I had a similar issue with a s370 dual socket board.
I had to disable completely the integrated graphics before being able to use a PCI GPU.

The option was not available with the HP OEM bios, but the board was an Asus one with customized HP bios. I flashed the Asus bios from their website and had access to much more options, including disabling the integrated ATI rage.

Good luck

XCVG wrote on 2023-07-08, 05:25:

So, after years of paring down my collection and generally staying out of retrocomputing in general, I suddenly got the insatiab […]
Show full quote

So, after years of paring down my collection and generally staying out of retrocomputing in general, I suddenly got the insatiable urge to build a computer.

This build would have a story of sorts. A legend, if you will. Lore. Kayfabe. The elevator pitch is to take an old server/workstation and turn it into a gaming/content creation PC, like some people do today, but back in 2006 or so. You can't swing a hot new Athlon 64 FX or Pentium Extreme Edition, wait for or afford Conroe, but you can pick up this old server with a couple Xeons in it and throw a graphics card in it. I've always thought dual processors was super cool, and I'm still miffed that my Athlon XP Shuttle XPC can't run Premiere Pro 2.0, so this sounded like a fun build idea.

I did find an actual decommissioned server that was the right age, but it was just too big for my real actual in 2023 space so I pulled the 15K SCSI drives out of it and passed it along.

Multiprocessor Xeon systems seem to be all the rage right now and I feel kinda like I'm jumping on a bandwagon. I had this idea months ago, but procrastinated a bit and had to wait for parts to arrive and now it looks like I'm following the leader... story of my life. Oh well, the parts are here now so let's build!

plumas_gpu.jpg

The first part to actually show up was the graphics card, a Radeon X1300 PCI. This seems about right for the era (2005-2006), wouldn't have been too expensive and has a kind of "I went to the computer store and they sold me something they promised was good for gaming, for way more than it was worth" vibe to it. And it was relatively cheap. And, hopefully, the PEX 8111 bridge on it can run on a 66MHz PCI bus so it won't suck as much. One of the constraints I'm giving myself with this build is to avoid using an AGP graphics card and instead limit myself to server-oriented boards with only PCI(-X).

You want to see the motherboard? Just a second.

I also bought a pair of 2.4GHz Prestonia Xeons for $4 each, shipped. These aren't the best CPUs for the board I chose, which will take faster Prestonias and maybe the Gallatin MP with extra cache, but hey, they were cheap. And I feel they're more "realistic" for this build. From what I've read, the Gallatins with extra cache weren't that much better. They're tres cool, and cheap today, but the hypothetical me in this scenario would have probably stuck with Prestonias. Probably the ones that came in the server that fell off the back of a truck.

Okay, fine, here's the motherboard. A Supermicro P4DP6. I wish it was an X5DP8, which supports 533MHz FSB Gallatins, has Ultra320 SCSI and gigabit ethernet onboard, but I got a deal on the P4DP6. And I think it fits the idea of a constrained, "picked up a server that fell off the back of a truck", build. Not something done by picking perfect parts, with the benefit of 20/20 hindsight, but just picking up what was available and what you could afford. Not the perfectly tweaked 440BX-with-1.4GHz-Tualatin we're building today (okay, mine's an Apollo 133A with a 1GHz Coppermine, but it's just about perfect for me) but the Celeron on a "BXPro" we actually had.

plumas board.jpg

Unfortunately, while the seller included the board, a CPU, a generous 4GB of RAM, and two (different) coolers, they didn't include all the mounting hardware for the coolers. In fairness, that would probably include an entire server chassis, as I'm pretty sure these originally bolted through the board right into the case. So all I could do was gaze longingly at this beautiful piece of computing machinery until I could make the pilgrimage to either Orange Coloured Chain Building Store or Tire Store That Also Sells TVs And Guns.

I did think about just plopping the heatsink on top and crossing my fingers, but with NetBurst it just didn't feel like a good idea.

So I waited and went to the store and came home with a bag of nuts and washers that would go with the screws they included. I bolted the mounting frames onto the board, which is probably not how it's supposed to be done, but whatever, it's fine. Unfortunately I don't have a good picture of it, but basically I just used nuts on the back of the motherboard where the chassis or a backplate would be. Then I squirted some MX-4 on the CPU, clamped one of the heatsinks down, wired everything up and dropped my iFixit Driver (not sponsored) across what I was 90% sure was the power switch header.

plumas fail.jpg

And... nothing. No POST, no beeps, not even a fan spin. Dead as a doornail.

