blog the making of the damagebox quad

The making of the Damagebox Quad

I haven’t written about this yet, but I probably should before too much time passes. A couple of months ago, I finally decided that the last iteration of the Damagebox, based on an Athlon 64 X2 3800 overclocked to 2.4GHz, had served me well long enough. It was time to upgrade my main PC.

Now, mind you, this system isn’t exactly the typical enthusiast desktop, since it sits in my office and serves as my main work PC. My willingness to tinker with it is limited by my fear of unproductive downtime. On top of that, I’ve migrated almost all of my gaming activities over to my GPU test rigs, since they’re attached to the gorgeous 30″ LCD that I, ahem, must have for graphics testing. (These things happen, and we must learn to accept them.)

My main PC’s workload became even lighter when I built a storage server that sits in the corner of my office handling file, print and other various services for my network. In the past, my own system served as Damage Labs’ main file and print server, which could become inconvenient when my wife wanted to print a document while I was halfway into installing new video drivers or otherwise tinkering.

So really, my Athlon 64 X2-based system wasn’t terribly burdened or slow at what I was asking it to do—mostly a mix of web surfing, email, IMs, Skype conferencing, MP3 playback, file downloads, web editing, spreadsheets, and image processing, sprawled across dual displays—but it was a little bit on the noisy side. When I built it, I put it into a CoolerMaster WaveMaster-style case that I happened to have on hand, and I told myself I’d move in into a quieter case soon. Well, time passed, and that never happened, even though I’ve had a Sonata II sitting in a box in the corner for most of the life of this system.

Plumber, leaky sink, you know the drill.

In fact, I’d been planning this upgrade long enough to watch some of the hardware I had set aside for the project go out of date. But the fan noise of the old system was grating, and I kind of figured I ought to be using Windows Vista on my main PC just to make myself more familiar with it.

So after rummaging around in Damage Labs for the appropriate parts, deciding which ones I didn’t need for test rigs any longer, I put together a system based on the following components.

  • Intel Core 2 Extreme QX6800 — Hey, there are benefits to being able to raid the parts bin in Damage Labs every so often. The QX6800 has been superceded by 65nm chips with faster bus speeds and by 45nm chips with faster, well, everything. Still, my testing showed that this 2.93GHz quad Core 2 processor isn’t limited by its 1066MHz bus in the vast majority of apps, and, heh, it’s a 2.93GHz quad Core 2 processor. This puppy does have an unlocked multiplier, but I haven’t taken advantage of it yet—and I may not. Acoustics are my big concern, and higher-bin CPUs typically don’t overclock especially well.
  • Gigabyte P35-DQ6 — After using the DDR3 version of this board on my CPU test bench, I went out of my way to have Geoff ship me this board from Canada. The P35 chipset doesn’t have PCIe 2.0, but it’s otherwise current. And really I like what Gigabyte has done with its newer motherboards. The BIOS tweaking page is full-featured and (relatively speaking, at least) easy to understand, and the board gets by with only passive cooling, which was a big plus for this build.

    I briefly considered using an XFX nForce 680i SLI mobo instead, but I was concerned about its relatively high power consumption (which turns into heat and thus noise) and support for 45nm processors. The Gigabyte P35 seemed like a better choice overall.

  • Visiontek Radeon HD 3850 512MB — This one may surprise those of you expecting to read about, in the Damagebox, a killer new PC. But hear me out. I had originally set aside for use in this system a GeForce 8800 GTS 640MB graphics card, with the expectation that I’d have a more-than-adequate video solution for occasional gaming (when I host small LAN parties for the guys in Damage Labs) that would also be nice and quiet most of the time. I even installed a GTS 640MB card in the box during the initial build.

    However, a thought struck me, and the logic of it was inescapable. Based on my own testing, the Radeon HD 3850 512MB offered nearly equivalent 3D performance with markedly lower power consumption, especially at idle, where this card would spend most of its time. I had on hand a Visiontek HD 3850 512MB card I’d ordered for use in a review and never actually needed, complete with a higher-than-stock clock and a very nice dual-slot cooler that ejects hot air from the enclosure. The HD 3850 GPU has HD video decode assist capabilities that the GeForce lacks, and on the P35 chipset, it can run in a dual-card config, which Nvidia prevents GeForces from doing. In what way was the HD 3850 not a better choice?

