First, just like to thank anyone in advance for any words of wisdom or advice for the following issue.
I recently bought 4 used servers. A pair of SGI 1200s (each with dual Slot 1 p3 700s, 100mhz bus) and a pair of IBM xSeries 342s (each with dual socket 370 1.13ghz Tualatins , 133mhz fsb).
Yes, they're old, yes, they're relatively slow, but that's exactly why I bought them. I want to sharpen my distributed computing programming skills. Here's my question, one that should test you guys pretty good.
The end-of-the-run SGI 1200s used an L440GX+ board that could use up the 1.0Ghz/100mhz P3. Remember those? Unfortunately, my boards are a slightly earlier revision that can only do up to the 850E (oh well).
I'm fishing around for some SL6BY core Tualatins, just to upgrade an overclock one or two of the xSeries machines. And I'd sure hate to waste the Tualatins I'm going to be yanking out.
Does anyone know of a trick or a way of running the 1.13 Tualatins on the SGI board at 800 or 850mhz? There are still some slocket adapters floating around, but I need to kill the FSB, and if memory serves, Intel chips didn't like it when that happens. (Oddly enough, I think 1.13GHz/133, when brought down to 100mhz fsb, equals 850 !) I also have to worry about going from the Tualatin pinout to the P3cu pinout. (Powerleap...?) Does anyone know if I have a chance in hell of making this work? Any experience to be offered? Any chips to sell?
I can't see why the CPU's FSB wouldn't downclock to 100mhz. It's been a long, long time since I've dealt w/ Slot 1 machines. If I'm not mistaken, jumpers were still heavily used on those boards. Check and see what jumper settings you have.
It's worth a try at any rate. Those slot 1 to 370 adapters are only $10 on ebay.
OK... i did a little more research. People have had much success with this on the 440BX chipset. No one has mentioned GX. But since the only real difference between 440BX and GX is memory support, I don't think it'll be a problem.
A couple forum posts stated that people have done this w/ 440BX and the CPU was downclocked from 133mhz x (whatever) to 100mhz x (whatever).
Do you really need to overclock them in the first place if your main purpose is to use them for experimenting? If you are experimenting with distributed computing programming, then you can simply lower the load given to each computer. For example, you can just offer simple or dummy computations to each computer. This way you can practice and develop the core architecture of a distributed program without being effected by a slower processor.
I know it would be more fun to overclock them, but if its going to eat up alot of time from your real goal then it may not be worth it.
First, mpjesse thanks very much for all of the information and the posts. you're just the kind of replier I was hoping would turn up here. I appreciate the links and your time.
MesaRectifier: I'd like to OC the 1.4s if I go that route, but that's more of a fantasy than anything. The servers did come configured with the 1.4Ghz chips, so I'm confident that would be plug and play. (or close). I haven't even really studied the mobo layout and such, but you're right, it probably won't let me select incremental freqs like an enthusiast mainboard. I think What I'd really like to do is just bump the FSB, because they seem to starve with the paltry 133 bus.
DrBlofeld: no, I really don't need to OC. I've been running SETI @ Home (the newest BOINC client version) on both my AMD 64 (5.0 hours_ & Pentium M laptop (6 hours) & now the Servers (13 hours), and while I'm impressed with the Tualatin, I'd just like to try my hand at squeezing some more out of it as something to do when I'm bored.
Again, thanks to all of you for your input, it's exactly what I was looking for
Tualatin was one of Intel's best processors. It was leaps and bounds better than the Pentium 4. But by the time tualatin hit street, Intel had already invested too much into the Pentium 4 (Willamette) to bother w/ improving upon Tualatin.
I've heard rumors that Tualatin was Intel Israel's first project. Of course, as we all know, Intel's new line of Core processors borrows heavily from the Pentium M, which was also designed in Israel. And the Pentium M was an improvement upon Tualatin...
So one could say, Tualatin was the beginning of Conroe. I'm being very liberal there. Wiki has a decent write up on Tualatin on their Pentium 3 page here.
I decided to give it a shot and see what happens. It'll be worth it either way. I just ordered a pair of nice slocket adapters that touted both dual processor + Tualatin support. All I have to do now is find a set of cheap SL6BY cores. There's not some bad deals on the SL5XL which seem to be easier to find. I'm assuming I want the sequentially later model.
If anyone cares, and if it works, I'll post some benchmarks on the p3-700 vs Tualatin @ 850
Having said all that, I'm not really an intel OR amd-fanboy. I just like what strikes me as cool. I came off an OC K6-2 400 (aircool to 523) to a p3Cu 533 and I couldn't believe the numbers the benchmarks gave me. From then, I salivated over the coppermine and later P3 tech. Later, I did the AMD Route, until now when you can finally buy this old stuff on the cheap. I always wanted to get my hands on the tualatin. And it's funny what you say in your history lesson, I had heard the same things, so I knew these chips were bound to impress. And they do! I was blown away by what my 1.6 Pentium M is doing on the seti Test. neat stuff
yeah, Pentium-M rocks. it's a shame that it took intel so long to realize it's potential. but again, i think internal politics held it back. for whatever reason, folks were ridiculously high on the P4.
I don't know if anyone was monitoring this thread.
Thanks to all above for their input.
I've learned a lot in the previous month about P3s, upgrading, and the whole business.
The long and short of it is this. the SGI 1200 uses the L440GX+ mainboard, which is a version of the good old 440bx board. However, intel claimed zero compatibility with slocket adapters, and even less compatibility (if possible) with Tualatin core stuff.
The cool 8) thing is though, I got it working. Using a pair of SLOT-T slocket cards from upgradeware.com, the system was up and running immediately at the guessed at 850 mhz. Unfortunately, it kept reading the L2 caches wrong, and then disabling them (boot times slower than a P1.. 100mhz memory carrying all the load).
After banging my head against a wall for a few days, I finally found a bios update at powerleap.com of all places that they list to be used with their junk. So I put that in, and some of the BIOS errors actually went away! (Interesting to note that the BIOS they had, by the numbers, was no where near the latest release). I'm not sure if it was tweaked or not, either.
After I had some limited success with that, I decided to take a chance and email upgradeware themselves (which seem to be an asian-pacific joint), and I got one of the nicest techies ever. He sent me a microcode update which actually worked on my bios (the one from powerleap). I'm assuming it'd work on a lot of bioses, but the fact it worked on mine period threw me through the wall. After installing that, I rebooted and... PEFECT! 846 mhz (listed in windows as 1130 mhz processors running at 846). No more bios messages, full caches running at full speed (CPU-z) and better #s than anything else at 850 mhz according to SiSoft Sandra. I couldn't be more pleased.
I put in a new U320 15K harddrive (plug and play, can't beat that. works right on the old SCSI2-LVD setup!) and linux boots in a flash.
A word to the wise for anyone trying this: I highly recommend using powerleap stuff. They use an independent voltage regulator module. SLOT-T gives you a jumper but if your mainbord doesn't go down that low, you'll massively over voltage your chip(s). Powerleap avoids the mainboard cpu voltage entirely (very cool). That's my last nailbiter