DIY Junk


Had time to pull the best guitar I built out. It’s a double-bound Toyota Spectra Blue ash body with a bound maple neck and Strat head, and SD pickups. Now, as we all know, the trifecta of acceptable guitars goes Gretsch, Jazzmaster, Telecaster. And this is one of those. If it wasn’t, it wouldn’t be part of the trifecta unless we were in Universe B. In that universe, I never gave up playing the cello. I like to think of that universe sometimes.

DIY Junk


I picked up a ScanGauge E, and fortunately the Element has a really top-shelf place to put it, especially if you are like me and are driven to rage by exposed wires.

The space behind that panel is occupied by nothing, so it’s a really simple matter to drill through and mount the ScanGauge. The cable run is really basic, too, as it’s just a straight shot over-and-up from the OBD port. Do excuse my use of Velcro straps and twist ties; I didn’t want to zip tie the cable down until I was sure it was a good location that didn’t rattle or anything.

The only problem is that the Element is a little stingy with the data, and doesn’t report fuel-flow information. So the ScanGauge’s MPG information is pretty far-off until it’s had a few tanks of gas run through the car to allow for calibration (it uses other sensors, like throttle position and MAF, to estimate MPG). I guess it’ll be a month or so until it reports worthwhile economy data. This is the reason I canceled my Automatic order; that thing doesn’t allow for calibration, and would be generally useless.

It’s pretty nifty otherwise, though. I was a little disdainful of the ScanGauge because of the thick, old-fashioned design and the high price, but on any level other than the superficial it really does seem a lot more well thought-through than alternatives like the UltraGauge. I guess my next step is to break out the Dremel and sink the ScanGauge into the panel. I think I’d need to come up with some sort of bezel to hold it, though (or take the ScanGauge apart and see if I could repurpose its own), since the location in the Element would be pretty tricky to do any actual work on.

DIY Junk


I thought this was mildly-interesting enough to write about. I bought a third-party-refurbished fifth-generation iPod Nano from They’re the last Nano with real buttons, which is a lot better for using while driving or at night. Anyway. It got here, and it’s pretty curious. The company that refurbished it seems to have dropped it in a plastic case which is pretty close to the original metal one. However, the interface is unusably slow. It just won’t scroll, or anything. My first thought was that it was a fake Chinese iPod. However, it plays Apple Store DRM’ed files, it takes Apple software updates, and it talks fine to iTunes. If it’s a fake, it’s really good on the iTunes integration front!


So, I wondered why it was so junky. These things are so simple that there isn’t much to break. There’s a Samsung ARM processor, a Cirrus audio chip, an Apple-branded power manager, and most of the rest of the space is occupied by a single chunk of Toshiba memory. Was the memory replaced with something crappier when it was refurbished?

One way or another, the only way to tell is to pull the thing apart. But then I’d be out eighty bucks. So I benchmarked it instead. Here’s what the Woot Refurb iPod got:

Okay then. There are two more 5G Nanos in the house — I have one, which I use with my headphone amp, and R. Monkeys has another. So I tested them both. Here’s mine:


They’re similar in some ways, but check out how much slower the refurb’s reads are. Of particular interest are those 4k random reads. Little random reads are what make any computer feel snappy, and the Woot iPod’s 4k random reads are majorly off the OEM Apple one. Let’s see the same category on R. Monkeys’ iPod:


It’s not quite as fast as mine, but it’s close — and way ahead of the Woot iPod. It seems as if the Woot iPod’s memory is just too slow to keep up with system functions.

I wish I had the $80 to pull it apart, though.

DIY Junk

Follow-up: Putting an Nvidia 670 in a Mac Pro

I returned the EVGA because it just wasn’t worth the fuss, but then Nvidia started releasing their own OS X drivers, so I thought I’d give it another shot. I got another 670, an Asus DirectCU II, and tried it again.

