninstall program basically rendered our drive image useless. We were not the first ones to experience these issues and unfortunately we might not be the last ones either in our opinion.
Another area of frustration was the fact that we could not use FNapps. This is the number two feature on the card and from all indications accounts for a large portion of the card's cost. This feature has been hyped ad nauseam since the inception of the card and we still do not have one working applet to test. In theory, this feature sets the card apart and could become a viable reason to own the card depending upon your needs.
We do not agree with the aggressive marketing techniques utilized by BigFoot Networks. At least it seems overly aggressive to us and we believe the tone set by BigFoot Networks has led directly to some of the backlash on the Internet. While the claims about performance improvements on the website seem a bit excessive based upon our results we can fully understand a company using them to promote the product. What we found to be an issue was the lack of easy access to the card's technical information that we found in the reviewer's guide. This type of information should be easily accessed on the website. It explains the technical reasons as to why offloading the network stack onto a dedicated processor can improve both frame and ping rates no matter how minimal the increases might be in most games.
We also discussed the viral marketing instances that we have seen along with some of the outlandish performance claims on Newegg with BigFoot Networks. Their reply was that neither they nor their marketing group were involved in these activities. They also stated there might have been some enthusiastic owners or people looking for information about the card on the various forums along with hecklers adding outlandish performance claims about the card.
We did see several emails from BigFoot Networks asking Newegg to delete some of the more outlandish review claims about the card. Newegg complied with this request in most instances. While we truly believe most of what BigFoot Networks explained to us we have serious doubts about the forum posts. There were too many junior or new members in forums across the Internet posting about the card at time of launch to have been a coincidental series of events. We call it viral marketing, others call it enthusiastic owners and fans of a product. Whatever it truly is, it makes us want to wear the wading boots and take a hot shower after searching through the forums.
(Editor's Update - We spoke at length with BigFoot Networks personnel and those of their advertising agency about the forum posts we noticed at product launch. We believe they are now very sincere about not promoting the Killer NIC through viral marketing and they reassured us this is not their intent or practice. In fact, we have seen these types of posts start to disappear and now instead see BigFoot Networks technical personnel on several forums assisting users. We commend BigFoot Networks for this type of customer service. We also understand the price of the card has dropped to $249.99 and can be found with rebates at this time. Their first FNapp has been released and we look forward to testing the card again in the near future.)
Our comments above have probably led you to believe we truly dislike the product or the company. Actually, this is far from the case. As an example, during testing we came across several driver and application compatibility issues. In every case, the personnel at BigFoot Networks worked to solve our issues and those of users on their forums. Each driver release solved the issues we or others noticed. We would receive driver releases over the weekend and late at night. The technical support provided to us was superb the majority of the time.
We fully realize this is a startup company and a new product introduction so there will be growing pains but so far in our experiences the technical support functions have been top notch. The personnel in the company have been very open and always willing to communicate with us. We also believe they have very good hardware engineering expertise based upon their product design and implementation. The drivers are starting to mature rapidly and to be honest the product should be launching at this time based upon the performance of the card and its newfound stability. We just wish the owner's manual was up to par. The multitude of options in the control panel that are available for configuration are briefly described from a technical viewpoint but the descriptions do not explain how or why changing these options would improve system performance.
When we discuss performance, it is not only about how well the card performs in games but how well it performed as a standard NIC. Up until the 126.96.36.199 and 188.8.131.52 driver releases we noticed several issues with extended system startups, sluggish Windows Explorer performance, long load times for Explorer, Firefox, or Opera pages, and download speeds suffering when compared to our onboard NICs. These issues along with the capability to use Skype while in game mode have been addressed in the latest 184.108.40.206 driver release. While Skype operation is not perfect yet it now works properly in game mode. You are required to start Skype in APP mode and then switch to game mode. We fully expect this step to be removed shortly. Our point is that when an issue arises or a feature is requested we find BigFoot Networks addressing it almost immediately.
In regards to game performance the card actually worked in improving frame rates in the top three titles that BigFoot Networks heavily promotes as show casing the benefits of the Killer NIC and its MaxFPS technology, those titles being F.E.A.R., Counter Strike: Source, and World of WarCraft. We witnessed improvements from 4% to 10% in these titles when compared to our standard D-Link PCI NIC. This sounds impressive and was a slight surprise to us actually. However, we did not see or feel any real differences in our game play experiences due to increased frame rates from the Killer NIC or by simply overclocking our video card.
Ping rates were a different story. We displayed ping rate results but it is nearly impossible to truly measure or benchmark these numbers. There can be a small reduction in ping rates on the host system due to the offloading and bypassing of the Windows Network Stack but these differences rarely show up with the one exception being World of WarCraft. We firmly believe the majority of our benchmark differences are due to the variability in our network connection and at the server.
In fairness to BigFoot Networks, we did notice our game play appeared to be smoother in F.E.A.R., World of WarCraft, and Counter Strike: Source when the server traffic was at its heaviest point. The likely cause for this is improved ping rates although any differences were not measurable during our test runs. We did not notice this same improvement in Battlefield 2, Quake 4, or Call of Duty 2 so it is obvious there are certain game engines the card works well with and others it does not. In our limited single CPU testing we saw further improved frame rates over our dual-core platform in the three optimized games but no significant differences in our other game titles.
This is the real irony of the Killer NIC as the systems that show the greatest amount of improvement (in a very limited number of titles) belong to owners that would never consider spending $279.99 on a NIC. Those who can afford the card are probably running system specifications in which the game performance improvements would never be noticed. In fact, we could simply overclock our systems by 5% or a little more and end up with the same frame rate improvements. That leaves a very small audience of buyers who would potentially purchase the card for the gee-whiz factor or the professional gamer who has the ability to take advantage of a 1ms or better improvement in ping rates in Counter Strike: Source or could tell the difference between 58 fps or 53 fps in F.E.A.R..
We truly wish BigFoot Networks success as their technology has merit on the desktop. We see data payload requirments increasing in upcoming game titles and their offloading technology could have a larger impact on improving your online game play experience. Our current opinion is, without FNapps, improved performance across a wider variety of titles, and a significantly lower price tag, this card is destined to be nothing more than an interesting footnote in the annals of hardware history.