Home The making of the new Damagebox

The making of the new Damagebox

Scott Wasson
In our content, we occasionally include affiliate links. Should you click on these links, we may earn a commission, though this incurs no additional cost to you. Your use of this website signifies your acceptance of our terms and conditions as well as our privacy policy.
EVEN THOUGH I run a computer hardware website and spend my days playing with all of the latest goodies, I don’t actually upgrade my own PC all that often. I learned long ago not to stick review hardware into my own PC, for fear of messing up the main system on which I do my work. Swapping cutting-edge hardware in and out of your computer constantly is like doing cartwheels on a high wire—you’re seriously asking for trouble. Besides, moving all of your stuff from one PC to the next is a trying process, and I’d rather avoid it whenever possible.

My last PC upgrade was just over two years ago, though, and I recently decided that it was time for me to upgrade again. My old system wasn’t horrible; it was based on an Asus SK8N motherboard with a single Socket 940 and 2GB of registered DDR333 memory. However, the SK8N wasn’t exactly a stellar board, based as it was on the iffy first-gen nForce3 chipset. This particular mobo was a pre-production job that I’d used for an early preview article and then snagged for my own use. On top of that, the thing was progressively dying. The NIC never did work right, and a few months ago, the secondary IDE controller (based on a Promise chip) gave up the ghost, robbing me of some storage capacity. (I lost access to my 40GB MP3 drive!) Then the system started locking up at random for no apparent reason—the kind of hard lock-ups that suggest hardware problems more than software, at least at first glance. I wasn’t about to spend hours trying to troubleshoot an intermittent problem of that sort.

That was the final straw. It was time to build a new PC.

Of course, it didn’t hurt that PCs have gained a broad range of nifty new features over the past couple of years. The combination of dual-core processors and Native Command Queuing for SATA hard drives promises new heights of multitasking bliss in desktop PCs, and I can’t very well prattle on about the creamy smoothness of SMP without partaking of its delights myself. The transition to PCI Express has also limited the upgrade options for older AGP-based systems, especially when it comes to cutting-edge graphics cards, and that’s no fun at all. Having a second PCI Express slot for some dual-graphics action might be nice, at some point, and you don’t get that with AGP. I also like a quiet computer, and the SK8N doesn’t support AMD’s Cool’n’Quiet dynamic clock speed and voltage throttling capability. All of this new stuff would certainly be nice to have.

Plus, I had some additional objectives that I wanted to accomplish with a new PC build. Because my PC is my main work system, our always-on local file and print server, and holds gobs of precious testing data, I needed a better storage solution—namely, a large RAID 1 mirror. I also wanted the system to look nice and have a few new bells and whistles that my last box didn’t have.

The parts
So I gathered together whatever hardware I could manage, and I built a new system. Here’s a quick look at the components, with my justification for each choice. Note that my hardware choices weren’t intended to be the most EXTREME possible, as if my PC were auditioning for some Mountain Dew commercial. I was aiming for a good price-performance ratio, and I wasn’t shy about using perfectly good components from my parts shelf or my old PC.

  • AMD Athlon 64 X2 3800+ processor — The Athlon 64 X2 is undoubtedly the processor of choice these days, and the 3800+ is the most affordable model. Given its likely overclockability, I don’t see much reason to pay more for a higher model. As I said in my review of the X2 3800+, taking this chip to 2.4GHz makes for a very livable overclock:

    Now, that’s a sweet overclock all by itself, but hitting 2.4GHz has the added benefit of bringing everything into line. When the memory clock is set to the proper divider for DDR333 operation and the HyperTransport clock is raised to 240MHz, the memory actually runs at 400MHz even. Lock down the PCI and PCI-E bus speeds using the motherboard’s BIOS, and you’re running virtually everything but the CPU and HyperTransport link at stock speeds. I was able to leave the RAM timings at 2-2-2-5, nice and tight. This is the sort of overclock I could live with for everyday use.