The card uses a different style of cooler than the 5870 and EVGA 670 FTW, which I thought might help with the heat concerns. It also looks nifty!
Installation was totally normal, as usual. I didn’t have issues with the EVGA in Slot 1, but someone was reporting trouble; as they’re both x16, I figured I might as well put it in Slot 2 to get some cool air ’round the back anyway. Probably made no difference to temperature or stability. Whatever. It booted up just fine, and I installed the newest OS X drivers from Nvidia, as well as Asus’ GPUTweak thing in Windows. That came with an older driver (301.something?), so I installed the newest 306 drivers on the Windows side.
So far, so good…
Luxmark is slightly worse than the EVGA.
Cinebench is only marginally better than the EVGA, and still worse than the Apple 5870.
Games were the same. Most things are slightly faster than the 5870, Source games have a tiny bit of microstutter (but are just fine), Civ 5 still has numerous graphical glitches, and the wrapped version of Skyrim still sucks (plays beautifully under Windows, of course). I was hoping for an improvement to everything, with the new drivers from Nvidia, but an improvement to anything would have been nice. Long story short, there still isn’t a worthwhile difference in game performance between the 5870 and 670 under Mac OS. Unless you only play WoW.
The card’s non-reference cooler was fine. I put in some vented PCI slot covers (as the Asus blows up, not out, and I wanted some way for the hot air to escape), and that was that. The PCI bay fan ran faster, but temperatures were the same. The Asus was significantly quieter than the EVGA or the 5870. I didn’t notice the rear hard drives (on which the Asus 670 exhausts) to get very hot; they reported about a 10°F increase over the 5870. The PCI fan must have been keeping most of the air off ’em.
The real problems are in Windows. Asus’ software is kind of crappy, but functional. Most games were great — Skyrim, New Vegas, and so on — but World of Tanks had really horrible microstuttering, regardless of AA, V-sync, or any other graphics settings in the game or Nvidia’s control panel. Hours and hours of screwing with settings, driver versions, and so on later, I pulled the goddamn thing out and threw the 5870 back in. Given how few reports of the problem there are, I expect it’s some arcane conflict with Apple’s Boot Camp drivers. Maybe I’ll try again this afternoon, or maybe I’ll just return it and use the $400 to go on a nice trip somewhere with R. Monkeys. It is to some degree worth all the crap (dang, Skyrim looked nice), but the point at which it isn’t approaches fairly rapidly.
Edit: Solved! Hardware fault, according to ASUS. Exchanged 670 for another EVGA FTW — works fine. Done!
DIY Junk

Putting a GeForce GTX 670 in a Mac Pro

Update from 2017: This post is five years old now, but lots of people still find it through Google. So I thought I’d update it just to say that I sold the Mac Pro with an Nvidia card to my dad, and it has been working fine through the various OS upgrades. Googling around suggests that the 10×0 cards work fine in Mac Pros too. I replaced my Pro with a MacBook Air and a custom gaming PC, but I think the 4,1/5,1 Pro remains a really good computer.

I thought I’d try dropping a GeForce GTX 670 into my Mac Pro, since apparently 10.8 can drive it (you just won’t get the EFI boot screens). I bought the 2GB EVGA 670 FTW, mostly since that has two six-pin power plugs. The top OEM card for the Mac Pro is the Radeon 5870, which is a good card but also pretty old, so — on paper — the 670 is a nice upgrade. The 670 is actually a little cheaper, too. The drivers in Mac OS are still pretty immature, so my goal was to get equal-or-slightly-better performance for now, and way better performance in Windows.

Installing it is pretty straight-forward, if you’ve ever done computer stuff at all. Here’s my Mac Pro as it sat, with the 5870.

The 670 is a little smaller.

The 670 dropped right into the 5870’s spot, and hooked up to its power plugs. These are standard parts, of course, so if it didn’t that would mean I’m really dumb. You can see that I also put in a little PCI fan (powered from the spare optical drive SATA port, and a SATA-to-Molex adapter). This pushed a lot of air, but it was really noisy, and it didn’t seem to make much difference; the only thing making much heat in the PCI bay is the GPU (the rear hard drives being idle most of the time), and both the 5870 and 670 keep their heat fairly contained in those plastic shrouds, and push it out the back. Maybe the fan did a little something, but it wasn’t worth the noise. So, the little PCI fan is gone now.

So I did that, and it booted. That was nice. You don’t get the grey boot screen (the video drivers loads when the OS does, so it appears to boot straight to the desktop). With the fast SSD booting, it was just, like, “push button, wait 20 seconds, be at desktop”. The obvious problem is that if you find yourself looking at the startup manager a lot, this is a bad solution — you no longer get that. In that case, it’s worth keeping things OEM or finding a flashed Nvidia card that gives you the boot screen.

Anyhow, it worked just fine. I went to run benchmarks, and LuxMark couldn’t load. No OpenCL. So I found a fix from Netkas, installed it, and hey, liftoff. So, here are benchmarks and observations.


Here’s LuxMark doing Sala. That’s a pretty solid improvement. I would have hoped for a bit more, but I’m not complaining. Of course, the real strength of an Nvidia card over ATI is CUDA; the 670 is probably about six million times faster in After Effects.