If you look closely, you'll see that I took out all but one of the memory sticks to keep things simple for testing. This will become important later.

I read the manual. Twiddled some jumpers. Read the manual again. Confirmed that yes, that is the power switch header. Read the manual a third time. Tried clearing CMOS. Scratched my head. Dug out my power supply tester to verify that my power supply actually worked. It did. Dug out my post card, got worried it would fry the 3.3V-only PCI-X slots, put it back. Took the RAM out, put the RAM back in. Took the CPU out, put the CPU back in. Went back to the manual, found the troubleshooting steps...

The battery on your motherboard may be old. Check to verify that it still supplies ~3VDC. If it does not, replace it with a new one.

Surely this isn't one of those boards that requires a working battery even to power on, right? Only Apple ever made those, right?

plumas booted.jpg

It is indeed one of those boards that requires a working battery even to power on.

I'm actually misrepresenting things a bit with that picture, though. I was able to power the board up once I changed the battery, but it didn't actually get anywhere close to POSTing. The fans spun. The lights came on. No video. No beeps.

I take the RAM out. It beeps angrily at me. Okay, that's expected. I put the RAM back in. It stops beeping, but doesn't boot. Hold on, it doesn't need RAM installed in pairs, does it? Three guesses as to the answer, and the first two don't count. I checked the manual, cursed, put a second stick of RAM back in, and then it booted. For some reason, I misremembered the E7500 chipset as having strictly single-channel memory.

I still feel justified in blaming Supermicro for this, though. No RAM, beep like the building is on fire. Incorrect RAM, stay silent as a mouse.

I did a little celebratory dance, then I had to run off. When I came back, I set to work installing the second CPU. Because what's the point of a dual Xeon rig without dual Xeons, amirite? I actually swapped both CPUs, because I had that matched pair. The CPU in the board was the same speed, but a different stepping. The socket looks a little smashed and I had to finagle the CPU in a bit, which was worrying. The first heatsink attached with clips, but the second one just had bolts. There was something vaguely resembling a backplate in the package, but I couldn't figure out how it was actually supposed to attach. So, nuts again it was. The spacers on the heatsink seemed too thick, the bolts too short, and I didn't think it would provide enough mounting pressure to keep the CPU from frying itself...

plumas dual.jpg

...but it seems to be fine, at least at idle. That was... actually pretty easy.

Okay, board is booting, dual CPUs are working. There was one more thing I wanted to try before declaring the base platforms ready to go and the project totally viable. I threw the Radeon X1300 PCI into the first PCI-X slot, which is an appropriate brown colour, and powered the system on.

The fans spun, the lights came on, and it didn't give me any video.

To be honest, I didn't really expect that to work. I reconfigured the jumpers to force the slot into 66MHz PCI mode and tried again. That didn't work either. Uh oh. There was much cursing, swapping of jumpers (including the one on the card itself which I have no documentation for), switching back and forth to the integrated video, swapping slots, blindly changing BIOS settings, and cursing. I was able to get a little further after changing some PCI slot settings. Still no video, but it beeped out an error code indicating "video configuration error". The only documentation I was able to find is for another board and tells me to change a BIOS setting that doesn't exist on this one.

I forget how, but I was able to boot it once with the integrated video enabled and the X1300 installed one time, at which point it promptly threw up a bunch of errors about a resource conflict and gave up midway through post.

I did try my only other PCI graphics card, an eBay special Rage XL, in this board, but if it hated the Radeon it loathed the Rage. Which is ironic, because the integrated GPU is a Rage XL. I think I maybe got the board to beep at me once, but it wouldn't post with the Rage in, even when switched to integrated graphics, and mostly it just sat there and did nothing no matter what I tried.

So, at this point, I'm kinda stuck. I'd love to talk about my plans to modernize a side of the road special edition case by bolting the front on upside down, or this cool soundcard I have that has firewire on it, or my myriad of storage options, but I'm kinda blocked on getting a graphics card to work for the build to work at all. There's one thing I can try without spending a lot of time or money, which is updating the BIOS. I mean, if it goes wrong, Supermicro's advice in this case is to contact their RMA department for a replacement BIOS chip. I'm not sure if they'll even respond to the email for a board this old.

