sean's place

The exploding power supply, the fried motherboard, and how 2.6GHz became 3.5GHz (if only briefly)

This last few weeks have either been 1) a little spot of computer hell or 2) a comedy of errors. At this point, I’m not sure which.

It all started with episkey, my main hosting server. For about 2 years, it had been colocated at Red Rocks Data Center (these guys are great, I totally recommend ‘em), just outside of Denver. After my largest tenant (an IRC server and most of the accessory services for the network) departed, I felt that the cost of the colocation far outweighed the tangible benefit to me; so, I pulled the server back home to live on my network and on my /29 from Comcast.

At the data center, episkey spent its computing hours cooped up inside half-depth 1U case; a case that would have been an ideal fit in my rack at home… if it weren’t for the 10,000 decibels of industrial-grade cooling system blowing out the back. Honestly, if the case had an airfoil and a runway, it’d probably beat “my” 172 to pattern altitude.

So, I ordered a new case (an Antec Sonata III), new quiet heat sink and fan, and a new heat sink mounting bracket (for former enclosure necessitated custom set). I built that all up in the new case and all was well, until: POP (and quite possibly, a few snaps). The power supply emitted (emut? new word, perhaps?) its secret smoke and that, my friends, was all she wrote.

The 500W power supply in the Sonata III was done; totally cooked. Oh, and it took the motherboard with it. episkey, if it were constituted solely by its motherboard, processor, and memory, was now dead. A faint miasma of death (which is astonishingly similar to capacitor smoke) still lingers to this day.

Thus, began the replacement process.

I decided to go “newish” and “low cost” — one thing I’ve realized is that for my needs, I don’t need the fastest out there. This “new” box is going to run FreeBSD.  It’s going to have a steady and predictable workload very similar to what the former Pentium 4 3.0GHz incarnation was able to do without so much as breaking an electronic sweat. So, in a nutshell, I didn’t need the latest nor the greatest. High quality but middle of the road (perhaps a little older, then), would do.

But first, a new case. If you weren’t already aware, Amazon has fantastic customer service. I explained what happened to them and they gladly sent me out a replacement case and power supply. It’s really nice to not have to worry about fighting a retailer for things like this, so kudos to you Amazon. I figured I was on my own for the board, so I started shopping around.

Ultimately, I opted for a used Supermicro board (a C2SBA+II) from eBay — it’s a LGA 775 board, has 6 SATA 3Gb/s ports, and can carry 8GB of RAM. I picked up this “baby” for a song (and $45) on eBay, then added on a 2.6GHz Pentium Dual Core and 2GB of RAM. I also added on a PCI 10base T (yes kids, that’s a whole 10 megabits per second!) card that I borrowed from Myles, as I discovered that my awesome eBay find has a bad Gigabit NIC onboard (PCIe replacement already enroute). All built out, I christened it “protego” (in keeping with my Harry Potter naming scheme) and installed (then -STABLE’ized) FreeBSD:

FreeBSD protego.dreamfire.net 8.0-STABLE FreeBSD 8.0-STABLE #0: Sun May  2 19:06:48 MDT 2010

root@protego.dreamfire.net:/usr/obj/usr/src/sys/PROTEGO  amd64

I like to beat up on machines a bit; and I like FreeBSD’s make buildworld for just that. So, for this evening, protego will be “building the world” in a continuous loop. As a bit of a benchmark, its first world took just a bit over 38 minutes (I believe episkey was in the few hour mark, avifors — my backup server — takes many many hours).

Because I’m ever the tinkerer, I spent a little time looking at the stenciling on the board itself for interesting clues and for things to turn on and off. I discovered the FSB frequency jumpers and a handy table with the settings to force it to 133MHz (533MHz FSB), 200MHz (800MHz FSB), 266MHz (1066MHz FSB),  333MHz (1333MHz FSB), and the ever interesting default setting of “Auto.” It had been set to the default — Auto — which for this processor was 800MHz.

I decided to bump and lock the FSB at 1066MHz. That jump kicked the CPU from 2.60GHz to 3.5GHz. Unfortunately, a make buildworld immediately caused a kernel panic. I went into the BIOS and tinkered with some of the settings; a little extra voltage here, a mysterious 15% addition (seriously, you can overclock it in 5% increments from 0 to 15%; I have no idea what that means either!) there, and presto: a make buildworld carried on… for a few extra minutes, until it panicked yet again.

Oh well. Can’t have it all. At least, it’s working (back in “Auto” mode).