Introduction

I knew a man once that always liked to say, "dreams are free." What he basically meant is that it's okay to talk about things that you will probably never have; that you can still have hopes of achieving great things in life regardless of your current situation. That's a good introduction to this article, because the simple fact of the matter is that few of us are likely to purchase a Falcon Northwest computer, just like there are few people that will buy (or even sit in) a Lamborghini during their lifetime.

Falcon Northwest has been in the business for almost 15 years (since 1992), always catering to the high-end enthusiast. In fact, high-end enthusiasts are a step below what Falcon offers, as many of us might actually purchase a high-end system. What Falcon Northwest offers is better classified as exotic systems, and if you want to go there they can create truly unique looking designs for you. Custom paint jobs that can cost over $1000 are available - you can even send in the pictures that you want painted on the case!

The cheapest systems start at around $1500 for just the system (i.e. no keyboard, mouse, monitor, or speakers), and prices can reach over $10,000 if you want to go all out. We really do mean "all out" for such a system: Quad-SLI, 30" LCD, 2x400GB RAID 0 HDDs, custom paint - the works. Do you need something like that? Almost certainly not. Would you lust after such a system, though? Some of you might not, but most of us geeks dream of having that much computing power at our fingertips.


Kelt Reeves, president of Falcon Northwest, likes to describe his company as the "anti-Dell". Dell sells millions of computers each year at pretty reasonable prices, all offering decent performance for the home and business user but usually nothing truly stand out. Dell systems are basically cookie-cutter designs: if you and hundreds of other people all go out and buy the latest Dimension system for $1000 from Dell, they'll all be close to identical - only their internal components might differ. Falcon Northwest on the other hand doesn't look to sell millions or even thousands of computers per month. Their goal is to offer a fully customized, ultra high performance computer that qualifies as something of a status symbol. Their target market is fundamentally limited, as they're looking for people that want the best money can buy, with cost being a distant concern.

Needless to say, it's pretty exciting to open up the packages from Falcon Northwest, and the all-black boxes with silver logos already start to give you the impression that you're opening something out of the ordinary. It's not every day you get to test a system that costs over $5,000, even if you write for AnandTech, and while most of us are more than capable of building a super high performance system on our own, we're still interested to see how one of the top boutique system builders does when it comes to putting together a dream computer. $5,000!? Yes, this is a very expensive system, and we certainly don't recommend that most people go out and spend that much money on a computer. You could build two great systems on your own for that much money, or you could build five moderate systems for the same price. The thing is, there are plenty of people with money that don't have the time to build or inclination to learn how to put together a hot rod computer. That, in a nutshell, is the target market of Falcon Northwest.

The unit that they sent for review is their latest FragBox SLI offering, and it's similar to the Monarch Hornet Pro that we reviewed a couple months ago. In fact, the two systems share the same motherboard. The primary difference is that Falcon Northwest has upgraded several parts, due mostly to the recently launched NVIDIA 7900 GTX. The goal of the system is to offer all of the performance that you can get in a regular desktop system, only put it in a smaller case. Ideally, if you were to call up Falcon Northwest to place an order, the only major difference between the FragBox SLI that we are reviewing and their Mach V offerings is going to be the size of the case and motherboard - which also means the expansion options. We'll run some benchmarks against a desktop system (not from FNW) to see whether or not they succeed in that goal.



Appearance

Opinions on what looks good always vary: clothing, hair, cars, computers - you name it, and some people will like it while others will probably hate it. There is also the reasonably large group of individuals that couldn't care less about appearance. Needless to say, if looks aren't important to you when it comes to computer systems, Falcon Northwest is definitely going to seem like overkill.

We were sent a "standard" unit, meaning it doesn't have Falcon's Exotix paint job. However, even without the custom paint job this case is a real looker. It has cold cathode lights in the interior and on the front panel, a sleek black anodized paint job, case windows on both sides, and even a carrying handle for hauling the system off to LAN parties. Any enthusiast that so much as glances at the case will recognize the brand and design, and regardless of what they might say, you can be sure they would love to have one if money is removed from the equation.

Click to enlarge

A picture is worth a thousand words, so we did our best to snap shots from a variety of angles. Online images don't really convey the experience as well as actually touching and handling a unit, but trust me: the design oozes quality. A switch on the back of the case can turn the lights on and off, which is always a nice touch if you have the system in a bedroom or the living room where the blue lighting might not be desired. (The dim light on the HSF doesn't shut off, however.) Most of our pictures were taken with the case lights off and in a brightly lit room, but we did want to include one good shot of the lit-up system in a dark room.

Click to enlarge

From the purely aesthetical standpoint, it's difficult to find any flaws. You might leave some fingerprints on the case windows, and you might want to dust the case (inside and outside) on a regular basis, but that's typical of any similarly designed system. The front panel has an awesome etched look, again not something that can easily be conveyed via images. Quite a few people have seen the system sitting on my desk during the past few weeks, and the responses ranged from "Wow!" to "Holy [expletives] that looks awesome!" Not a single person has had anything but positive remarks for the FragBox case.

Click to enlarge

How does the case compare with competitor offerings? The Monarch Hornet Pro goes for a different style, and I'd have to say the FragBox looks nicer. The acrylic face plate and curved carrying handle on the FragBox definitely set it apart from typical computer cases, and the Hornet Pro looks rather boxy in comparison. Moving to the Shuttle SN26P, that still has the "small and sleek" factor going for it, which some might prefer. It doesn't have the pizzazz of cold cathode lights, windows, logos, etc. - though of course you could add them - but that's okay. Just talking about looks alone, the FragBox gets my pick, followed by the SN26P and the Hornet Pro - but opinions differ, and what's really important is what you like.

Click to enlarge

We mentioned the custom paint jobs that Falcon can do, and we've included a sample above. If that's something that piques your interest, we recommend you check out the gallery over at the Falcon Northwest site. As you can see, they'll customize your case, keyboard, mouse, monitor, etc. - provided you're willing to pay, naturally. If you're looking for a fast PC that can also be used as marketing for your product, the customized paint jobs are certainly an interesting way to advertise. They're also a great way to truly make your PC unique. Excessive? You betcha', but that's Falcon's modus operandi in a nutshell!