    Also, I have to confess to having a soft spot for the Radeon HD 3850 512MB. AMD produced one of the best video card values in years, yet it was largely overshadowed by the GeForce 9600 GT. In any other year, a card like the 3850 512MB would have been the hottest thing going.

  • Dual WD Raptor 150GB drives — The new Raptors weren’t out yet when I built this system, so these drives were the fastest SATA hard drives on the market than to their 10K-RPM rotational speed. Performance is definitely snappy. Their relatively low 150GB capacity wasn’t a big concern because my storage server holds most of my data, anyhow. I went with RAID 1 for the obvious reason: if one of the drives fails, I’m not out of commission. Not only that, but RAID 1 is simple; the recovery process isn’t dependent on a RAID controller understanding much of anything about the array in order to rebuild it. I don’t really trust motherboard-class RAID 5 or the like not to botch everything if I happen to mix up the SATA plugs after working inside the PC, let alone in the event of a drive failure.

    However, my storage setup is the source of one of my biggest complaints about this build: Raptors are frickin’ loud on seek, and Vista keeps the hard drive busy almost constantly, for reasons I mostly understand but still can’t quite accept. The acoustic problem is multiplied by the fact that the two Raptors seek together in RAID 1. It’s pretty bad—especially, heh, with almost no fan noise to drown it out. I’ve told myself I’ll buy a pair of newer, relatively quick 7200-RPM drives to replace the Raptors eventually.

  • Antec Sonata II — Yep, it was time to pull the Sonata II out of the corner. My first act, after unboxing it, was to remove the power supply, since it was old enough not to have the requisite complement of power connectors for my mobo and only a single PCIe six-pin aux lead. After that, the build was a snap, since the Sonata is simply a well-designed case with a smart layout. Of course, it’s also pretty darned quiet, with its single, 120mm exhaust fan. Without using the default Antec power supply, I couldn’t get PSU-based speed control for that fan, but the big fan comes with a three-way manual speed switch. I was able to leave it on its lowest setting at the end of the day, which is slow and stealthy.
  • CoolerMaster 850W PSU — Ok, so I’m writing this on an airplane, and I don’t recall the exact model PSU I used. It’s an 850 CoolerMaster with something like a 130mm single fan in it and enough power connectors and capacity to support a quad-core CPU, dual graphics cards, and a full complement of drives in the Sonata’s bays. Perhaps it’s overkill, but I had it on hand and wasn’t using it otherwise. And it’s blessedly quiet.

    Initially, I installed another PSU with a big fan, an OCZ GameXStream 700W. I’ve used these in my test systems for ages, and they’ve quieted down the confines of Damage Labs considerably. However, I was surprised to learn that once it was installed in this case and powering everything, the GameXStream tended to crank up its fan speed quite a bit—even at idle. That simply wouldn’t do, and swapping it out in favor of the CoolerMaster bought me a nice reduction in overall fan noise.

  • Zalman CNPS-9500 LED — If you’re getting the idea that achieving the best acoustics in this box was an iterative process, good for you. The Zalman was perhaps the biggest surprise of this whole build, even though I’ve known for ages that these CPU coolers are excellent. Initially, I decided to go with a stock Intel cooler, for several reasons: because I thought it would be sufficiently quiet at idle, because it had a four-pin connector for linear fan speed control, and because it would fit inside of the Sonata II’s funky cooling shroud thingamajig.

    Turns out, based on my casual evaluation, that shroud is pretty much no help, either with system temperatures or acoustics. I soon gave up on it, and it’s no surprise Antec left it out of the Sonata III.

    The stock Intel cooler wasn’t horrible, but it was loud enough to bug me once I had the PSU fan under control. The QX6800 is a 130W processor, and I considered dropping down to a Q6600. However, my major concern was idle noise, and SpeedStep would remove any difference between the QX6800 and Q6600 at idle. So I decided to try the Zalman.

