Nice build, great log. And it's done the job ... I feel re-motivated. We are slightly hampered this side of the pond. There always seems far more refurbished 4P units and cheaper in the US than europe. But I'll get there.
Not sure how far on your server rack is coming on. I would suggest some long diagonal slats down the back and sides to increase stiffness and extend the life of your corner joints. Especially important if you will be pushing it around on those casters when loaded up. If you haven't done so already.
(I was lucky enough to pick up a 40u server case from work when we moved offices recently. That will save me a whole heap of woodwork.)
Looking forward to the next installments.

hah. You're all over it. Nice one.

Looking good!! Ready for the next pictures!

The rack is essentially finished, at least to the point where it can house the 4p servers. I broke down and bought paint, the only out of pocket expense so far.
 

 
Getting ready to fasten the left side panel.
 

 
The H8QME-2 with the 8439's is in place and folding.
 

 
And finally the G34 server is in place and folding.
 
Still to do:
 
Install the bottom and third shelf
Install two 1500 VA UPSes (in transit, servers are on a surge protector for now)
Fabricate a screen door or equivalent (it is screened with temp fastenings right now)
Fabricate rear ducting and install a blower to the outside window
Build the 4p 8425 ankle-biter for the third shelf (all parts on hand)
 

Looking gooood!
I like the way this is coming to together.
 
On a side note to self I need to crank up the air compressor, time to blow some dust.
I noticed a couple of my rigs temps starting to climb more than they should be.
Its been hot here the past 2 days, but looking close I can see the dust foruming pretty good on a couple of them on there HS fins.

Black is a stealth color.  He's been sneaking up into the stats.  If he went red well, we'd all know he's out to take the crown!

Great news (apart from the Black) 

Since the basement folding rack is functionally complete and the two 4p servers are stable, productive and have been moved into it, it is time to build the 4p 8425 Ankle-Biter for a much needed PPD boost.
 
Over the past couple of months I have gathered the bits for this build. I ended up with an extra Socket F board after I bought one that already populated with 8216's and 16 GB of DDR. After deciding to build a basement folding rack with some room to grow it only made sense to my folding addicted mind to complete a third 4p since I already had just about everything except CPUs. I had originally decided to wait until I could find a smokin' deal on some 8439's before going ahead with the build, but with prices for those holding steady at $175 and up over the past couple of months it didn't look like I was going to find any hot deals any time soon. Through chit chat in another thread I found out about a seller who was willing to entertain some low ball offers on multiple 8425's and I managed to pick up four of them for a good price a few weeks ago. The only impediment to building the third 4p (aside from time and the wife) was getting the basement folding rack finished enough to clear my workspace of the other two 4p's. With that accomplished it was time to start the H8QM3-2 build.
 
Of course, having gathered the parts over the course of two months or so, they were scattered here and there in my NOC. I pretty much knew where everything was except for the updated BIOS chip I had bought back in April. After locating everything except that chip after searching high and low for it, I almost hauled out a floppy drive and the 8216's to attempt a flash. I just happened to notice the glint of a little anti-static bag wedged between some of my computer detritus about 12 inches from my daily driver's keyboard, it was the QM3's BIOS! Yay!
 

 
Here are all the goodies waiting to be united for the good of mankind. The BIOS chip is the little bag next to DDR bags. The H8QM3-2 already had active chipset coolers so i didn't have to dig out any 40mm fans. The two passive coolers to their left (NEC chip for PCI-X and LAN ports and LSI chip for SAS) run plenty cool, at 51 C and 45 C respectively, according to my IR gun, so they don't need any active cooling. Time to start slapping things together!
 

 
The second pitfall I encountered (after the missing BIOS chip) was that the 1U copper slab heatsinks wouldn't clear the caps in three out of the four CPU sockets. I'm glad I took the time to check it before applying TIM to the CPUs. I grabbed a sharpie, labeled each heatsink and traced off each area that would need to be relieved to clear the caps. They were so close to fitting, maybe 0.5 of a MM. Off to the basement shop I went. I have a cross slide vise on my drill press that allows me to use it as a poor man's milling machine, but the same job could be accomplished with a  rotary grinding tool like a Dremel. Copper cuts pretty easily as it is a soft metal. The CPU #1, 3 and 4 heatsinks each need a trim only on one side to clear.
 
I had decided this would be a "quick and dirty" build with no lapping, just schlap it all together and see what happens. Since the 8425's have a TDP of only 79 watts and an average power consumption of only 55 watts, I decided to try the stock 1U copper heatsinks to see if they would get the job done.
 