Features

We've talked about the appearance of the case, and we've talked about prices. There's a reason for the prices, of course, and that's the features, components, and performance that you get with the system. It's no secret that you get greatly diminishing returns on many hardware upgrades. A $1000 system is rarely twice as fast as a $500 system - at least, if you discount the difference in performance between IGP and a discrete graphics card - and $2000 systems are even less of a performance jump from $1000 builds. Doubling or tripling cost to get into the $5000+ range will get you about the fastest system currently available, but many of the components will carry a 10-20% performance increase for twice the cost. So here's a look at the components Falcon includes in one of their top FragBox SLI systems.

Falcon Northwest FragBox SLI
Motherboard:	EVGA 131-K8-NF44-AX NF4 SLI (NVIDIA nForce4 SLI chipset) Micro-ATX
Processor:	Socket 939 CPU up through FX-60
Heatsink/Cooling:	Zalman CNPS7000Cu (most 939 HSFs will fit) 1 x 40mm rear fan; 1 x 80mm HDD fan; 1 x 120mm PSU fan
RAM:	Four DIMM slots supporting up to 4GB DDR200 through DDR400
Expansion Slots:	2 x PCIe X16 with X8 data connections (including SLI bridge connector) 1 x PCIe X16 with either X1 or X16 (configured by jumpers) 1 x PCI
Expansion Bays:	2 x 3.5 inch internal bays 1 x 3.5 inch external 1 x 5.25 inch external
Audio:	RealTek AC'97 7.1 (ALC850)
Power Suply:	SilverStone 600W Modular PSU 1 x 24-pin; 1 x ATX12V; 4 x SATA; 5 x 4-pin Molex; 1 x mini Molex (floppy) 2 x PCIe 6-pin Additional cables can be added if necessary
Motherboard Connections:	1 X FDD 1 X IDE 4 X SATA 3.0Gbps 5 x 3-pin fan headers 24-pin ATX; ATX12X; 1 x Molex
Front Ports:	2 X USB2.0 1 x 6-pin FireWire 2 X 3.5mm Audio (Headphone and Microphone)
Rear Ports:	4 X USB2.0 1 X FireWire (6-pin) 6 X 3.5mm Audio S/PDIF Optical and Coax Out PS/2 Keyboard and Mouse LAN (GbE) 1 X Serial and 1 X Parallel
Extras:	8-inch cold cathode light Acrylic faceplate with logo and light On/Off switch for lights on rear of case Carrying handle on top (removable) Case windows on left and right

Fastest CPU currently on the market? Check. Fastest graphics cards currently on the market? Check - at least somewhat. See, NVIDIA is launching their Quad-SLI platform, and the system as tested "only" comes with two 7900 GTX 512MB cards - BFG overclocked models, no less. If you want Quad-SLI, Falcon can actually work that into the FragBox case as well. Who needs Quad-SLI? To paraphrase Kelt Reeves, it's pretty much only for the filthy rich people running around with 30" Apple Cinema (or Dell 3007FPW) displays; 2560x1600 can actually bring even a mighty 7900 GTX 512 SLI or X1900 XT CrossFire setup to its knees.

Click to enlarge

What was that about using the FragBox with X1900 CrossFire? As you're likely aware by now, CrossFire and SLI are competing standards, and both companies currently refuse to allow their multi-GPU setup to run on the other's chipsets. ATI has at least worked with Intel to get CF running on 975X, but what we'd really like to see is for both companies to take off the shackles and let people use whatever multi-GPU setup they want on any motherboard with dual X16 PCIe slots. The uATX/SFF market needs help even more, as only recently have we managed to get an SLI uATX board and there are still no CF uATX designs. Given that there are quite a few games where X1900 XT CF comes out ahead of 7900 GTX SLI these days (and of course there are still areas where NVIDIA leads ATI), ideally the decision of what GPUs to run should be up to the customer. We don't need official support for CrossFire on SLI or SLI on CrossFire, but we'd love to see the driver locks removed once and for all. Is it likely to happen? No, but as we said at the start, dreams are free.

Other areas continue to go ultra-high-end as well. The system has two 150GB Raptor drives in RAID 0. If you prefer, you can get a couple 400GB Seagate drives for less money, or you can go for a full terabyte of storage complements of two 500GB Seagate drives for even more money than the Raptors. If you don't want RAID 0, you can also opt out of that feature and simply have one or two drives installed, or you can go with the safer route and have the system configured for RAID 1.

Click to enlarge

The one area that falls short is the audio. Realtek ALC850 isn't terrible, but it certainly isn't as good as even the original Audigy, let alone many of the other high-end sound cards. We talked with Falcon about that and asked to get a couple 7900 GT cards with an X-Fi sound card to use as an alternate configuration. Sure, you sacrifice some GPU performance, but depending on the monitor and resolution you run, it may or may not matter. The Realtek ALC850 is adequate for most tasks, but even with $55 Logitech speakers there was a clear difference in quality between the integrated audio and the X-Fi card. If you're planning on connecting the FragBox to some really nice speakers or headphones, we would seriously consider upgrading the audio subsystem. (That might mean you have to ditch the FragBox and go for the full Mach V setup if you're looking at FNW systems.)

So the system comes equipped with the best parts money can buy, but there's still the problem of that price sticker, right? How can anyone justify spending thousands of dollars on a computer? FNW charges a pretty hefty price premium for putting your system together, but they really do go all out in building and testing the system. Included with every FNW computer is a binder of documentation, and part of the documentation lists performance and stress testing results. Scores from all three of the latest Futuremark 3DMark applications are listed, so you can always refer back to them to verify whether or not your system continues to run at peak performance. There is also a total of about 3 days of stress testing conducted on the systems, including running Prime95 (one instance per CPU core) for 12 hours and a loop of 3DMark06 for 12 hours. You can be relatively sure that whatever FNW ships out will run fully stable - just remember to "pop the hood" and dust out the fans and heatsinks twice a year to maintain optimal heat dissipation. What it boils down to is that Falcon employs some very knowledgeable technicians, and if you want a really skilled computer person assembling and troubleshooting your system, you're going to pay extra. For some people, it's worth the peace of mind and support that you get.