The more expensive and time consuming options I can think of are:

• Try modifying the BIOS. I've heard this is easier than you'd think. I don't actually know what's going wrong, though.
• Get another PCI graphics card. Both of the ones I have are potentially iffy for compatibility. There's no guarantee that's the issue, though, so I could end up with another card I have no use for.
• Give up on this motherboard and get something else similar. Maybe a Tyan S2720. I've heard Tyan is less picky. But it's not as nice a board overall- in particular it only has one PCI-X bridge.
• Give up on the idea of PCI-X only and get something with AGP. Maybe an IBM x225 motherboard, which seems to be a rebadged MSI E7505 Master-LS2

At this point, though, I'm kinda putting it aside to think about where to go next. And, of course, solicit ideas from people who might know a thing or two I don't.

PCI cards can work - https://www.supermicro.com/support/faqs/faq.cfm?faq=2176 - so maybe something about the X1300 it doesn't like (did you try it at 33MHz in any slot?)

Board manual & BIOS updates can be found on TRW downloads link at https://theretroweb.com/motherboards/s/supermicro-p4dp6

PC Hoarder Patrol wrote on 2023-07-08, 12:08:

PCI cards can work - https://www.supermicro.com/support/faqs/faq.cfm?faq=2176 - so maybe something about the X1300 it doesn't like (did you try it at 33MHz in any slot?)

Board manual & BIOS updates can be found on TRW downloads link at https://theretroweb.com/motherboards/s/supermicro-p4dp6

This card was flakey when I tested it in my Shuttle XPC, too, so I am suspecting the card is picky about compatibility, or even marginally functional in general. The Rage XL I was trying is one of those cheap eBay ones with known compatibility issues so I'm not too surprised it didn't work at all. I don't want to spend a lot on this project, but I'll keep an eye out for another PCI graphics card and see if something comes up for cheap.

I did grab the manual and an updated BIOS already, I just haven't had a chance to flash the latter yet. The manual was helpful, but it would have been more helpful if I'd actually read it from the beginning 😅

After weeks of procrastination addressing other matters of utmost importance, I finally got around to updating the BIOS. For some reason, the flash tool didn't want to run from the boot disk I created in XP, or my usual DOS boot disk, complaining about not being able to run on top of a memory manager. I ended up using an (original!) DOS 6.22 install disk. I also used the floppy drive and too-short cable I pulled out of one of the servers, because it was the closest to hand.

A tense few minutes later and the deed was done. To be honest, I didn't really expect this to make a difference. After verifying that it at least didn't make things worse and that the system booted up on integrated graphics, I shut it off, installed the X1300, and powered it up again.

To my great surprise, it booted right up with no complaint. I have no idea if it's running at 66MHz, I have no idea if it's stable, or even if it'll boot up reliably, but this is a lot further than I got before. For the time being, I'm calling this a success.

So, what's next? I think I'm really going to start looking for a case. I've seen a lot of candidates comes up locally, but I've been too lazy to pick one of them up none of them sparked joy. I'm probably going to go with a generic beige or black box, but I haven't ruled out something a bit flashier if one comes up locally for cheap. I don't want to use a rare or desirable case, because I'm probably going to have to bash the drive cages out with a sledgehammer to fit this too-big motherboard in.

I'm going to go for something with the older style layout of the power supply on top, which should give me enough room for mounting that salvaged floppy drive and an optical drive of some description above the motherboard. For storage, I think I'm going to go with two of the 36GB 15K SCSI drives, plus a big IDE drive that I don't have yet. As for mounting the hard drives, I'll cross that bridge when I come to it.

I keep thinking "it would be nice to have" various things with this build. It would be nice to have a motherboard with Ultra320 SCSI, it would be nice to have RAID, it would be nice to have 72GB drives instead of 36GB ones... but that's missing the point. I don't really want to spend a lot of money on it, and the whole "story" of the build is "mid-2000s nerd hunts for bargains". Yes, it'll be compromised, but that's the whole part. Both behind and in front of the curtain, it's a build made out of the parts that were available, not the best parts for the job.

Still, I have to slap myself sometimes and go "bad XCVG" when I start thinking "but what if..."

I do have two PCI-X RAID cards out of those old servers, though. I have no idea if they'll even work outside of a Compaq motherboard but I might try to get one working.

It's been a while (almost two months exactly), but I've made some progress. I have a case now!

