|
|
| Overclocking Blue & White Power Mac G4/400 |
| Written by Josh Long | ||||||||||||||||
| Sunday, 25 December 2005 | ||||||||||||||||
|
This information comes courtesy of the "the JoshMeister" and is used with permission. Visit his site at http://users.aol.com/furballguy/oc/MacOverclocking.html
When my friend Ben got a 450 MHz G4 Cube, I got jealous. I have a B&W "Yosemite" G3 with a 400 MHz processor, which is a pretty good machine. However, when one of my best friends became the owner of a better Mac than mine, I began to want a G4 to be able to take advantage of AltiVec (aka Velocity Engine), and an extra 50 MHz or more would also be cool. Well, a G4 ZIF upgrade was way out of my price range at the time, and a G3 ZIF upgrade costs almost as much but doesn't have AltiVec. (Frankly, ZIF upgrades aren't all that cheap even as of late 2005, when you consider the price/performance ratio compared to buying a new computer.) A while back I'd read a bit about overclocking, and I remembered a site called "Accelerate Your Mac" (www.xlr8yourmac.com). I checked it out again and went to the pages on G3 overclocking. Following the instructions outlined there, I decided that it was worth voiding my warranty (if it hadn't expired already) in order to get some extra speed. All it would take was changing a few jumpers, and I should be able to overclock it at least 50 MHz, no problem. Well, that's what I thought, anyway.
I went again to Radio Shack within about an hour of trying to overclock my G3, and picked up a Pentium CPU Cooling Fan with heat sink (about $15, RS #273-248, no longer being sold as far as I can tell). I took it home and disassembled it, screwing the fan onto my G3's heat sink and plugging it in where the hard drive was getting its power, and then piggybacking the hard drive onto it. Then I had to decide what to do with the heat sink. It didn't seem like it would cool my processor any better than the standard one, so I removed the mounting clip and placed the sink next to my G3's heat sink, touching what was exposed of two chips on the processor board. (According to Mike from Accelerate Your Mac, these are the "cache chips"—L2 cache I believe—and he warns, "make sure no metal contact [sic] with the leads.") I then bent the mounting clip and held the new sink in place by bridging the clip somewhat tightly between the G3's sink and the Pentium sink.
I powered on and tried it out for quite a while (still at 450 MHz, in hopes that cooling was the only issue and that the new additions would be sufficient). I don't think the distributed.net client crashed after that, but Unreal Tournament demo and Excel still crashed after using them for a while. I set the jumpers back to 400 MHz and left the new fan and sink installed. Total cost so far: over $16.50 after tax. Benefits so far: basically none.
The next day, however, the processor temperature went to 39 degrees Celsius again after some heavy usage. Apparently, the desk fan helped a little bit, but was more efficient when the air outside the case was cooler to begin with. Disadvantages: It's certainly not as glamorous as an in-case solution, and on the High setting it's a bit louder than I would prefer.
Later I obtained two Cyber Cooler ATX ball bearing case fans (80 mm, 36 CFM for each fan). The cost including shipping was $14.95 (bringing the total monetary cost of the project to about $33.50; I didn't buy anything after these fans). I had to "ghetto-rig" (in the words of my PC overclocking friend) the fans in the case. I used twist-ties and clear tape in order to mount the fans sturdily. I set up one of the fans so it rested on the little ledge next to the standard internal fan, and I attached the other to the bottom of the Zip drive, facing the back of the case. Both fans blew air onto the heat sink and processor. Since installing the fans, the processor temperature was consistently reported as 35 degrees C when clocked at 400 MHz.