The rest of the documentation provided with the system is also superb, and echoes my feelings on several points. For example, they suggest installing all of your games into C:\Games rather than into the default directory. This helps to keep your hard drive more organized, as the Program Files folder has simply turned into the new root directory as far as applications are concerned. Let's not even get into the discussion about games that install to C:\Program Files\[Publisher]\[Developer]\[Game Name] -- even when the developers haven't released any other games! I can't say that I really learned anything new from the manual and documentation, but it does have quite a few pointers that many people (and companies) could find useful.



Construction and Assembly

Still too expensive for you? There's an alternative you might consider if you really like the FragBox case. Falcon offers the FragBox as a DIY chassis for $299, or you can get it with a 600W modular PSU (with a "short cable" kit) for $518. Either way, you get the case as well as cold cathode lighting setup (including an inverter for the lights). Note that if you go the DIY route, you can only get the "FragBox" faceplate and not the "Falcon Northwest" option, which makes sense. Exotix paint jobs are also not offered for the DIY setup.

The case and PSU are made by SilverStone, a company known for building high quality cases. Add in the EVGA SLI motherboard and you're looking at about $700 total for a "barebones" configuration. If you really like the looks but you don't want to pay the price premium - not that $300 for a chassis is cheap, of course, but it's not too bad all things considered - you can buy the DIY kit and add your own modular ATX PSU and motherboard purchased at any of the typical online resellers. That will cut around $125 from the price of the barebones, though whether or not you're getting equivalent quality is up for debate. If you compare that price with the Shuttle SN26P and the Monarch Hornet Pro, the FragBox SLI DIY system is pretty competitive. Finally, you can also go with the SilverStone SG01 for $150 - that's the basis of the FragBox design. At that point, you lose the lights, acrylic face plate, side windows, and carrying handle. It's still a reasonable uATX case, but it's certainly not going to turn heads like the FragBox.

How does the system work as a barebones offering? First, it has to be said that the case isn't super easy to work with, but any competent enthusiast will be able to handle it. Falcon makes no claims about offering a simple-to-assemble barebones - quite the contrary. From their website: "Small Form Factors are by nature not as easy to work in as towers. There's not much room for knuckles, and sometimes a part will need to be removed to access another part. Patience is important." That pretty much sums up the situation. Once the system is properly assembled, you really don't have to worry anymore about how difficult the assembly process might have been. It's slightly more difficult than working with a spacious tower or mid-tower ATX case, but it shouldn't take more than an hour or two to put everything together. If you're the type of person that frequently upgrades components, the case is certainly not geared towards making your life easier; the more common infrequent upgraders will be far more concerned with appearance and performance than ease of assembly.

For those thinking of going the DIY route, we did take the time to dismantle the case, swap some components, and take some pictures. Here's a brief commentary on the process, as well as plenty of images. Larger 1280x960 images are available for the interested, but be warned that they weigh in at 150-200 KB each. We did not completely dismantle the system, as you will see in the images, but we did get down to the point where the only installed components remaining were the motherboard and a few other minor peripherals. You can also see where the wiring was placed by Falcon, which you may find useful if you're trying to assemble a system yourself.

Click to enlarge

Starting with the power supply, this is a full-size ATX unit. As you can see, it's also a modular power supply, although most of the cables are used for the FragBox SLI configuration we were sent. We mentioned before that this is a 600W power supply, and judging by the way this is a true 600W unit, and not one that simply makes that claim. The PSU has a large 120 mm fan, and the rear of the PSU has hexagonal perforations that allow for good airflow. In use, the power supply is relatively quiet, although as we'll see later this system as a whole certainly isn't silent.

Click to enlarge

Click to enlarge

A hard drive tray sits at the front of the case and can be removed for easier installation of the drives. More importantly for the high-end users, the hard drive tray is actively cooled by an 80mm fan, helping to reduce concern for data loss when running dual HDD configurations. This is particularly important if you plan on running RAID 0 - and we would still advise you to make frequent backups in that case. Once everything is installed, the various wires and cables tend to restrict airflow around the hard drives, but the active cooling should compensate for that.

Click to enlarge

Here you can see the shots of the system in a mostly disassembled state. You can already see that the number of wires coming off the power supply requires a lot of effort to fit into a case this small. Falcon lists the power supply as coming with a "short cable kit", but even then there's a lot of wires to deal with.

Click to enlarge

With just the motherboard installed, access to the CPU socket is relatively unrestricted. A support beam does run across the top of the case, but it doesn't get in the way. Considering that this is supposed to be something of a "small form factor" computer, you might be surprised at just how large your heatsink can be. Falcon ships prebuilt systems with Zalman CNPS7000 copper cooler. It's a relatively tight fit once everything is installed, as you can see from the images, but the Zalman cooler should keep temperatures lower as well as help to regulate noise levels. One issue that you might have with the Zalman heatsink comes from Falcon's top-end 1GB memory modules. They use Corsair TwinX 2048MB 3500LL Pro modules - complete with flashing lights - and due to their taller than average height and the large Zalman HSF, you can't install more than two DIMMs. FNW does give you the option of using some slightly cheaper and slightly lower performing Corsair TwinX 2048MB PC3200 modules, however. That As the move to Windows Vista and potentially a 64-bit OS and applications will make 4GB of RAM a conceivable upgrade during the life of the system, we would recommend avoiding the 3500LL Pro memory. Ideally, Falcon Northwest will offer another high-performance RAM option in the future -- or you could always just buy your own RAM when it comes time to upgrade.



Assembly (Cont'd)

Once the processor is installed, getting power supply back into place -- along with all the cables -- definitely requires a bit of finesse. Because of the way the wires are routed, we found it easiest to slide the power supply into place from the GPU side.

Click to enlarge

The first picture above has the power supply partially in place, and you can see that you have to have the optical drive pushed forward to accomplish this. A shelf helps to hold the power supply up, but the rear of the unit does tend to sag a bit. As we said, it's a relatively heavy PSU, which was good for quality and cooling does tend to put a bit more strain on the case. Once everything is installed, you'll find that the rear of the PSU basically rests against the rear of the DVD drive. Then all you have to do is clean up the cable clutter -- Falcon's generous use of zip ties certainly helps here, and we would recommend you use something similar.

