Working on the G5 quad liquid cooling system

I don't think that "pushing its limits and thus unreliable" is really true. IBM JS21 blades are available in 2.5GHz and are air cooled. Heat production and power consumption of the G5 are inline with other offerings of the era. For example, something like the Megahalem (http://forum.corsair.com/v3/showthread.php?t=82954) might air cool the Quad just fine, and quiet, if mounting was feasible to do and it fit in the case.

Btw. reading your repair description again, did you check that the pumps actually work? Did you open and inspect them? If one pump fails/failed the ASD, wouldn't that be the most obvious cause of overheating? Iirc, the pumps have around 50000 hours mtbf and if one pump has been pumping air in a low coolant situation that would probably shorten its life quite a bit.

Tore down the system to bits. A neat detail: the inlet barbs of the CPU blocks appear to have some sort of screen filters in them. The filters were pretty full of brownish ... nn... stuff, which explains why the metal bases were clean.

Haven't tried it. I briefly had a 2.3 DC unit, and it didn't run very hot. The fans never revved up from the minimum. I'm sure that if the heatsink mounts into the quad, which wouldn't sound too far fetched, converting a quad to a 2.5 DC would be much easier than fixing the LCS. Even if the heatsink needs some duct tape and such to mount, it would still be easier imho. The quads do start even without the pump connected, so, not having the LCS won't be a problem. But one has to have a broken 2.3 DC and at least, in these parts, those still cost well over 100 € used.

It might even be easier, than to fix the LCS, to take the metal framework of the LCS unit, keeping the heatpipes, and retrofit it with a couple of standard pc heatsinks. There's plenty of space for extra holes for mounting the heatsinks. The fans are already there, so only proper tunneling of air might be needed.

I feel that a bigger longevity problem with all these G5s is the BGA joint issue with the motherboards.

zampanone, one more thing I'd like to ask you. How well did you check that the radiator was actually clean. For example, pouring hot water in, shaking, draining, and checking the water for pieces of residue would be one way of doing it. Or did you just trust that the acid treatment worked? This is something I've been wondering about the other rebuilds as well. Most of the times the radiator has been in use for quite some time and if the rest of the system is full of crystallized residue (as in the Dual 2.5 cases) it'd be a small wonder if the radiator wasn't.

I did a couple of white vinegar treatments and used boiling water. Couldn't get crystal clear water out. And the radiator didn't really seem to be clogged to begin with, i.e. clean barbs and good flow with the pump. I think there might be bigger clumps that move around and sometimes cover the outlets.

I was once told by a service tech that if you re-apply thermal paste in a G5, thermal calibration need not be run. I looked at the darwin sources (AppleMacRisc4PE has the relevant bits), and there appear to be quite a few parameters in the rom. Stuff that's inherent to the processor from time of manufacture (see https://www-01.ibm.com/chips/techlib/techlib.nsf/techdocs/9CE44BF501282CD287256F CB0050EAB6/$file/970mp_using_therm_dio_an_a.pdf and https://www-01.ibm.com/chips/techlib/techlib.nsf/techdocs/3D6E60DFC12B58B4872570 960048D859/$file/970mp_therm_calib_value_collection_a.pdf) and stuff that's surely affected by the thermal transfer of the cooling unit. I would find it surprising, however, if improving the cooling system would suddenly fry a cpu that has the original/factory calibration info and was limbing along with an impaired cooling system.

Slowly answering my own questions:

Got replacement o-rings just in case. Expert opinion was that the originals were silicon. The replacements measure 31,37 mm x 1,78 mm and are EDPM. I wouldn't mind hearing why Apple, or whoever, chose silicon for a mostly high temperature water system. Viton, apparently, can't take hot water very well, so it isn't recommended for this sort of application.

But silicone is not good with water vapors. Parker O-ring guide lists silicone as "moderate water resistance." And most importantly, EDPM is recommended everywhere for water/glycol coolants and looks like the standard practice.

So, tried to refill the LCS a few times. I used a vacuum method: pulled a ~29 inch Hg vacuum through the service port, closed the valve, connected coolant bottle to the port, and let it draw in the coolant. An air bubble of around 1/2 inch to 1 inch was always left there. The method itself was a lot easier and probably quicker to do than what people have reported of their refill procedures/struggles, but the end result apparently was no better.

Well, I can report a partial success.

It's the single pump version alright and I think it was easier to work with. I got air bled out of the system and rebuilt it today. Bleeding was rather easy, actually. However, I seem to have broken the pump by accidentally applying 12V to the tach pin; the pump pumps fine, too fine, since it's going at full rpm all the time. Fails the ASD test with "speed high" or something like that.

Otherwise the machine is working great. All CPU temps in the low 40s, at "highest" cpu setting, and all the other fans stay at min. We'll see whether it stays that way..

I wrote a work log of the project mainly for myself and, in hopes of it being useful to others, sanitized it and posted it online at

http://www.saunalahti.fi/~holindho/g5-quad-v1-lcs-repair.html

I will update it with photos and other info as time permits.