This beige box showed up at the thrift store one day and I snapped it up. The computer in the case was mildly interesting, had one of those oddball ECS boards with a soldered-on Duron, and surprisingly it POSTed (though the hard drive was dead). But that's a story for another day. I just needed an old case with the old style layout; power supply up top. There was no way the new motherboard was going to fit in a normal ATX case, at least with the drive cages in the way.

So I started cutting...

If I was smart, I would have left more of the drive cage intact, but I wasn't and I'm not. Oh well. Like I guesstimated, there was still enough left to mount the optical drive (which I think came out of a Power Mac) and floppy drive in weird 3/4 height adapter salvaged from one of the servers. The drives actually stiffen up what's left of the drive cage and it's a lot more stable than I'd thought.

I have a vague plan for mounting the hard drives. The original hard drive cage is basically two pieces of sheet metal riveted to the optical drive cage at the top and the floor of the case at the bottom. I just cut through it at the top, but I carefully (ish) drilled out the rivets on the bottom. I mean, I still bent them pretty badly getting them out, but it'll buff out. Interestingly, there's a second pair of holes in the bottom of the case to mount the drive cage about an inch over, away from the motherboard. That should give me enough clearance as long as I shorten the cage to clear the bottom CPU cooler. I was thinking of bending the top pieces over and attaching them together somehow, to make a sort of u-shaped drive cage, but I might be able to get away without that and just rely on the drives for stiffness.

In the meantime, though, I wired up everything else and tested it out. I even threw in the cards I'm planning on using, to get them out of the way if nothing else. There's some good news and some bad news:
- It still works!
- The cards mostly play nice together. Mostly. At the very least, none of them have let out the magic smoke. Okay, I'm reaching a bit here.
- The video card continues to work, however, no matter what I do it refuses to go above 33MHz PCI. This is disappointing, but not terribly surprising.
- It does not like the Smart Array cards I pulled out of the servers at all. The machine does POST with them installed, runs into conflicts if you don't disable the onboard SCSI which is expected, but hangs just after initializing the option ROM. This is also disappointing, but not surprising.
- I still get intermittent bus/address conflict errors. I have no idea why. Disabling the onboard networking (I have a PCI gigabit card installed) might help.

What's next? Probably the hard drive cage, and hopefully less than two months from now. I'm still not sure what the final drive configuration is going to be. I know I want to boot off one of the 15K SCSI drives, but I might give up on RAID and just run a single drive, maybe with another one of those as an apps/games drive. I still need a bulk storage drive, I still don't have one, and I'm not sure if I can fit three drives in my cut-down cage. I think I have a 320GB SATA drive somewhere, so I might go with that, either on a SATA card or SATA to IDE adapter. I want to put something snarky here but honestly there are just still a lot of unknowns here.

I've started thinking about the OS a bit, too. At first I was thinking XP, because it's about the right period and the hardware is a good fit for it. But then I started thinking, well, this is turning into a cursed build, really janky and weird and different, and that just seems too much like the safe option. There was an exciting new operating system that came out in 2007 that a brave soul might try running on their bodged-together "workstation"...

It hasn't been two months yet, but it's still long overdue for an update.

Modifying the drive cage was pretty quick and easy. I cut it down, bolted the pieces into the case through the existing holes, shoved a drive in and called it a day. That was done weeks ago.

Then I tried and failed to get the machine to actually work with that drive.

The Apple-branded carrier-loading DVD-RAM drive (chosen specifically for its weirdness) refused to boot my Windows Vista DVD. In fact, it refused to boot pretty much anything. I gave up on the Apple-branded carrier-loading DVD-RAM drive and Windows Vista and threw in a boring but known good CD burner and Windows XP disc, at that point giving up on doing anything fancy and just hoping something would work. After several reboots, I was able to get the installer to boot and went through the first steps. No matter what I did, though, I couldn't get to the next step where it's supposed to boot from the drive.

After spinning my wheels poking at the BIOS and jumpers I realized I don't actually know if this motherboard can boot from a SCSI drive. I kind of assumed it could, but there doesn't seem to be any option ROM which I'd expect to see. I gave up on the onboard SCSI and ordered an Adaptec RAID card.

Weeks later, said card arrived, I shoved it in awkwardly (the SCSI cable doesn't really fit in the space between the slots but I couldn't be bothered to rearrange the cards) and it completely failed to do nothing. Some cursing later, I realized the card wasn't in all the way, and after a bit of shoving it was in and I was booting into the Adaptec utility. It sat there spinning for ages as the drive made ominous chirping noises.