Click to enlarge

With everything else installed, all that's left is to decide what sort of graphics you want to run, install them, clean up the last few stray ends, and close the system back up. You want to go with two single slot graphics cards and a sound card, or do you prefer the higher performance offered by to dual slot GPUs? That's about the only area where you really have to compromise in going with the micro ATX case and motherboard.

In terms of assembly and upgrades, the GPUs and expansion slots can be accessed without too much difficulty. The extra fan blowing across the GPU area does need to be removed first, but we'd rather deal with that than have the graphics cards overheat. In fact, we would say the rear case perforations that allow this fan to draw in air could actually do with being a bit larger in order to allow even more airflow, but that's a minor concern. This rear fan blows across the expansion slots towards the front of the case, where the hard drives are located and where a second fan serves as an exhaust.

The one area that requires quite a bit of work to access is the CPU socket. You need to remove the PSU to get at it, which means you also have to remove the screws from the optical drive and slide that forward. You will need to remove the GPUs as well in order to get the PSU out of the way. The process necessitates unplugging all of the internal connections to the PSU, of course, at which point it's relatively easy to get at the CPU socket. Naturally, if you were to replace the motherboard, you would need to remove everything from the case first, but that's a rare task.

Overall, the case design is decent, but it's geared more towards performance than ease of use. If we return to the aforementioned competitors, the Shuttle SN26P is still noticeably smaller, but it also lacks the ability to use two dual-slot GPUs - or two single-slot GPUs and an add-in sound card or TV tuner. The fact that it comes as a pre-assembled barebones with all of the necessary cables pre-routed also makes it the easiest to assemble by far. The Hornet Pro is about the same as the FragBox in terms of difficulty of assembly: some areas are easier, others take more time, but overall they're about equal. Ironically, the SN26P works great with the large Corsair DIMMs, though of course you can only use two DIMMs total in the Shuttle units.

If you're after SFF/uATX with SLI - or you could get a different motherboard with the uATX systems and forget about SLI support - any one of these three DIY/barebones options are worth considering. They're all on the higher end of the price spectrum, but they're also on the higher end of the performance and design spectrum. After all, you don't need SLI in order to play games, and you certainly don't need a smaller case - consider the higher prices something of a luxury tax for people that don't want your typical PC. For a high-end SLI setup, you're looking at about $200 extra to switch from an ATX case to one of these cases. $200 is a lot of money for budget and mid-range systems, but if you're looking at spending $3000+ on a top-performing system, an extra $200 really isn't a big deal. If you can't afford an extra $200, you probably shouldn't be looking at a $3000 system in the first place; whether you actually want the smaller case or not is the real question.



Benchmark Setup

A review of a system without comparisons with competing solutions wouldn't be much of a review, so we've pitted the FragBox SLI against two alternatives. For the first, we've used the Shuttle SN26P SFF, this time upgraded with a couple 7900GT cards - this is the maximum level of GPU performance currently available for the SN26P, as you're limited to two single-slot GPUs. The second system is one of the top configurations you can get in the full ATX market: a custom-built computer using a DFI LanParty SLI-DR motherboard.

We ran several different variations of the FragBox and DFI systems through the benchmarks, though the DFI motherboard failed towards the end of testing and didn't allow us to test every potential configuration. (If you're wondering, we'll take the blame for the motherboard failure - during testing, things can get a bit hectic at times, and trying to hurry while working with electronics is an iffy proposition.) In terms of configurations, we have the high-end setups with two 7900 GTX 512MB cards, and then we have a lesser configuration with two 7900 GT 256MB cards. We ran the 7900 GT cards with an X-Fi sound card as well as with integrated audio. Falcon also shipped us two 512MB DIMMs and two 1024MB DIMMs, and we tested both options.

We didn't bother testing the DFI or SN26P with 1GB of RAM, because we recommend 2x1GB memory setups for high-end computers. We also didn't get a chance to test the DFI board with X-Fi and two GTX cards - the motherboard failed as we were getting ready to run that particular configuration.

While we won't include new performance results for the Monarch Hornet Pro Revenge we tested last month, we will include the system noise levels for reference. In all the remaining tests, since the Hornet SLI and the FragBox SLI use the same EVGA motherboard, performance should be the same with identical configurations.

Here are the specifics of each configuration we benchmarked:

Falcon Northwest Fragbox SLI
Motherboard:	EVGA 131-K8-NF44-AX NF4 SLI (NVIDIA nForce4 SLI chipset) Micro-ATX
Processor:	AMD Athlon 64 FX-60
Heatsink:	Zalman CNPS7000 Cu
RAM:	2x1024MB Corsair TwinX 3500LL Pro 2-3-2-7-1T 2x512MB Corsair TwinX 3200C2 2-2-2-7-1T
Graphics Cards:	2 x BFG GeForce 7900 GTX 512MB OC 2 x BFG 7800 GT 256MB OC
Hard Drive(s):	2 x Western Digital 150GB 16MB 10K RPM Raptor Drives configured in RAID 0
Optical Drive:	Plextor PX-740A 16x DVD+/-RW
Floppy Drive:	Mitsumi 3.5 Floppy plus 7-in-1 USB Card Reader
Audio:	RealTek AC'97 7.1 (ALC850) Creative X-Fi ExtermeMusic
Power Suply:	SilverStone 600W Modular PSU
Drivers:	NVIDIA nForce4 6.70 NVIDIA ForceWare 84.43
Operating System:	Windows XP Professional SP2 with Updates

Shuttle XPC SN26P
Motherboard:	Shuttle SN26 proprietary form factor NVIDIA nForce4 SLI chipset
Processor:	AMD Athlon 64 FX-60
Heatsink:	ICE (Integrated Cooling Engine) 4G
RAM:	2x1024MB Corsair TwinX 3500LL Pro 2-3-2-7-1T
Graphics Cards:	1 x BFG GeForce 7900 GTX 512MB OC 2 x BFG 7800 GT 256MB OC
Hard Drive:	Seagate 250GB 8MB 7200 RPM 7200.8
Optical Drive:	BenQ 16X DVD-ROM
Audio:	VIA 24-bit Vinyl Audio 7.1
Power Suply:	Silent X 350W
Drivers:	NVIDIA nForce4 6.70; NVIDIA ForceWare 84.43
Operating System:	Windows XP Professional SP2 with Updates