 
 Because I like to experiment I mounted the #1 and #2 fans in blow mode and the #3 and #4 fans in suck mode. Blow outperformed suck by about 2 degrees C. The 80mm fans are mounted on 70mm to 80mm fan adapters, which provides a small air plenum between the fans and sinks to improve airflow. I dug through my screw boxes to find tiny screws that wouldn't unnecessarily deform the heatsink fins to mount the fan adapters. I had to buzz one edge of the two adapters used on CPUs #1 and 2 with a belt sander so the adapters would be centered on the sinks. Pitfall #3 I guess. Not really a pitfall, but worthy of concern, occurred when I was installing the RAM. I got to the D bank of DIMM slots and noticed that the DIMMD 1B slot was oriented in the reverse of the 31 other slots! How weird is that? My first thought was that somehow I had gotten a bad board with a production error. Would it work with DDR in the DIMMD 2A and 2B slots instead? I decided to hit google and see if I could find out if this was normal. No luck with 4 or 5 different queries. I decided to see if there was a board image available anywhere that was detailed enough to see the DIMMD 1B slot. Sure enough there was and it clearly showed the reversed slot, yay!
 
I just need to plugin the PS and drives and we're ready to fire up!
 
 
 
 It booted up first try, in fact it actually tried to load the FreeBSD 7 install it had been used in. Wait! The hard drive had been used in a FreeBSD box as part of a GMirror array. This had already bitten me twice before on the G34 and 8439 builds and is a major PITA to fix in Ubuntu before the install will successfully complete and make the drive bootable. I slapped the drive into my eSATA dock on the QX9650 and wiped it seven ways from Sunday, but no matter what I did, I could not get it to not be listed as "Type: RAID" in diskpart. Made it GPT then made it MBR, used the "clear" command on it, swore at it, swore at M$, but it was still Type: RAID. Ok, the hell with it, I'll see if Ubuntu 12.04 LTS installs and if it doesn't, I know how to fix it.
 
After loading the optimized defaults and checking the hardware monitor, I flipped the #3 and 4 fans to blow mode. tweaked the BIOS settings to my preferences and it was time to install Ubuntu. Aside from the usual annoying "Installing Language Packs" delay it went off without a hitch and to my surprise it completed the install successfully and booted right up into Ubuntu without any crankiness. I guess the drive fiddling in windoze did the trick enough to make Ubuntu happy.
 

 
I did an apt-get update and installed the lm-sensors package before I blew in the [H] F@H install so I could monitor the CPU temps under load. Started up fah6 and it immediately downloaded an 8105, yay! While it cranked away on the 8105 I checked the temps and at 76 F ambient they seemed to be running a mite high at least for my taste. CPU #1 was running 49 C and after 10 minutes or so had climbed up to 52 C. Maximum operating temp on these CPUs is between 72 C and 76 C depending on who you listen to so it's not too bad, but I like to keep CPU temps under 50 C as a matter of course. I decided to let it cook while I blew in preferred customizations to the OS. I set up sshd, generated keys and swapped them with my other boxes, put my temp monitoring perl script on it, added the desktop background so I know which machine is on the monitor when it is on the KVM, added a root password (it's behind a FreeBSD firewall box that blocks virtually everything inbound) so I don't have sudo this and sudo that, changed my shells to tcsh and added my preferred .cshrc/.login script and everything was good to go. TPF on the 8105 initially came in at 24:08, but has stabilized at 23:43 since. I checked on it after dinner and it seemed pretty happy. Checked it a few more times before bed in the wee hours of the morning. Aside from the slightly high temps it appeared to be a happy machine.
 
After getting some coffee going, the first thing I did after waking up was check the Ankle-Biter. CPU #1 was up to 56.2 C and #2 and 4 were over 50 C with #3 at 49.5 C. WTH? Ambient was the same as last night and I had expected the temps to come down slightly, not go up. Looks like these unlapped 1U copper slabs just aren't going to cut it with quiet fans, especially when basement temps top 80 F in the heat of the summer. I slapped a couple of helper fans on #1 and 2 and they came down below 50 C and sympathetically #4 also dropped below 50. The helper fan I put on CPU #1 is a Vantec Tornado 84 CFM screamer and running at full blast it brought #1's temp down to 42.5 C. I used a speed controller to back it down to 3650 RPM from 5500 RPM so it is audibly tolerable and the temp is holding steady at about 47 C. Obviously the six buck 34 CFM Antec 80mm fans won't cut it, but if I am going to keep using the copper slabs, what fans do I want to put on them? I'm out of Rosewill 46 CFM's, all eight of them are on the 8539's Hyper 101's. I wish I had bought more of them. I could use the 80+ CFM 80mm screamer Nidecs that got swapped out of the SM G34 2U case, if the Rosewill speed controller that I have will handle them. I don't think there would be a need for those to run at full song in the basement unless ambient gets over 85 F and I am sure that four of them at full tilt will be audible on the first floor. Six of them flat out sounds like a freaking jet engine. I'll have to experiment with the speed controllers. I have a set of four Dynatrons, but four of those would be louder than I care for. I have four CM 212's, but I would hate to use them on the Ankle-Biter, they're in stock for another future G34 build. I would also have to lower the third shelf in the rack to fit those. I have a set of stock lapped 2U SM Socket F heat pipe coolers, but I was less than impressed with them on the 8439's and mounting  fans is a real pain on those.
 