    Having a pair of K8 cores running at 2.4GHz should make for decent performance, too.

  • DFI LANParty UT nF4 SLI-DR motherboard — I’d used this motherboard for quite a bit of CPU testing, and it had been solid and trouble-free for me. Not only that, but the tweaking options that DFI has built into this board are extensive, and the mag-lev chipset cooling fan is reasonably quiet for an nForce4 mobo without a fancy heatpipe like the Asus A8N-SLI Premium.

    Also, this board happens to look totally pimp under black light—not that a mature, responsible individual like me would care about such things. But I did use the included bright-yellow I/O cables to connect all of the drives, and there are a couple of cold cathodes in the case. Fo shizzle.

  • Four Corsair XMS3200XL 512MB DIMMs — I’ve had 2GB of memory in my Pee Cee for years now, and no way was I going to settle for less. Corsair is a great brand, but frankly, I wouldn’t hesitate to choose any of the major enthusiast brands of memory these days, including Kingston, OCZ, Mushkin, and Crucial, to name a few. All feature lifetime warranties with no-hassle replacements, and with the exception of Crucial, they all use the same pool of chips to make their products. (Crucial is an arm of Micron and doesn’t tend to shop around like the others.)

    As for whether or not it’s really worth paying a premium for fancy low-latency or high-clock-speed DIMMs, well, we just published an article examining that very issue. The bottom line: value-wise, you’re probably better off going with cheaper RAM that runs at slower timings. Nevertheless, I went with some fancy DIMMs that could give me very low latencies through tight 2-2-2 timings, because I had them on hand.

  • XFX GeForce 7800 GTX 256MB graphics card — This card was the Editor’s Choice winner in our 7800 GTX round-up, so I definitely liked it. I have to admit, though, that I picked this card mostly because I already had it on hand and available. If I were buying a new card for myself, I’d probably be choosing between a GeForce 7800 GT and a Radeon X1800 XL. The 7800 GT would most likely get the nod on the basis of broader availability, current street prices, and its ability to run in SLI mode on my nForce4 motherboard.

  • M-Audio Revolution 7.1 sound card — This one is a transplant from my previous system. The Revo initially captured our hearts a couple of years ago when it did what few other cards at the time could do: reproduce audio streams with clarity and 24 bits of precision using decent quality DACs for under $100. Since then, the perennially retarded sound card market has made little progress, with the notable and happy exception of Creative’s Sound Blaster X-Fi audio processor. M-Audio has done almost nothing to improve the Revo’s drivers, which have never been all that great for gaming. I still enjoy the Revo’s crisp sound, but I may pony up for an X-Fi card soon and relegate the M-Audio card to service in a home theater PC.

    This motherboard also has relatively decent built-in audio thanks to DFI’s Karajan audio module, which situates the Realtek audio codec on a separate riser card. Rather than disable this second sound subsystem, I connected some external audio ports on the top of the case to the integrated audio module so that I can use a headset for voice-enabled apps without having to disconnect the speakers from the Revo’s outputs around back.

  • Two Maxtor MaXLine III 250GB hard drives in RAID 1 — The MaXLine III is a 7200-RPM drive with SATA, Native Command Queuing, and a 16MB onboard cache—not a bad drive. It’s very similar to the DiamondMax Plus 10, which we liked pretty well when we reviewed it. Even more importantly, Maxtor’s NCQ implementation performs well in multitasking scenarios, which is important given the way I use my PC. Also, I had two of these, pre-production eval units, on hand. I wouldn’t trust a single pre-production drive with my data, but two of ’em in a mirror is OK, I suppose.

    I seriously considered dropping the cash on a pair of 400GB WD Caviar RE2 drives instead, because I think they’re the best all-around 7200-RPM drives available.