Custom Full ATX Desktop
Motherboard:	DFI LanParty UT SLI-DR NVIDIA nForce4 SLI chipset
Processor:	AMD Athlon 64 FX-60
Heatsink:	Thermaltake Big Typhoon
RAM:	2x1024MB Corsair TwinX 3500LL Pro 2-3-2-7-1T
Graphics Cards:	2 x BFG GeForce 7900 GTX 512MB OC 2 x BFG 7800 GT 256MB OC
Hard Drive:	Western Digital 74GB 8MB 10K RPM Raptor
Optical Drive:	BenQ 16X DVD-ROM
Audio:	Karajan Module - RealTek ALC850 Creative X-Fi ExtermeMusic
Power Suply:	OCZ PowerStream 600W
Drivers:	NVIDIA nForce4 6.70; NVIDIA ForceWare 84.43
Operating System:	Windows XP Professional SP2 with Updates

All systems were tested with the latest drivers and BIOS versions available, though we opted for the "beta" 704-2bt BIOS for the DFI board, as this is regarded in the enthusiast community as being one of the best overall BIOS offerings for the platform. Not all of our gaming benchmarks support testing with sound enabled, but for several of those that do we'll be running tests with and without sound.

In order to speed up total testing time, there is one potentially significant difference between all three platforms: the hard drive. The Falcon system ships with two 150 GB Raptor drives, but we used two different hard drive options for the other systems. The DFI system used a 74 GB Raptor, and the SN26P used a Seagate 7200.8 250 GB hard drive. In most instances, the different hard drives won't have a major impact on performance, but we will note in the benchmarks when hard drive activity appears to be a factor.

It is important to remember that benchmarks are not the be-all, end-all when it comes to the "ideal" computer system. Features, performance, appearance, and acoustics are all important factors. If all of the systems perform similarly to each other in our testing, that will be the best possible outcome: you can then make your purchase decision based on the other areas.



System and Application Performance

We'll start with a look at system performance using several common business benchmarks. All of these tests are to varying degrees "synthetic" benchmarks, as user interaction is eliminated from the equation. For some tests (i.e. word processing, email, spreadsheets) the user is the primary bottleneck. In other tests like multimedia encoding, though, system performance can play a more significant role. We've included results for Winstones 2004, PCMark05, and Worldbench 2005. PCMark04 would not report a score on any of the tested systems, but this is a known problem with the benchmark when running on the fastest AMD dual-core processors. The third multitasking benchmark always reports a failure on the grammar test. In our experience, PCMark05 is more demanding of a system than PCMark04, and the scores are relatively similar, so this isn't a serious problem.







Winstones has a margin of error of about 3%, though it can be as high as 5% in regular testing. Most of the scores fall within the margin of error, and the faster hard drive subsystem of the FragBox almost certainly plays a small role as well. The end result is that all three systems, when comparably equipped, perform similarly. PCMark05 stresses the hard drive subsystem even more, with a couple test that measure maximum hard drive performance. There are also 3D rendering tests, accounting for the slight difference between 7900 GT and 7900 GTX configurations. The overall performance of the Falcon Northwest FragBox comes out on top by a small margin, which is important considering the price. We could almost certainly match performance if we used the same hard drive setup in our custom-built desktop system, but the main point is that the FragBox lives up to the promise of delivering performance matching that of the best desktop systems in a micro ATX form factor.

In terms of memory, having 2 GB of RAM versus 1 GB of RAM has little impact on these application benchmarks. That's not too surprising, given that none of these tests are designed to use more than 1 GB of RAM. Someone odd is the fact that the falcon Northwest system scores noticeably (about 10%) higher in the content creation benchmarks when equipped with the 7900 GTX cards. With the DFI system, the reverse is true: the 7900 GT configuration scores somewhat higher. We're at a loss to explain this discrepancy, as Winstones generally isn't affected by graphics card performance, short of using integrated graphics.



Encoding Performance

For our encoding tests, we used AutoGK 2.27 with Xvid 1.2 and DivX 6.1.1. Note that we have upgraded to the latest versions of AutoGK as well as more recent versions of Xvid and DivX. DivX and Xvid should now show improved performance with multi-core configurations, and scores are not directly comparable with older AutoGK results that we've reported. We're still encoding the same content (chapter 9 from The Sum of All Fears), though, and if you're interested in seeing how the change to the new versions affected performance, we did include results using the earlier versions on the FragBox system. (The earlier results are shown in red.)





There are slight variations in encoding performance, but basically all of the system configurations are tied. Again, this is what we like to see, because it generally means that all of the computer configurations are pretty well optimized.

Memory Latency

CPU-Z includes a program called latency.exe that can be used to get a rough estimate of cache and memory latency. We used version 1.33 and converted the latency from CPU cycles into nanoseconds.



While there are noticeable differences in latency, even when using the same memory timings, it's also important to realize that a 10% difference in memory latency will generally only result in a small difference in actual application performance. Here at least the EVGA and Shuttle motherboards do not appear to be as well optimized as the DFI board. There is of course the possibility that a future BIOS will improve performance slightly, but it's a minor concern.



Synthetic Gaming

While some might argue otherwise, we really have trouble qualifying Futuremark's 3DMark applications as anything but synthetic benchmarks. They basically represent another potential 3D gaming engine, but unlike actual games, the engine is not used for anything in the real world. That doesn't make them useless, and they do provide yet another gauge of overall system performance. However, given the choice between 10% higher 3DMark scores and 10% higher scores in a real game, the latter would get our overwhelming vote.

















Our DFI system comes out on top by small margins in virtually every configuration, but there's nothing that we would really consider significant. Unlike PCMark, 3DMark depends almost exclusively on graphics card performance, and to a lesser degree on CPU performance. The CPU scores naturally focus more on that area, and the FragBox lags behind slightly in 3DMark03 CPU scores. In other tests, the tighter memory timings of the 1GB setup moves that configuration to the top of the charts. An odd twist is that the FragBox with the 7900 GT performs better in the CPU test of 3DMark05 than with the 7900 GTX; we are not entirely sure why that occurs, and we wouldn't place too much significance on it. These are synthetic benchmarks, and the end result is that we've once again basically got a tie.