Wait, crap. Check the GFX details; the 5870 is the first result, and the 670 is the second result. That’s annoying. Cinebench is pretty dated these days, but the numbers went down! My precious numbers! Well, Kepler drivers are immature. Hopefully this changes over time.

Now, games. Everything was tested at 2560×1440 at the highest detail settings.

Starcraft 2 gained about 10 FPS (61 vs 50, average) on the 670, tested playing a Bly v. Tarson replay. So that was cool; you can see that Barefeats got the biggest improvement on a 570/580 from SC2 at those settings. Then I tried Civ 5; not only was it slower on the 670, it was kind of glitchy. Source games don’t really tax the 5870, so it’s hard to see much improvement in practical play, but the 670 was about 15% faster across Portal 2 and L4D2. L4D2 did have a weird bit of stutter when levels started. Dirt 2 ran great on both cards, although it remains a super-annoying game. Cities in Motion was, I think, slightly slower on the 670, but played fine. It gets jumpy with a huge map and the fastest game speed. The Sims 3 was just about as fast, too. Weirdly, Diablo 3 didn’t improve at all — you’d think whatever worked for SC2 would work for D3, but no.

My wrapped version of Skyrim was the only game here that wasn’t really playable on the 5870 at 2560/Ultra, so I was excited to see what the 670 did. It did nothing, actually; the game was still jumpy. If I had to cork sniff, I’d say the 5870 was more consistent — the 670 seemed to do a little better as long as you were standing still, and a little worse when you turned. This isn’t really a reflection on either card.

Everything else worked fine. I’d heard people saying that Steam didn’t work with PC GPUs — it did. All my non-game programs worked as well as before. Temperatures were about the same as before, and the 670 managed its fan just fine (it’s quieter than the 5870, which made me nervous). Restarting in Windows showed off what the 670 could actually do, of course, although I found the difference marginally underwhelming in stuff like New Vegas and World of Tanks. But then, they’re not taxing anyway, and the only tricky stuff for the 670 to add is anti-aliasing. Anyhow, the 670 in Mac OS is pretty much as fast as the 5870 except if you like Civ 5.

I ended up a little conflicted. The 670 is better than the 5870, depending on what you’re doing, but not by that much in most cases except for CUDA. It’s unsupported, and there were a couple glitches. The 5870, in comparison, is an OEM card that developers (presumably) test against, so you won’t see much in the way of issues, and it’s still the fastest OEM Mac GPU by a comfortable margin (over the iMac’s 6970M). However, the 670 is definitely faster in Windows, and there’s a lot of potential for the Mac drivers to improve and let the 670 really go nuts.

DIY Junk

2009 Mac Pro W3680 Swap Addendum

I got a bunch of questions regarding the 3.33 six-core swap in my Mac Pro, so here are answers:

– When I say 10°F cooler, I mean that it idles at around 95°F now, instead of ~105°F. My Mac lives in a cabinet, so 95°F ain’t bad. It gets up to 125°F when I play games or render stuff.
– I got a good deal on the W3680, so the total cost was $450 (I haven’t sold my W3520 yet). If I hadn’t gotten a good price, I probably would have gone with the W3570, which is the 45nm 3.2 four-core. Actually I probably should have gone with that anyway; I don’t think there’d be much difference in 95% of the stuff I do, and the W3570 is $250. You can also use the LGA1366 i7s if that saves money, although they’re not really any cheaper than their equivalent Xeons.
– No, you can’t use the W3680 in dual-processor machines. The equivalent for dual-socket machines is the W5590. They should be around $500 each these days for used chips. That’s a pretty screaming deal, since the retail price is still $1500 each.
– I don’t have a Kill-a-Watt, so I can’t measure power consumption, but I don’t think it went up by much, judging from Apple’s published specs for the 2010 vs 2009 Pros. Putting in the newer video card probably did a lot more.
– Performance isn’t really any different now that I have the matched set of 1333MHz RAM in there. On the plus side, my RAM is all the same color now, which looks fancier.
– The 5,1 firmware is totally stable. Hasn’t crashed or hung once. I upgraded to the 10.8 GM, and it’s totally fine.
– One guy asked if it would be a good upgrade for a MacBook Pro. Um, not really.
DIY Junk