At some point, the first hard drive died completely. I don't know if it was already on the verge of failing or if it really didn't like running with no cooling and/or only three screws into a shaky mount. Fortunately, I have a stack of them, and all but one of the others seemed to work. Although I went out of my way to pick up a RAID card, I've pretty much given up on the idea of RAID at this point. I feel putting two 15K SCSI drives in is tempting fate and they might just cook each other or shake each other to pieces. I'm not sure if there's enough clearance for two drives with SCA adapters anyway.

I did find a 160GB IDE drive which I think would fit well as a storage drive. I haven't installed it yet, because I'm still working out cooling for the drives (very necessary for 15K SCSI drives I think) and I might have to remove the cage for that. I haven't installed an operating system either yet, because I don't want to tempt fate with the uncooled SCSI drive.

I'm probably going to take the lazy way out and just mount a pair of slim 80mm fans on the provided mounting holes. I've thought about trying to hack out the metal and cram in a 120mm fan or hack out the bezel and jam a thick 80mm or 92mm fan in the front, but I'm not really feeling it. I also had plans to integrate a USB hub connected to the USB card to provide USB 2.0 to the front panel, but I'm not really feeling that either. To be honest, I think almost all my initial excitement for this project has faded at this point, largely (but not entirely) because of the technical issues I keep running into. One month later, three months after starting the project, and I'm closer to having a working system but still not quite there.

It's been over a month, and it's time for another update. Have things gone well? It's more like things have, well, gone.

I decided to go with the laziest possible solution for the front fans and ordered a pair of 80mm slim fans. It was something like $10 on Amazon for both.

They don't really cool the drives. Maybe the top one, a little bit. I'm sure it will be fine.

I tried to install Vista from DVD shortly after the previous post, but it failed early in the process. I suspect it was a bad burn, but it took me a while to get around to trying again. This time I put it on a USB drive and used Plopboot to boot from the drive.

Hey, now we're getting somewhere! This looks like a perfectly cromulent Vista install.

Unfortunately, it's completely unusable.

The installer ran fine, but problems started to crop up in OOBE (first boot) before we even got to the desktop. It was really, really, painfully slow, with UI that took seconds to respond. Setting up the first user and date/time was excruciating. Then it got stuck on "please wait while we check your computer's performance" even after I let it sit for half an hour. I think it was technically running, just running super slowly. I looked up a sketchy YouTube video on how to skip OOBE in Vista and slogged my way through a horrifically unresponsive regedit.

Things did not get better at the desktop.

Not only do we get some interesting graphics artifacts...

...it also runs slower than maple syrup in a Montreal winter. We're talking several seconds for a screen redraw. Open the start menu, move a window, close something, doesn't matter.

I tried installing the graphics driver. It made things even worse. At least we got to see some advertisements for circa-2010 games in the installer while we waited.

I rebooted it once more just to be sure, then gave up and shut the system down.

So, Vista is a complete bust. I have absolutely no idea what is going wrong. It seems to be an issue with the graphics- disk access is okay and CPU load isn't that high- so maybe the GPU, one of the many buses between the CPU and the GPU, the graphics driver, the chipset driver, PCI configuration, yeah that doesn't really narrow things down that much.

The scope of this project has changed a bit, or at least firmed up since the initial kickoff. The driving question has always been "was building a workstation/gaming pc out of an old server a good idea in 2006" and I think the answer to that is a pretty resounding no. Even if performance was acceptable, this behemoth is way too hot, loud, heavy, and problematic to be worth it. I do, however, still want to get an idea of what that performance would be.

I'm going to try XP, which hopefully will work a little better and does put the machine on a level playing field with my other retro machines. I'll run some benchmarks, try some games, and then relegate this abomination to the dustbin of history.

One last update before the conclusion. I'm not sure when I'm going to get to writing that, but at this point I'm done with the actual build so I wanted to post an update on the last few things I did with it.

Vista didn't work, but I was able to get XP installed and run some benchmarks!

Well, sort of.

I may have cheated a little bit.

I tried installing XP right after the failures with Vista. The first part of the install went fine, but it bluescreened after rebooting into the second step of the installer. I tried a different disc and got the exact same thing. I have no idea what the problem was, if it's an issue with the RAID controller or something else in the hardware. As I mentioned in the last post, my attitude toward this project has definitely shifted towards just getting some benchmarks so I can answer the originally posed questions.