Introduction to Real-World Gaming Benchmarks

Now we come to the benchmarks that most of you are probably really interested in seeing. After all, you don't buy a system with two 7900 GT/GTX cards just to surf the Internet, run benchmarks, and work in office applications! With any system of this level, most games are going to be CPU limited at lower resolutions. However, there are many people that use 19" LCDs at 1280x1024 resolution, so we'll start the benchmarks at that resolution and move up through 2048x1536.

We'll also include the increasingly common widescreen resolutions of 1440x900, 1680x1050 and 1920x1200. (The lowest widescreen resolution will generally be CPU limited as well, but even that is useful information.) Many games support those resolutions natively, but quite a few others require some hacks to get proper widescreen support. If you're not already familiar with the site, we definitely recommend owners of widescreen displays check out the Widescreen Gaming Forum for solutions on getting most modern games to run properly.

We've tried to reduce the total number of graphs to keep things readable, so the charts will show resolution scaling, and we've separated most of the GTX setups from the GT systems. The charts for the 7900 GT configurations will also show the shuttle system with a single GTX card, as well as the FragBox with GTX SLI. That should help to keep things in perspective. All tests will be run with 4xAA and 8xAF, with the exception of FEAR. We will run FEAR with 0xAA and 8xAF as well as with 4xAA, due to the fact that FEAR is more GPU limited than the other games.

On a separate note, let's talk a bit about gaming benchmarks. We'd love to include some newer benchmarks, and some non-FPS benchmarks would be particularly helpful. The problem is in creating repeatable tests as well as making effective use of our time. Without the ability to record and play back a demo file, benchmarking is more difficult and less consistent. With many games shifting to fully 3D environments, we can only request that more developers add proper benchmarking support. Elder Scrolls: Oblivion is the latest entry in a list of great 3D games that make it difficult for us to use due to the variability in benchmarking as well is the time required. That's why you'll see us devote separate articles to games we feel are important to benchmark, but we may not continue to use those games in all future reviews.

If you're involved with the development of any commercial game and would like our input on benchmarking features, please feel free to send us an email. Considering that companies like Futuremark earn money solely from the creation of benchmarks, it seems like the repeated use of a game for hardware reviews can only serve as a form of nearly-free advertising. We encourage all development studios to make benchmarking features an integrated part of their gaming engines, as an extra week or so of development/testing time could very likely pay back in other ways. This request goes double for strategy and RPG games, where we're still lacking any recent games with adequate benchmarking tools.

With that said, the games we're testing today are all essentially FPS shooters. Popularity also plays a role in the games we select, and so we have a set of six games that use five major 3D engines. There are other games that we've benchmarked in the past (and some of us continue to use these), but at least one of the games tested here will generally have similar performance characteristics to just about any other current game. We're always looking for new titles to include, of course, and hopefully we'll be adding a few new titles in the near future and phasing out some of our older games.



Battlefield 2, v1.22

We've updated our BF2 benchmark (again), which means that old scores have been invalidated. It's really difficult to understand why a game patch necessitates new demo files, and ideally gaming companies will start to make demo file support backwards compatible if at all possible. If there's a good reason for the change, that's acceptable, but DICE has never been forthcoming about what "upgrades" have been made to the demo recoding/playback features.

We've tried to keep the demo file somewhat consistent, so we're still using the Operation Clean Sweep map with 16 players and a variety of camera angles including shots for infantry, planes, helicopters, tanks, etc. Other levels in Battlefield 2 are not quite as demanding on the system, but we wanted a large map with water to present something of a worst-case scenario. (Granted, getting a demo with 64 players would represent a better worst-case scenario, but creating such demos on our own is a bit difficult.) We didn't even bother running tests without 4xAA, as the game is almost entirely CPU limited on the top SLI configurations we're testing.





All of the platforms offer reasonable performance under Battlefield 2. Even the Shuttle system with a single GTX card manages to remain playable, and in fact it gives the 7900 GT SLI configurations a run for the money. Considering a single GTX card starts at $480 while two 7900 GT cards will run $580, not to mention the fact that you don't have to deal with SLI incompatibilities or headaches, and you can make a good point for going with a single fast GPU. That's something we'll keep an eye on throughout the rest of the game benchmarks, but of course the bottom line is that if you want maximum performance, the FragBox SLI system is about equal to the best full-ATX desktop systems.





Enabling audio causes quite a bit more load on the CPU, making Battlefield 2 largely GPUs limited. With dual GTX cards, the FragBox is about 17% slower than the DFI system -- this despite the fact that they both use the same Realtek ALC850 codec. Looking at the 7900 GT numbers, equipping the FragBox with an X-Fi card doesn't really help performance, though again we must say that it definitely sounds better. Something else to pay attention to is how much having only 1GB of RAM hurts performance in this configuration. It didn't seem to have much of an impact with the GTX cards, perhaps because they have 512MB of RAM each, but the 1GB GT SLI configuration is easily outperformed by a single GTX with 2GB of RAM.

We also tracked loading times for Battlefield 2, and though we don't have a graph made up, I can give you a couple quick numbers. The FragBox takes 100 to 140 seconds to load a level with sound when using to GT cards, the X-Fi, and 1GB of RAM. Upgrading to 2GB of RAM cuts the load times down to 50 to 80 seconds. That's the amount of time it takes to start a Battlefield 2 and load a level, but transitions between levels will also be substantially faster with 2GB of RAM.



Doom 3 v1.03

Doom 3 has been around for quite a while, but the engine is still impressive and there are several upcoming games that will use tweaked versions it. We continue to use the included Demo1 file for testing - imagine, a demo file that shipped with v1.0 of Doom 3 that still works after three version updates! While Doom 3 was an incredibly stressful game in terms of GPU requirements when it launched, the configurations we're testing today have become almost completely CPU limited without enabling 4xAA. We won't include the 0xAA results, but suffice it to say that frame rates were steady at 118-120 FPS all the way up to 2048x1536, which finally dropped performance of the 7900 GT system all the way to 115 FPS. If you don't feel 4xAA is necessary, you can pretty much set the game at the highest detail settings - even including Ultra quality - and spare no thought for graphics performance.





All of the tested systems perform about the same, whether they're equipped with two GT cards or two GTX cards. Even at the highest tested resolution, GTX SLI only holds an advantage of about 21% over GT SLI. Meanwhile, the advantage of GT SLI over a single GTX card ranges from 20% at the lower resolutions to almost 50% at 2048x1536. While Battlefield 2 didn't show a major advantage for GT SLI over a single GTX, Doom 3 clearly benefits from the additional memory bandwidth.