Swapping a 3.33 6-core into a 2009 2.66 4-core Mac Pro


Use the long Allen key to remove the heat sink. It’s also a great time to blow all the dust out of it. When that’s done, you can pull it off and get this:
Note that the standard application of thermal compound is actually pretty bad. At least, it was bad on mine. The processor side will be like so:
Again, not great; there’s bubbles in the compound, and goop falling off the side of the heat spreader. Use the thermal compound remover (ArctiClean in my case) to get this crap off the heat sink and processor, and the surface purifier (where applicable — that’s the “ArctiClean 2”) on the heat sink. Then, you can undo the latch and carefully take the processor out, being careful not to touch the contacts on the processor or the daughterboard. The new processor drops straight in, like so; make sure to get the alignment right:
And latch it up.
Then, after you’ve made sure the surface is clean, it’s time for new thermal poop. On these Xeons, Intel recommends that you don’t cover the whole thing in thermal paste. Instead, taking the alignment arrow to be pointing south-west, do a bead running north-south (this one’s thinner than it looks; you don’t wanna cake it on there):
Once that’s done, you can bolt the heat sink back on. Put it over the alignment pins to get it on straight, and then screw it down evenly. You’ll have to push down a bit on the Allen key to get it to engage. They will have a firm stop once they’re tight; obviously, you don’t wanna exceed that.
Pow! Put the processor tray back in and go!
It’s faster than the 3520 to a fairly nutty degree. My system already has a SSD and 5870, so it was already pretty speedy as far as Macs go, but the W3680 was like getting a new computer all over again. As a side note, you can now use 1333MHz RAM. You don’t have to, though, as the W3680 is fine with 1066 (I’ve got four sticks of mis-matched 1066 in here; I ordered 16GB of 1333, but it hasn’t come yet and I was eager to try the new processor).
As a side-bonus, the stock thermal compound was so junky that my system actually runs 10°F cooler at the CPU now.
Here are benchmarks.
Before on the left, after on the right. Full results are linked. These are 32-bit scores; 64 seem to be about 1000 higher for each.
I didn’t do any serious testing with games, but improvements (at 2560×1440/full details) seem to be anywhere from 0 to 30 FPS, depending on the game, with X-Plane being the high end. Most are around 10 FPS faster. The big improvements are in multithreaded apps, of course, since games don’t really use those two extra cores.
A ton faster and a bit cooler. Neat!
DIY Junk

This was a fun project…

As a result of the shitsville Mac Pro “update”, I got a 2009 Mac Pro, which is pretty easy to upgrade to the current spec. You just need a newer Xeon and one of Apple’s insanely overpriced and outdated GPUs. A quick firmware bebop, and yer done.

First these showed up, which is fun: a 480GB SSD for OS X, and a 240GB SSD for Windows. These will join an ATI 5870 and a Xeon W3680 (3.33 six-core) in making the thing newer. It came with a 2.66 quad-core Xeon and a GT120. The 5870 is pretty old, and Nvidia releases Mac drivers for all their cards now, but the 5870 seems to keep up with Nvidia’s 570 in the real world, at least under Mac OS, so I decided to go OEM.
I’ll go into detail on the upgrade later, though, after I have time to install everything and stop weeping after having to pay $450 for a 5870. The fun project was tearing apart enclosures to harvest the delicious meaty platters within.
It was a straightforward job of butchering Seagate, Iomega, and Western Digital enclosures. The WD one was sensible — the disk was more or less suspended in a lightweight housing — but the Seagate one was a nightmare. The drive was more or less mummified in metal and plastic. I don’t know how the failure rate on these was as low as it was.
Worked out fine, though. 1TB for backing up the Mac SSD, 1TB for backing up the Windows SSD, and 3TB as a local copy of my media server.
My elementary school’s after-school program had “take-apart day”, where they encouraged clouds of pre-teens to demolish electronics, without providing any hope of reassembly. This may have been before people knew how much crazy toxins that would release. Actually, that might be why I don’t have a memory. At any rate, it was fun to re-live, and also to carve open drive enclosures with a Ka-Bar.
In other news that doesn’t matter at all, how ’bouts that purple. I used to think one didn’t see many bright yellow websites around, but then I saw some, so now we’re doing bright purple. Wait, I looked at Flickr. Okay now we’re doing orange. Now I looked at Ars Technica. Um, weird green, we’re going weird green.
DIY Junk

Fixing one’s climate control

I couldn’t find a detailed guide to this, so I made one just in case.