    This meant tearing the motherboard out of the now the fully built system. The Zalman’s mounting bracket must be installed on the underside of the board, so there was no way around it. On top of that, Gigabyte mounts a big hunk of copper below the CPU socket on some of its DQ6 boards, and that gets in the way of the Zalman’s bracket. I had to remove it, which is possible, but still a pain.

    Gigabyte made it up to me by providing near-perfect automatic speed control for my older Zalman 9500’s three-pin fan. I test cooling and acoustics on most new builds using the combination of a multithreaded Prime95 torture test, a windowed 3D graphics demo, and whatever temperature monitoring software works with the mobo. Happily, Gigabyte’s smart fan control kept the Zalman spinning at very low RPMs at idle and only ramped up incrementally as needed under load, with no obvious transitions from one speed to the next and—blessedly—no tendency to flail back and forth between speed levels.

    Not that I could really hear the Zalman’s fan, anyhow. The thing just wasn’t working very hard. With the stock cooler, CPU temps were flirting with 80°C with all four cores fully loaded, nearly into thermal throttling territory. With the Zalman, temps dropped to between 60 and 65° max, even with relatively low fan speeds. At idle, it was less than a whisper. This was a nice reminder about the worth of a good aftermarket CPU cooler, both for temperatures and for acoustics.

  • 4GB of Crucial DDR2-800 memory — With a 64-bit OS, I can actually use all of this memory, and I do need it for editing multiple eight-mega pixel images alongside all of the other multitasking I do—especially, heh, with Vista’s memory footprint. The RAM itself is decent memory from a quality vendor, but nothing particularly special. I don’t see the point in paying extra for slightly lower latencies or higher clock speeds.
  • Auzentech X-Meridian — I had to replace my trusty Revolution 7.1 sound card because it didn’t have drivers for 64-bit Vista, and frankly, I was ready to be free of M-Audio’s lousy driver support, anyhow. Before this C-Media-based replacement arrived, I used the DQ6’s onboard audio for a while, and the sound card was a noticeable improvement. This card produces clear, crisp audio with no annoying overemphasis on highs, lows, or mids. If anything, it’s an improvement over the Revo, although I’d have to do a back-to-back comparison to say how much. The drivers are fairly lightweight, unobtrusive, and seem to work fine.

And that’s about it. I installed some model of dual-layer SATA DVD writer, of course, and a 273-in-1 flash reader. The system is attached to the same keyboard, mouse, dual 20″ Dell LCDs, and APC UPS that I’ve had forever. I may add a TV tuner card and a remote for use with Vista Media Center, but I haven’t done that yet.

The end result brought a huge reduction in fan noise, but the constant seeking of the Raptors kind of spoils the effect. There’s more work to be done, I suppose. Overall, though, I’m still fairly pleased with the results.

I guess I should say something about the performance gains, but honestly, they’re not blowing me away. Yes, this system is snappier than my Athlon 64 X2, but as I said, I didn’t have many complaints about performance before. The X2-based system rarely seemed slow, except when booting, paging from disk when memory got full, or launching a new application. Occasionally, rarely, both CPUs would get occupied when multitasking or when a program went sideways. The new PC seems quicker in all of those cases.

As I’ve mentioned before, though, I’ve run into serious problems with Vista network file sharing performance. This is far from ideal, since most of my data resides on a separate box. I’ve spent quite a bit of time looking into the issue since my last blog post on the subject, and I still haven’t found a fix. I’m not sure whether to blame Realtek’s drivers, Microsoft’s Vista SP1 update, or a combination of the two for the problems, but I’m currently leaning toward Realtek. Some of their more recent driver drops have had changelogs that mention fixes for some scary problems. Unfortunately, they haven’t resolved the problem causing my PC to drop connectivity to the file server yet. In all likelihood, my next step will be moving to a discrete NIC to see if that helps.

Beyond that, well, I’ve had almost no problems with application compatibility or system stability in Vista x64.  Of course, I did some planning ahead of time to make sure that my hardware was compatible, including replacing the Revo 7.1 sound card and moving my HP DeskJet all-in-one over to a WinXP-based print server to ensure full support for all of its capabilities.  These were reasonable accommodations to make for older hardware, though, and I had planned to make those changes during my next upgrade, regardless.