Quake 4, v1.2

Quake 4 uses the same engine as Doom 3, with one major difference: the latest official patch to Quake 4 adds multi-core CPU support. Call of Duty 2 is the only other PC game currently shipping with any form of SMP support, though our initial tests with the feature weren't particularly successful. The Q4 SMP support is now further along in development with the 1.2 patch, and there's a menu item that allows us to enable the feature. Unfortunately, our old demo file no longer works, so we recorded a new demo covering the same game sequence (basically, the first minute or two of gameplay after the introduction).

Amazingly, the demo files for Quake 4 are absolutely massive: our relatively short demo checks in at a whopping 140MB! In contrast, our two HL2 demo files combined are 5MB, and the 4 minute BF2 demo file and camera file are only 1.5MB. Needless to say, we don't plan on sharing our Q4 demo file - you're better off creating your own. Doom 3 also has relatively large demo files, though they compress a lot better; we're also happy to see that we can still use the original demo1 that's included with the initial shipping version of Doom 3.





Frame rates are substantially lower in Quake 4 than in Doom 3, but part of that can be chalked up to the fact that for testing with Ultra Quality mode in Quake 4. The higher resolution textures didn't seem to do much in Doom 3, but for whatever reason I can definitely see a difference in Quake 4. Starting with the GT graph, all of the systems are grouped together -- even the single GTX configuration falls in line with the GT SLI performance. It appears that the amount of texture swapping that occurs with 256MB graphics cards hinders performance, as the GTX SLI configuration is 65 to 100% faster.

Moving to the GTX performance, Ultra Quality once again demonstrates a game that can make use of more than 1GB of RAM, though the benefits seem to taper off as the bottleneck is moved back to the graphics card. SLI also offers a huge performance boost in Quake 4, as the single GTX in the SN26P is outperformed by 75 to 92%. The combination of dual core processor support with a typically GPU limited game definitely puts multiple GPU configurations in a good light. As for the FragBox, performance is once again basically equal to a similarly equipped desktop system.



Far Cry v1.33

Far Cry is easily the oldest of the titles I'm still using for benchmarks. Amazingly, the graphics quality still competes well against the latest games on the market. The rendering of a bright, tropical environment is certainly different from most other titles, and we're looking forward to seeing CryTek's Crisis game in the future. With Far Cry, we record average frame rates from three passes on the following Ubisoft demos: Regulator, Research, Training, and Volcano. We then average the results from these four maps to come up with a composite score. If you're interested in the level-specific results, we can provide those as well, but in order to avoid generating a ton of different graphs we averaged the results.





At lower resolutions, Far Cry is still CPU limited, though with frame rates in the mid-100s, you're not really losing much in the way of performance. As resolution increases, frame rates eventually begin to drop, but everything short of 2048x1536 is more than playable with a single GTX card. The systems are once again pretty close to each other, with the DFI system being a bit faster at lower resolutions, while the FragBox takes the lead by a small margin at higher resolutions. Far Cry is another game that clearly doesn't benefit from more than 1GB of RAM, though as with most games, loading times are noticeably faster.





Enabling sound in Far Cry doesn't cause nearly as much of a performance hit as in other games, although the FragBox scores do drop about 4%. On the other hand, the shuttle and DFI systems, as well as the FragBox X-Fi configuration, all show less than a 4% performance decrease. That typically means the bottleneck is something other than the CPU, or else the sound subsystem is simply not as CPU intensive as in other games.



F.E.A.R. v1.03

Whereas many of the other games become largely CPU limited with systems this powerful, FEAR is a game that can pretty much take as much GPU power as you can afford to throw at it. Performance shows a steady downward trend with increasing resolution, even on 7900 GTX SLI configurations. Enabling 4xAA (or soft shadows if you think those look better) only accelerates the trend. If the FEAR engine is a sign of things to come, more than a few people are going to need to upgrade their PCs over the coming year. Thankfully, for the time being it appears to be the exception rather than the rule - though Oblivion now seems to lead in terms of overall system requirements.

Incidentally, FEAR v1.04 was released while I was performing benchmarks. I did install that and run a few test benchmarks on one of the systems, and found that the new version performed essentially the same as the older version. I also performed some quick testing with sound enabled, and found that the game is so GPU limited that the impact of turning on sound is less than 3% in almost all cases -- not enough to worry about, in other words. Also, I did run a few benchmarks with soft shadows enabled. Performance was far worse than with anti-aliasing, and was generally unplayable. ATI's X1900 CrossFire setup apparently does better, but basically the soft shadows are more of a check mark than something most people can currently use and still get playable frame rates.









Given that FEAR is so GPU limited, it's not too surprising to see most of the systems performing all but identically. The DFI results are missing for the 7900 GT, due to the motherboard failure mentioned earlier. As for the GTX results, I was about ready to retest them when the motherboard gave up the ghost. As you can see, the scores are clearly better than a single GTX, but they're nowhere near as fast as the FragBox. I have a feeling that something was misconfigured in the NVIDIA drivers, but I was unable to verify this.

That does bring up something worth mentioning: don't expect every game to work immediately with SLI, even the ones that are supposed to be SLI-ready. Patches to games can sometimes cause issues, and often times you will be forced to use a tool such as nHancer in order to create an SLI profile that works. Most likely, the DFI system was running in a sub optimal SLI mode, so I left the numbers in to illustrate a point. Plan on doing a bit of extra legwork now and then in order to get games running properly with SLI.



Day of Defeat: Source and Half-Life 2

The Source Engine games so far show similar and yet different performance. The games are largely CPU limited except at the highest resolutions. The major difference is that where HL2 ranges from 110-170+ FPS depending on level, DODS tops out in the high 60s, and the Lost Coast level strikes something of a middle ground. DODS and Lost Coast also include a form of HDR support - we say "a form" because they don't use FP rendering, allowing them to do both antialiasing and HDR even on older hardware. At present, only ATI's X1000 series cards support AA+FP rendering, which unfortunately means that most games aren't even looking at supporting the feature. I decided to focus on DODS and the original HL2, so no Lost Coast numbers this time. (It's only one level, so I can't really consider it a true game benchmark.)