So I took the nuclear option, ripped out all the cards except the graphics card and USB card, and installed XP to the IDE hard drive. It's a much less interesting machine now, but it did work. There's no sound or ethernet (I forgot to put the jumper back in to enable the onboard) but I don't need those to run benchmarks.

I don't have a lot of other machines anymore, but I do have two that should make decent points of comparison. I have a Shuttle mini-PC with a late Athlon XP and a Dell laptop with a Core 2 Duo and dedicated graphics. I pulled both of those out and ran the same benchmarks on all three machines.

I don't want to spoil the conclusion too much, but I've had a hunch for a while that this wouldn't really be worth it, even against period machines, and at a glance I think I was right. I do still need to go over the data, try to remember what the landscape was like in 2006, and try to answer the questions of whether this was a good idea back then and whether building a machine like this is a good idea today.

EDIT: Typos.

This was a good read and a pretty cool experiment.

Something that stood out to me, while not having much to do with your actual build, is that you mention an ECS motherboard with a soldered-on Duron. I actually ended up picking up one of those a couple weeks ago in a mixed lot of old PCs, and I don't think I'd ever experienced something quite like it before. It is a K7SOM+ with a Duron 1.2 - which seems to have been marketed by ECS as an "1800+" even though AMD never actually made anything like a "Duron 1800+". Definitely an oddball board, which I could not get even remotely stable in Windows 98, but which seems to run perfectly fine under XP. It is bare bones and the BIOS is mostly locked down. A weird piece of kit that really has no use to me, but might just stay in my collection as a novelty.

Alright, conclusion time! This is going to be a long one, long enough to break up into sections.

How did it do?

I ran a few synthetics: Cinebench R10, 3Dmark06, and PCMark05. Unfortunately, the PCMark05 System benchmark didn’t complete but the sub-scores are interesting and we’ll take a look at those in a moment.
I ran the same benchmarks against two systems for comparison. One is a Shuttle SN45G with an Athlon XP 3000+ and ATI Radeon 9600 Pro. The other is a Dell Precision M4400 laptop with a Core 2 Duo T9600 and Quadro FX 770M (similar to a 9600M GT or desktop 9500 GT). I wish I had a later P4 system and an early Core 2 (Conroe) machine, but I don’t anymore. I did guesstimate some CPU scores for an early Conroe system based on the M4400 benchmarks and the difference between a T9600 and an E6300.

With two CPUs and hyper-threading, performance scaling in Cinebench from single-thread to multi-thread is impressive. In single-thread Cinebench, it’s at the bottom of the chart. It does beat the Athlon XP by a good margin in PCMark and a small margin in 3DMark. Theoretically, though, even the E6300 should handily beat it, and the T9600 with higher clocks and more cache obliterates it.

This confirms what I already suspected. The real surprise was how close the Athlon XP got in the mixed benchmarks. Maybe it shouldn’t have been, though; back then single-thread was king and probably weighted heavier in tests, and the Xeon is basically a 2.4GHz Northwood P4 on a 400MHz bus, which the Athlon XP handily beat.

To be honest, this data isn’t very useful. I suspect the PCI bus is really hurting this system in the Cinebench OpenGL test since it loses that one but beats the Shuttle with its 9600 in 3DMark,but there are too many variables to safely draw that conclusion. I will say, though, wow, that mobile Quadro is fast. I knew it would beat the older cards, but I’m pretty surprised at how wide that margin is.

I don’t have an AGP or PCIe version of the same card to do a direct comparison, but Techspot reviewed the card back in 2006 and compared its performance with a PCIe version of the X1300 Pro. You can find the review here, but in short, the PCI version gets about half the performance. Ouch.

I would like to highlight two of the PCMark sub-scores that illustrate that PCI bottleneck. In 2D Transparent Windows, this system scored 93.283 windows/s. The Shuttle was four times faster at 396.210 windows/s. In 2D Graphics Memory, 64 lines, it scored 138.453 FPS, which sounds fine, until you look at the Shuttle’s result of 692.165 FPS.