  • WD Raptor WD740GD 74GB hard drive — Even without NCQ, the Raptor is still the fastest SATA drive around, thanks to its 10K-RPM spindle speeds. Unfortunately, the Raptor’s usefulness is seriously hampered by its relatively small 74GB capacity. I had one Raptor on hand that I considered making my Windows boot volume, perhaps as part of a RAID 0 array with a second one, but I didn’t want a single drive’s failure to force me to reinstall the OS. In the end, I decided put the Raptor’s 10K-RPM speeds to use conquering a pet peeve of mine: slow level load times in games. The Raptor now serves as my game install volume. If I lose the data on that drive due to a drive failure, it might crimp my leisure time, but it won’t knock me totally out of commission. I’ve also configured Windows to use the Raptor for its page file, giving me an additional, speedy physical volume for paging.

  • Sony DVD-U10A DVD RW combo drive — Transplanted from my old computer. No, it’s not dual-layer, but it writes DVDs on either media type (plus or dash), and it hasn’t given me much trouble. Also, it has a black front face, so it matches the scheme. I couldn’t be bothered to look for a faster drive at this point. I also threw in a Sony DVD drive to make disc-to-disc copies easier and to give me another slot for game “play” CDs.

  • Mitsumi combo floppy/6-in-1 flash card reader — Also with a black front face. This gives me the ability to read a floppy disk in case I really must, but it also enables the computer to read virtually all of the popular flash card formats, including CompactFlash, SD, SmartMedia, and Memory Sticks—all in one 3.5″ external drive slot. These things rule.

  • Generic 56K modem — This allows me to record phone interviews, and I could use it to dial up and get online in the event of a broadband outage. Otherwise, I hate modems.

  • Enermax 565W power supply — From the parts shelf. It’s nothing too special, but it’s from a decent brand and has a nice, big rating. Should allow me to run SLI if I want.

  • CoolerMaster WaveMaster ATX case — This enclosure has a swanky brushed-aluminum look, a built-in case window, and doesn’t run too terribly loud. I really liked the Antec Sonata that housed my old PC, but its PSU and cooling were iffy for newer hardware. The Sonata was quieter than the WaveMaster, though, even with an auxiliary fan installed. I’m seriously considering moving this new system into a Sonata II soon. The WaveMaster looks great and will serve for now, but its drive cage placement makes for cramped quarters and messy cable routing.

Those are the basic components. I don’t think I’ve missed anything too important. I didn’t have the production budget or aesthetic aspirations of a Superbowl halftime show, but I did want a nice, fast, solid PC.

To x64 or not to x64?
One of my big goals at the outset of this project was to install and run Windows XP Pro x64 Edition as the OS on this new PC. To that end, I checked around to make sure that my components were WinXP x64 compatible. Sure enough, pretty much all of ’em had 64-bit drivers by now, at least in beta form, so I ordered up an OEM copy of the operating system. After it arrived, I realized that I hadn’t checked on drivers for my HP all-in-one printer/copier/fax/scanner/toaster oven. Turns out that HP doesn’t yet have 64-bit drivers available for this printer, and the OS doesn’t ship with even basic drivers for it. It’s currently not possible to print to this printer from a WinXP x64 machine, even when printing over a network through a print server.

I like the idea of going to Windows XP x64 Edition. I think 64-bit computing is a good and useful thing, and I like that the Windows Explorer interface is more multithreaded in x64. It feels faster. Also, WinXP x64 has better scheduling for multi-core processors.

But I darn well need to print from my computer, and I wasn’t going to give that up. So back to the 32-bit version of Windows XP it was, and the x64 Edition sits on the shelf, waiting for a better day. I’d bet that lots of folks have had similar experiences with this OS, and it’s a shame that 64-bit stuff hasn’t gotten more traction in desktop computing. I’m afraid that it won’t until Windows Vista arrives.