Half-Life 2 is pretty much completely CPU limited, even at the highest resolutions. 2048x1536 does begin to show some slight separation, but essentially all of the tested setups are more than fast enough. Day of Defeat: Source is in a similar situation, only with much lower frame rates. However, at 1920x1200 and 2048x1536, the GTX SLI configurations take a clear lead. What happens when we enable sound?









Being CPU limited at most of the tested resolutions, we basically see a mirroring of the earlier results. At the lower resolutions were the CPU is the bottleneck, enabling sound has a substantial impact on performance -- at 10 to 20% performance drop, roughly. At the higher resolutions with the GPU starts to become the bottleneck again -- and this is illustrated best by the 7900 GT scores -- enabling sound doesn't have nearly as much of an impact.



Noise and Power

The last area we would like to look at is the noise levels and power requirements of the systems. We did not include results from the custom desktop system, as it has completely different options. Results for the Monarch Hornet Pro Revenge are taken from the previous review - note that eight uses a slightly different graphics card configuration.

Noise Levels

Noise testing was conducted using an SPL meter with A-weighting at a distance of 1 foot and 4 feet. The ambient noise level was under 30 dB during testing -- the limit of our meter.









It is unreasonable to expect any system this powerful in a case this small to be anything near silent. Due to the differences in parts used, you should also not read too much into differences of one or 2 dB. During testing, the FragBox never got any louder or quieter. The cooling is adequate to keep the system stable -- we didn't have a single crash that we would attribute to overheating (flaky applications on the other hand...). At 51 dB, the FragBox is about as loud as a typical enthusiast desktop system, though there are plenty of water-cooled systems that are far quieter. While it would be nice to have something a bit quieter, the reality is that performance and stability are far more important for systems like this.

System Power Draw

As usual, we measured system power draw at the outlet using a Kill-A-Watt device. Idle power draw was measured at the Windows desktop, while 100% load was generated by running two instances of Folding@Home while looping 3DMark05.





Far more impressive than the actual system power use is the fact that the FragBox SLI matches the Monarch Hornet Revenge while offering substantially more performance. That is not say that the Monarch system couldn't equal the performance level, but rather that technology has progressed in just the last few months to the point where you can get more performance without needing any more electricity. We're looking at dual core Athlon FX-60 compared to single core Athlon FX-57, as well as the changing graphics cards. The chief reason that the FragBox is able to match the maximum power draw of the Hornet Pro Revenge is largely due to the process shrink from 110nm to 90nm that occurred with NVIDIA's release of the 7900 series.

If anyone would really like to see numbers from the SN26P, let me know and I can put the system back together. Basically, due to the lesser graphics cards used, it should be about 40-80W lower in terms of power draw. The important thing is that the Shuttle system was able to handle an FX-60 processor as well as two 7900 GT cards with its standard 350W power supply. The SN26P was completely stable throughout all testing.



Final Thoughts

How does the Falcon Northwest FragBox SLI stack up against the competition? If the overall purpose is to match full ATX desktop performance using a micro ATX motherboard and case, I have to conclude that they've succeeded in that goal. However, there are still reasons to consider going with a full ATX system. Expansion options will always be better with the larger case.

With the FragBox, if you want to install two of the fastest graphics cards on the market, you're stuck with whatever integrated components come with the motherboard. All of onboard options will get the job done, but we really would like to see something better than the ALC850 audio codec in the future. If you're hoping to add something like a TV tuner down the road, you'll have to go with an external solution. These problems aren't serious enough that people should avoid going with the smaller systems, of course; small form factor computers have always traded features and expansion options for size. As a whole, then, the FragBox does an excellent job at meeting the demands of a niche market.

I can't really evaluate the support that comes from Falcon at this time. We're working on trying to come up with some reasonable ways to conduct "tech support benchmarks" for the future, but for now all I can say is that Kelt Reeves was more than happy to help out with any questions I had, and hopefully that same level of dedication is present in the rest of the company. Such attitudes often start at the top, though, so it's good to know that Falcon's President and founder is sincerely interested in helping out when possible. Interplay Productions (R.I.P.) used to have a slogan: "By gamers, for gamers." Based off my contact with Kelt Reeves, FNW has a similar attitude: these are systems built by enthusiasts, for enthusiasts.

In many ways, the results are almost a foregone conclusion. The Falcon Northwest FragBox SLI is fast - very fast. In fact, it's the fastest computer I've ever used/tested! It's also the most expensive system I've ever tested. For the DIY crowd, the price/performance offered is almost laughable. Any serious enthusiast can build an overclocked system that will come very close to the performance of the FragBox SLI - and perhaps even exceed it - for about half the cost. However, there are many people out there that have no interest in overclocking, and they don't want to bother assembling the system on their own.

Some people enjoy buying a sports coupe and modding it until performance can come close to the most exotic cars on the market. Others prefer to just go out and buy a factory built Ferrari and not deal with any of the dirty work. Neither choice is right, and it's a matter of determining how much time you want to put into upgrading and maintaining your car vs. the money you spend. Of course, both of those groups are a tiny minority in comparison to the vast numbers of people that just want a decent, reliable car for a reasonable price.

That analogy applies relatively well to the computer industry. There are people that want the best pre-built system money can buy, others that want to build the fastest overclocked system possible for minimum cost, and a huge group of business users that wouldn't know a benchmark if it walked up and bit them on the face. (Ed: That happened once!) For the first group, Falcon Northwest is among the elite when it comes to building exotic systems, and they can help interested buyers put together a screaming fast system. FNW also happens to get access to limited release technologies (7800 GTX 512 MB and Quad-SLI, for example) that may be difficult to pick up on the retail market.

Not everyone needs or even wants that level of performance, and more importantly few people are willing to pay the price. We're certainly not going to tell all of you to go out and spend a small fortune on a top-performing computer that will be middle-of-the-road in a year's time and merely average in two years. Then again, we're not going to suggest that everyone out there stick with moderate performance for a moderate price either. As the saying goes, if you have to ask how much it's going to cost, it's not for you. If you're more interested in how fast you can make it run, Falcon Northwest can surely help you reach performance nerd-vana. The rest of us can only dream.