What does anyone think an acceptable sustained max temp is for the 8425's? Is 55 C anything to be concerned about? Honestly, I wouldn't be all that upset if one of them cooked itself, it would be a good excuse to upgrade, but I don't want to have to do that for a while.
 
Anyway, that's the latest chapter in my 4p adventure.
 

After successfully testing a modified Rosewill fan controller on a Nidec 80x38 80+ CFM screamer fan and successfully testing the combination on one of the 1U copper heatsinks of the Ankle-Biter, I modified three more of the fan controllers and mounted them on the other heatsinks.
 

 
Out of the box the Rosewill fan controllers are a spaghetti factory of wires. I briefly considered making an external fan controller panel out of them, but I really don't want a mess of wires running all over the place. This is the Ankle-Biter, I want it to be as simple to deal with as possible. The fan controller is set up to take 12v power from a four pin molex connector, supply variable power to the fan with a three pin connector and supply the tach signal to the motherboard from another three pin connector. Since the SM mobo is designed to supply pretty high current from the three pin connectors to screamer fans I decided to simplify the controller wiring into a three pin in and three pin out deal and make the controller board fan mountable. I keep all kinds of connector parts in stock to make custom cables when needed so it was no big deal to re-engineer the cables. A quick buzz with a drill made one of the board mounting holes big enough for a fan screw and a snip with the Xuron cutters took off the corner of the board that intruded into the fan inlet area.
 

 
It's sort of a Mickey Mouse setup, but it works absolutely great. I closed off the room and shut off the A/C to heat things up and with the Nidecs at full song with the ambient temp at 86 F (reading a different thermometer, not the one in the pic) the hottest CPU, #1, didn't get over 47.2 C. This experiment was a complete success! I think this setup would cool 8439's adequately, especially with some heatsink lapping. To my surprise, the Nidecs at full blast (and I mean literally BLAST) are nowhere near as noisy as they were in the 2U case, which generated all sorts of funky harmonics. I think running them flat out in the basement folding rack won't cause any noise problem on the first floor so this setup should nicely take care of things on even the hottest days of summer. Dialed down to about 3500 RPM they are acceptably quiet sitting 18" away from my workstation. I would have no problem talking on the phone right next to them.
 
Once again, in keeping with the budget Ankle-biter folder, it was a no cost mod as I already had the fans and had the controllers. I'll let it run up here for a couple of days before moving it to the rack in the basement.
 

Well I suppose it is a ghetto build so it's appropriate!
 

Hey, I even have to manually short pins 1 and 2 on the front panel header as the power switch. C'mon, that's ghetto!
 

Coincidentally, I made up a couple yesterday afternoon since the other 4p has my test bed switch on it right now. I had a couple of tall jumpers that were perfect for this mini-project. The core was held in by the same sort of tab catch used on 3 pin fan plugs so it was easy to extract without damage. It was literally the perfect width to fit between the contact legs of the momentary pushbutton switch. After soldering both sides I snipped out the jumper bridge and trimmed the jumper housing to fit with a razor blade. The second one is identical to this one. In keeping with the budget folder theme, I already had the parts so it's another no-money mini-mod.
 

I always set the bios to power on if it is available, but with the PSU buried behind the board in the basement folding rack it will be much easier to use the motherboard switch. That way I won't have to worry about one of the vicious Nidecs biting me if I am not being careful
 

Wow this is very impressive.
 
So those temps you got, while you turned off the A/C in your room under full load your CPU's never got higher than 47.5C?
 
 

According to AMD the temp sensors on a lot of the K10 parts are unreliable and shouldn't be trusted. in linux k10temp has a "force" parameter to ignore the erratum. also, on the H8QMx-2 boards the fans can be pwm controlled in linux by setting the "reset" module param for wm83793 hw monitor chip.
Right now my 4 8423's are reporting temps 10 below ambient. at least on my h8qm8-2+ the bios seemed to report valid temps, but linux and windows still reported them incorrectly. my h8qm3-2 bios readings seem much closer to win/lin reported values.