Paging Oskar Wu!
I chose the DFI LANParty NF4 SLI-DR motherboard because it had been a model citizen for me when I’d used it in past testing for reviews, but when it came time to build my own system around it, things changed. Seems like they always do when I’m working on my own, personal PC. I can pick my way around problems with finicky pre-production hardware everyday while working on a review, but when I start working on a PC for myself, the wheels come off. Previously stable hardware develops a serious need for counseling, and my IQ drops 60 points, making troubleshooting nearly impossible.

Foolishly confident in the DFI board and in my ability to build a system, I installed pretty much all of the core hardware for this system into the case, attached everything, and then fired it up. No go. Thus began a long troubleshooting process in which I disconnected absolutely everything unessential from the DFI board, removed it from the case, and was still unable to get it to POST. Power would come on, the screen would remain blank, and three of the board’s four boot-time troubleshooting LEDs would light up. Then nothing. I spent what must’ve been hours working through the problem, swapping out different configurations of video cards, memory, and processors.

After some work, I was able to get the system to POST, but only intermittently. The problem seemed to be related to several things, including:

  • The number and positions of the DIMMs in the memory slots
  • The PCI Express slot used for graphics and possibly the type of card
  • The type of memory used

I found that when I varied these things, the system seemed more likely to POST. With lots of different types of cards, CPUs, and DIMMs on hand, I was able to rule out any one component as the sole source of problem. Of course, I tried flashing it to the latest BIOS rev, and I attempted various BIOS tweaks, like very conservative memory timings and higher DIMM voltages. Nothing I did would make the thing POST reliably from a cold boot.

I googled around and found that lots of other folks had run into similar problems with this board, but I didn’t find any definitive solutions. Then I got lucky, I suppose, and discovered something about this problem, at least in my case, that was something of a fix. Turns out that if you leave the board on for a while after powering it up, like two to five minutes, it will pretty much always go ahead and POST eventually. You’ve just got to be patient.

In the end, I basically tweaked the board for stability and performance, using the built-in Metest86+ bootable option in its BIOS to test, and left it at that. I don’t reboot all that often, and so long as I know that the thing is going to come back out of its momentary coma and POST, I can live with the intermittent boot-time delay. Perhaps a newer BIOS or some magic memory timings tweak will fix the problem once and for all, but until then, I’ll live with it.

Installing Windows and.. installing Windows again
My next challenge was installing Windows XP, a straightforward task I’ve done tens, if not hundreds, of times. To make things even easier on myself, I decided to use an old trick that would let me avoid having to make a floppy disk with the NVIDIA RAID drivers on it: install the OS on a single drive, then install the RAID drivers, and then use Ghost to copy an image of the boot drive to the RAID 1 array. I’d done it multiple times in the past, and it seemed like the right way to go.

So I installed the OS from an installation CD, immediately installed Service Pack 2, and then installed the NVIDIA chipset drivers, because those include the basic level drivers necessary for things like Ethernet networking. Upon rebooting, though, the system wouldn’t boot into the OS properly. After several failed boot attempts, I decided that the NVIDIA IDE drivers must have been causing a problem, so I told WinXP to boot from the Last Known Good Configuration. Doing so allowed the OS to boot, but it also made this Windows install into a total head case. I uninstalled and reinstalled the NVIDIA drivers multiple times, but every time the NVIDIA storage drivers were installed, they rendered the system unable to boot. (Yes, I know I could have simply used the Microsoft drivers for storage, but that won’t work for RAID, and it robs you of NCQ.)

Believe it or not, I had to start over again—wipe the drive and reinstall Windows. Hard to believe that one would run into such basic problems with the nForce4 at this late stage in its lifespan.

The second install went more smoothly, even though I followed the exact same procedure as before: install the OS on a single drive, install SP2, and then install the NVIDIA chipset drivers, including the IDE drivers. This time, the OS would boot just fine with the NVIDIA storage drivers installed. However, I didn’t yet have the NVIDIA RAID drivers installed, because the BIOS wasn’t set to RAID mode and I didn’t have any arrays defined. I was stuck in a catch-22: turning on RAID mode would render the system unbootable, but I couldn’t install the NVIDIA RAID drivers without having RAID enabled.