The W123 HVAC is pretty funny, because half of it relies on vacuum actuators and stuff (if your center vents don’t work, that’s why), and the other half relies on PCBs soldered at 90° to each other and then subjected to (usually) diesel vibration for 30 years. Failure of the latter can cause the HVAC not to come on, to come on sporadically, or to come on full-heat or -cool. So re-soldering is a good way to go. Let’s do that…

DIY Junk


R. Monkeys and I both realized we liked Pokémon a lot c. 1998, so we decided that it’s on over the holidays. And because my response to challenges is to dump a lot of energy into areas that are tangenitally-related at best, I had to craft a Game Boy of untold power to wield, and it will slowly corrupt me like the Soul Edge. So It has the buttons and de-labeled back of a Pokémon Edition, the front of a yellow regular one, replacement little rubber button pads from a DS Lite, and the frontlight from an Advance SP. So I guess it’s going to shoot a stream of energy into the sky and start using me as its host.

At least it didn’t cost money that wasn’t spent years ago. I used to collect handheld game thingies, and while the good stuff was sold in, like, 2003, stock Game Boy Colors and a Pokémon Edition in rough shape are among the things still around. Those and a Game Gear. Hm, there was rarer stuff, like the Turbo Express, but I think the funnest were the Nomad and Neo Geo Pocket Color. The former was just a Genesis, and the latter had a bunch of weird Japanese RPGs and Match of the Millennium. There aren’t enough submarine RPGs.

Too bad I can’t find my old Red cartridge. My crack team of killers probably died long ago. I’ll have to do a “getting the gang back together” montage.

DIY Junk


I have a few of the old colorful iBooks (a Tangerine and a Blueberry), which I like because they’re indestructible, and have a much more naïve design than modern Macs, which are quite Germanic and boring. The Tangerine is overclocked by 133mhz, has CompactFlash instead of a hard drive, and I re-celled the battery so it can get about 12 hours of life. It makes a great simplistic writing machine.

Of course, the essence of technological simplicity is needlessly complicating and overthinking it, then rationalizing more nonsense purchases. I like the FireWire iBooks because the colors are nicer (it’s less translucent and more white), and I also wanted to try a shady Chinese SSD instead of CompactFlash. So I got a preposterous KingSpec 16gb IDE SSD, and another iBook, a 366 Indigo. Here’s the procedure and a little test of the KingSpec.

Here’s the SSD. The mark of quality… must be around here somewhere…

There’s the computer with its keyboard off. In a nice world you could get to the hard drive from here, but the Clamshell iBooks (and every other iBook) are actually among the more notorious Macs to re-hard drive.

There’s the removal of the top case. Surely one can now get to the drive….

…oh. Then the EMI shield comes off. There’s the 60gb 7200RPM drive that I transplanted from a P4 2.8 Toshiba. I was going to overclock this computer to 433mhz like the older one, but Apple moved the resistors that you re-solder to do so in these second-generation clamshells, and I have no idea where they went.

At this point, I became paranoid that the Chinese SSD was playing cards, or nothing, so I took it apart.

There was actually real memory inside there. And it’s from Hynix, a brand I have heard of! I didn’t take the sticker off the controller to see who made that (I thought it might let the ghosts out). It isn’t JMicron, though.

There, now it’s in. I got this down to about 50 minutes from start to finish, so it doesn’t seem that hard. Putting the display back on is kind of annoying because the clutch cover has to go over some tabs, but it’s not bad.

Anyway, then I tried it out. It starts reasonably quicker—49 seconds from power-on to opening a browser, compared to 1:35 on the hard drive. The battery meter on the OEM battery jumped from 3:55 to 6:20, 13 hours on my custom battery, and of course it’s also silent. Here’s the XBench score; SSD on the left, HDD on the right.

You can see the SSD is a lil’ faster with small sequential writes, slightly slower with sequential reads and big seq. writes, and way, way quicker with random reads. That’s nice, because random reads are what a computer spends most of its time doing. For modern comparison, here’s my MacBook Pro’s 500gb 7200RPM SATA HDD (on a much faster bus, of course):

So the cheap Chinese SSD may have an awful controller, but it overpowers a modern HDD by a factor of 10 in small random reads, and that’s handy for making a computer feel quick. Although this one’s pretty CPU-limited. It’s a lot quicker than the CF “SSD”. So I’m pretty happy with this experiment. It runs OS 10.4 and WriteRoom just fine. My rationalization for keeping the other iBooks around will be “emergency Starcraft 1 LAN party laptops”. That actually happens sort of frequently around here, so I am being prudent.

DIY Junk


I got new headphones, and then left them alone overnight, and they got better!