I haven’t run a PCI graphics card in a very long time, but I did run an eGPU setup with limited PCIe bandwidth for a while. With that kind of bottleneck, you’re really in uncharted territory and performance doesn’t scale like you’d expect. In some workloads, it didn’t matter, in others, it was crippling. I remember that desktop composition was smooth, but Photoshop never ran well. There were games where I could run 4K Ultra at 30 FPS easily but couldn’t hit 60 no matter what settings I changed. It’s not as simple as “divide expected performance in half”. Unfortunately, I think a lot of the kind of professional workloads you might have wanted to run on a dual-CPU machine in the mid-2000s are going to fall into the problematic category.

I only tried a few games, unfortunately most of my collection is either too old or too new and I was frustrated even further by installation issues. I gave up on getting benchmarks pretty quickly but still wanted to get an idea of what gaming on it would actually be like. I played a round of Quake III Arena, which was mostly fine with a few slowdowns. That game came out in 1999, so that’s almost damning by faint praise. The only other game I both had and was able to get working was Crysis. It technically runs Crysis, but it wasn’t playable. There were graphical glitches and the framerate was terrible.

In summary, this machine is a lot like a weedy little four-pot in a cheap compact car. It makes a lot of noise, but doesn’t have a lot of go.

Did it make sense then?

Almost certainly not. It depends what you’re upgrading from and how cheap you can get a used server- and unfortunately I have no idea what the used market would have been like in 2006- but in most cases you’d be better off staying with what you have, upgrading your current rig, or saving up for something new.

If you have a late Athlon XP system or mid-late P4, the choice is easy. Dual Xeons look cool, but it’s not going to get you more single-thread performance than what you have, and in 2006 that’s what matters the most. That’s not to say there aren’t applications where this machine would be faster, but for a gamer or creator-who-games, it’s not worth the tradeoffs in single-thread and graphics performance. That will change in the coming years, but this 32-bit, PCI-X system is hardly future proof. , maybe upgrading your graphics card or adding some RAM to tide you over.

If you have an older system that’s significantly slower than this one, even in single thread- say an early P4, or an original Athlon- then it depends how cheap you can get this old server. You’ll need a more expensive (and worse-performing) PCI graphics card, and you’ll have to deal with a lot of weirdness, but it’s still faster than your current machine. If your old system didn’t have AGP anyway, then you’re not losing that either. If you can’t get the server cheap, though, you’re probably better off either buying a different used system or holding out just a little bit longer.

Does it make sense now?

The short answer is that you already know the answer. Either you’re already building something similar, or you’re running far away very fast.

In my mind I group retro builds into three broad, somewhat overlapping categories. Practical generalist, practical specialist, and silly fun.

As a practical generalist machine, there are better options. If you just want something to play XP games, your best option is the same one our hypothetical buyer would be looking at in 2006: Conroe. Although they’re slowly ticking upward in rarity and value, you can still get a Core 2 machine for dirt cheap. Throw something like an 8800GT in it, maybe an SSD, and you’re good to go. If you’re looking for something period-correct to a specific era, or maximally powerful, you might want to go a different route, but probably not this one.

As a practical specialist machine, I have a hard time imagining what niche this would fit into. It doesn’t make sense as a gaming machine. In its original configuration, it could fit into a period homelab setup, or simulated datacenter, or some other similar server/infrastructure project. If you want a powerful, period-accurate workstation, there’s a slightly different (and slightly saner) option I’ll get into in a moment.

As a silly fun machine, it’s certainly something different. While the machine didn’t turn out to be practical in the end, and to be honest I can’t see myself using it over my other retro machines for much of anything, that was never the point. The build itself was the point, along with a story and a hypothetical, and it’s something weird and interesting to try.

A slightly better plan?

There is a similar system that makes more sense that I alluded to in the first post: a dual Xeon workstation, with a more appropriate motherboard, 533MHz Gallatins and an AGP graphics card. I’m not sure if that’s really a better option for most retrocomputing purposes than, say, a later Core 2 Duo system or a contemporary P4. But it would certainly be a neat system, it would fill the niche of a period-correct early 2000s workstation, and it would probably be a lot better performing and a lot less buggy than how this one turned out.

Am I going to build one? Not any time soon, but maybe someday.

In Conclusion

This project really, really didn’t go as planned. There were a lot of issues along the way, and you can see both my expectations and excitement fade over the course of the build. It was a frustrating experience, and there were times where I just wanted to toss the whole thing into a dumpster. But I was really looking to try something completely different, and this certainly fit the bill. I was able to (more or less) answer the original question posed in the first post, too. I don’t regret doing this build, but at the same time, I wouldn’t do it again.