I deftly stepped around this problem by installing drivers for the auxiliary Promise SATA controller on the DFI board and booting from that. Then I was able to boot the system into Windows with NVIDIA RAID enabled and install the NVIDIA MediaShield storage drivers. Smart, eh? So I thought.

My next step was to create a RAID 1 array and copy an image of my Windows boot drive on to the array. Then I’d be up and running, with all of the proper storage drivers installed. I created an array in the RAID BIOS, imaged the drive’s contents to the array with Ghost, and attempted to boot.

No go. Operating system not found. The system would not boot from the RAID 1 array, even with the MediaShield drivers installed.

After pulling out several clumps of hair, I went and created a floppy disk with the NVIDIA RAID drivers on it. I used the F6 trick to specify additional storage devices during the Windows install routine, and all was well. I brought some of this on myself, I guess, but I still can’t believe it took three attempts just to install Windows. Building a PC isn’t always easy, even though we’ve made big steps in the past five to ten years.

Overclocking in concert with Cool’n’Quiet
Once the snags with the DFI board and the storage drivers were finally resolved, I was able to install the rest of the device drivers and begin overclocking and tweaking. However, the mission that I’d laid out for myself contained a couple of things that generally work at cross-purposes with one another. AMD’s Cool’n’Quiet technology dynamically scales back the CPU’s clock speed and voltage whenever processor utilization is low. That’s a great thing to have if you’re looking to build a nice, quiet system, but it’s the polar opposite of overclocking, which generally involves locking the CPU clock speed and voltage at higher-than-stock settings. On most Athlon 64 motherboards, you’re forced to choose between Cool’n’Quiet and overclocking, especially if you’re looking to overclock the CPU far enough that it would need extra voltage.

Fortunately, DFI’s LANParty NF4 boards have a way around this problem in their newer BIOSes, and it’s very, very slick. You can specify the amount that you want to overvolt the processor as a percentage, and the motherboard will supply that much extra voltage consistently as Cool’n’Quiet slides the CPU voltage up and down through its range of possible values.

DFI’s BIOS allows one to specify a percentage above
stock CPU voltage for Cool’n’Quiet operation

In order to get my Athlon 64 X2 3800+ running stable at 2.4GHz, I knew that I needed to give the CPU some extra juice, so I set the BIOS to supply 110% of the voltage that it would at stock—in other words, a 10% overvolt. Configured this way on a 240MHz HyperTransport link, the X2 3800+ runs nice and stable at idle and under extreme loads, with a peak clock speed of 2.4GHz. Here are some CPU-Z shots from the box at idle and under load, so you can see how the CPU clock speed and voltage varies.

At idle, the CPU runs at 1.2GHz with lower voltage

Under load, the voltage and clock speed scale up beyond the CPU’s defaults

Fancy, huh? I could probably back off to the BIOS’s next lowest voltage option, 104%, but I haven’t tested that yet.

In my case, I wasn’t interested in overclocking the memory, but the DFI BIOS does provide options on that front. As you probably know, newer Athlon 64s can be set at a lower multiplier, so that it’s possible to run a CPU at a relatively conservative speed on a wildly overclocked HyperTransport link (whose clock also governs memory speeds). For instance, I might choose to run my HT link at 280MHz and lock my X2 3800+ at a multiplier of 8, yielding a 2.24GHz CPU clock and 560MHz memory. Such spectacular feats of memory overclocking haven’t, in the past, been possible in conjunction with Cool’n’Quiet, because C’n’Q will ramp the processor up to its highest possible multiplier as soon as the system’s under load. On my X2 3800+, that would result in a 2.8GHz clock speed and a very nasty crash. The DFI BIOS, however, allows the user to specify a maximum CPU multiplier value for Cool’n’Quiet, neatly solving that problem.