For the first time after installing the Cyber Cooler fans, I tried 450MHz. This time I changed the L2 cache ratio to 5:2 (180 MHz) rather than the standard 2:1 (200 MHz). I changed this setting using G3/G4 Cache Profiler from Powerlogix. Temp was 39 degrees C immediately after booting. Able to run Unreal Tournament Demo for a while without crashing or temperature increase. [Weird side effect (possibly related to G3/G4 Profiler INIT): Extensions Manager would crash with Error Type 2 upon first open. Next open would crash Finder, and give a strange error message -- the typical "There are extensions installed that are not in this set" dialog, except the letters in the sentence w e r e s p a c e d o u t like so! To try to remedy this, I restarted holding space bar, got the message (sans weird spacing), and had it Update the extension set. After bootup, I opened the Extensions Manager and was disappointed to get another crash, this time Type 1. Normally Type 1 (in my experience) rarely has to do with amount of allocated memory, but I upped EM's preferred memory to 512K from 484K and it didn't crash again.] Finally, a little while after trying UT Demo, the frontmost application crashed (ironically as I was typing results for this page) with same error message that I used to get after a while when trying to run at 450 MHz: "Unimplemented Instruction at 00000000." The only way to recover from such a crash is to force-restart the computer. Discouraged, I set the cache ratio back to 2:1, disabled G3/G4 Profiler INIT, and set the jumpers back to the standard 400 MHz setting. The temp was back to the normal 35 degrees C. Since then, I rearranged the Cyber Cooler fans in the case. The one attached to the bottom of the Zip drive is still blowing onto the processor/heat sink (facing toward the back of the case). However, I decided to move the other fan so it could blow hot air out of the case. To do this, I removed two metal PCI slot covers (the bottom two, leaving one still covered, and the top one occupied by the SCSI card). I then carefully fastened the second fan to the slot openings, making sure to leave clearance so that the case could be closed all the way. It took some tweaking to get the case to close as it should, but I accomplished the task. Knowing that hot air rises, I thought that I might gain some benefit from siphoning hot air from the top of the case down and out through the bottom fan. I cut an old flexible tube to the right length and tried to position it so that it could suck air from the top of the case out through the fan. However, this was much more trouble than it was worth: having the tube installed didn't decrease the processor temperature, and likely didn't do any good for the overall case temperature because of the blockage of airflow it caused. Plus it was very difficult to open the case with the tube attached to the fan. I decided to remove the tube. I've considered peltier cooling. However, the maximum output of the G3's power supply is 200 watts, and peltiers generally recommend 300 watts. Thus, the only way to properly add a peltier would be to either swap in a better power supply (not practical), add a second power supply in the case (impossible, wouldn't fit), or get a cheap PC case with a 300+ watt power supply and power the peltier from that (possible, but not very practial). A PC tower case would allow me to house my external SCSI CD-RW drive in addition to powering the peltier, but a second tower would take up precious desk space (and with wires and cables going between the cases, it would look very "ghetto"). Considering the potential dangers of peltier cooling, this potential solution seems a bit extreme -- especially considering the possibility that 400 MHz may be the max that my particular chip can handle. (As mentioned several times on www.xlr8yourmac.com, even chips in the same batch can have different maximum clockable speeds.) Interesting note: 8 May 2001 was a very hot day in my area, and the extreme external temperature brought my processor temp up to 42 degrees C (107 degrees F) according to Gauge Pro -- even with the desk fan on High. (Typically it has stayed around 38 degrees C recently. This is at the default 400 MHz clock speed.) The system seemed to run fine despite the heat, so this leads me to believe that heat alone is not why my machine is unstable at 450 MHz. Another interesting thing I noticed was that if I quit my distributed.net client, the temperature dropped to 38 degrees C fairly quickly. Upon relaunching the program, the temp would reach 42 degrees again within a couple minutes. I wonder if my system would be more stable at 450 MHz if I wasn't running the distributed.net client. Norton System Info 6.0 ratings (processor MHz/L2 ratio/bus MHz):
I'm curious why the slower L2 cache seems to have resulted in faster CPU performance for 400/5:2/100 than for 400/2:1/100. (Note: All tests had 4096K disk cache and VM on, against System Info recommendations.) *Note that this configuration seems to be unstable on my test system. **FPU for 400/2:1/100 scored 905 in later tests, which I consider to be more accurate.
On 30 October 2001, I read a couple articles on AMDMB.com (now pcper.com) referenced by Slashdot, and I decided that they're well worth mentioning here: After reading these articles, I decided to again rearrange my Cyber Cooler 80 mm fans. I turned around the one attached to the PCI slots so that it would blow air into the case instead of out of the case. I thought this might help improve internal air flow since the power supply (at the top of the case) blows air out. Next I moved the fan attached to the bottom of the Zip drive so that instead of blowing air sideways onto the G3's heat sink, it would blow air from the side of the case opposite the door onto the heat sink/fans -- this is close to where I originally put the other Cyber Cooler fan. (Before this latest rearrangement, I had suspected that the fan might have been counterproductive where it was because it had been opposing the heat sink fan in airflow direction.)The same day, I decided to look up some info on certain chips on the motherboard, to see what might be worth getting a miniature heat sink for. Here's what I found out:
I personally don't plan to try to get a mini heat sink for the PCI/memory controller because the only safe way I know how to attach it would be to epoxy it, and I don't want to put any kind of epoxy inside my case, thankyouverymuch. 7 February 2004 - It's been a couple of years since I last tried to overclock my G3. Now I'm running Mac OS X 10.3, which is a much more advanced and robust OS than 9 ever was. Since I was junking my Performa (the same one I had cooled previously -- (see below), I took out the old heat sinks and fan (one sink being the thin one on the 80486 processor on the OEM PC compatibility card). I thought about the possibility of putting the 486 heat sink on the Grackle PCI/memory controller, as mentioned above. It was the perfect fit! My previous concern about adding a sink to the controller was that I wouldn't be able to safely get the sink to stay in place. Well, this particular sink fit snugly and stayed in place with some heat sink compound and a tiny drop of superglue attaching a corner of the sink to the white plastic ZIF base. Beautiful! Now it was worth a try to overclock again. ...And so far so good this time! I've run SETI@home (which takes 100% of the unused CPU cycles and uses them to analyze data to aid the Search for Extraterrestrial Intelligence) for hours straight without experiencing a crash. I've also been using the computer for various everyday tasks without problems so far. So what's the speed increase? Unreal Tournament framerate is about 5 frames per second higher (about a 7 percent increase). The benchmarking tool AltiVec Fractal Carbon indicates that the processor runs about 27 MegaFlops faster (an 8.6 percent increase). I haven't yet booted in OS 9 to see if it would eventually crash like it used to when running at 450 MHz, so I don't yet know if OS X or the heat sink on the Grackle or a combination of the two (or some other factor) may be responsible for allowing me to overclock to 450 MHz without abnormal crashes. I cannot test the processor temperature since there are no utilities that are compatible with my machine under OS X. Your mileage may vary -- VersionTracker has a list of temperature testing apps for OS X. Here's the current configuration (presently working without crashing at 450 MHz, up from 400 MHz):
Mike from Accelerate Your Mac commented that the heat sink on the PCI/memory controller should not have been necessary since I only overclocked the CPU and not the system bus. This is true, but since that chip seemed to be one of the hottest on the board I felt that it wouldn't hurt to keep it a bit cooler to potentially keep the system a bit more stable. I can't verify whether this has helped stabilize the system significantly, if at all, but I've kept the heat sink in place anyway. 21 December 2005 - It's been over 22 months (close to 2 years) of running at 450 MHz, and I still have not seen any related system stability problems. My G3 is currently running Mac OS X "Tiger" v10.4.3 without any problems (and quite smoothly for such an old machine; I'm glad that Apple has continued to support the Yosemite G3 for the past several OS updates).
Tips
If you have any personal experiences with overclocking that you'd like to have shared on this page, or to give me some advice related to overclocking or computer cooling please e-mail me to let me know!
Write Comment |
||||||||||||||||
| < Prev | Next > |
|---|
|
|


I pulled off the OEM jumper block and added my own jumpers (under $1.50*,
Next, I took another trip to Radio Shack and bought a small tube of silicon-based thermal grease (about $2,
One night I mounted a standard Holmes Air brand desk fan, about 8.5 inches (22.6 cm) in diameter, approximately 6 inches (15.2 cm) away from the right side panel of the case. I tilted it slightly downward so more air could blow underneath the case. (I got this idea after reading about the
Note: I've discovered that Newer Technology Gauge PRO 1.1 seems to give a more accurate temperature reading than PowerLogix G3/G4 Cache Profiler 1.3 and Sonnet Metronome 1.4.5. Unfortunately, Newer Technology has ceased operation as of December 2000. I have mirrored Gauge PRO 1.1