The end result is that I have a system that runs slower than stock when I don’t need any extra performance, with lower voltage and less power consumption; it then runs faster than stock, with higher clock speeds and a little extra voltage, when I need it most. These options may be available in the BIOSes of other Athlon 64 boards, but I haven’t seen them. I think they should become a practical requirement for an enthusiast motherboard’s BIOS.

Stress testing the new box
With all of this overclocking and cutting-edge hardware with brand-spanking-new drivers, I’d pretty much expect to run into some problems. One can always test to make sure that a system’s stable at a basic level, though, so that’s what I did. Here’s a screenshot I took during my over-the-top, uber-stability stress test.

I have running here two instances of Prime95’s torture test, with the CPU affinity for the two instances tied to the X2’s two cores. (Prime95 is a great CPU stress test, and will let you know if it encounters a single computational error.) The high-dynamic-range graphics demo in the corner does a nice job of exercising the GPU, and Winamp is playing some music to get the audio subsystem into the mix. Also running are two instances of the Windows Explorer, one of which is copying a few gigs of data over the network, while the other copies files to a 2GB USB key drive. Meanwhile, my usual mix of background apps is running, including Outlook Express, Trillian, Skype, DynDNS updater, Avast! Antivirus, POPFile, the HP printer/fax/scanner monitor, and Steam. CPU-Z shows that the CPU is running at 2.4GHz, as expected, and Task Manager confirms that CPU utilization is pegged. Up in the top left-hand corner is nTune, monitoring CPU, system, and GPU temperatures, voltages, and clock speeds.

I let the system run like this for at least 18 hours, just to make sure that it could handle it. I would have let it run longer, but temperatures were stable and there were zero problems, so I called it good.

Since then, in over a week constant use, I have had one system crash, a blue screen of death caused by nvtcp.sys. At the time, I was using NVIDIA Ethernet port with ActiveArmor TCP acceleration, but the crash confirmed my fears about the robustness of NVIDIA’s Ethernet implementation. (I’m all for ActiveArmor, but not for ActiveHarmer.) I would recommend looking for a motherboard with a good auxiliary PCI-E NIC onboard for this very reason. I’ve now switched over to this DFI board’s other onboard GigE NIC, driven by a Marvell chip. I’m willing to bet that I won’t have similar problems with this NIC.

Overall, I’d call this build a success, provided that the system doesn’t go up in smoke any time soon. Subjectively, I can tell the difference between this box and my old one in a number of ways. Yes, first and foremost, it’s much smoother in daily use thanks to the presence of a second processor core. I’ve spent a lot of time with SMP over the years, and I didn’t expect anything less. However, I’ve already run into problems with games not doing well on an SMP system, and in that sense, it feels like we haven’t come very far from the dually Celeron days. Surely these problems will be less common as dual-core processors become more prevalent.

This system is also a little bit louder than my previous box, and that’s why I’m considering moving it into a Sonata II case. In fact, I’d already have pulled the trigger on one of these purty Zalman coolers if I weren’t worried about it fitting inside the Sonata II’s ducting. Then again, better cooling won’t fix one source of noise that I somehow failed to anticipate fully: the RAID 1 mirror. Turns out that running two drives in tandem tends to make twice as much noise as a single drive. Who knew? If you have a good, regular backup solution, you may want to avoid RAID 1. That said, I am enjoying the benefits of 250GB of redundant storage.

Also, as fast as this system is—and it’s very quick indeed—my experiences with it so far have only reinforced the degree to which hard disk drive performance is the primary bottleneck in everyday PC use. I can’t help but start dreaming of a 15K-RPM SCSI drive while fiddling during a long reboot or application launch. I suspect the additional noise from the RAID 1 mirror has made me more aware of disk I/O and its delays, but the point is valid, nonetheless. I’d probably pony up for some exotic SCSI drives before going for, say, a second graphics card for SLI or the fastest possible FX or Extreme Edition CPU.