Working on the G5 quad liquid cooling system

I have a G5 quad with the one pump Delphi LCS ("version 1"). I'm guessing the LCS is clogged because when the machine sits idle (even in reduced CPU setting, which cuts the GHz in half), the fans and the pump slowly, over several hours, go to full speed. Temperatures of CPU B, particularly the second core, shoot to near 100C when loading the CPUs. ASD 2.6.3 usually passes, but sometimes ends in checkstop and overtemp. A local mac service center, which I had the machine checked at (in hopes of getting free repairs because of a leak), said it's fine.

As a simple first measure, I tried replacing the thermal paste on the CPUs, with little effect. The LCS didn't seem to be leaking or corroded anywhere, but when handling it, it sometimes made a bubbling noise, so, there's probably air in there. I thought I'd take the LCS apart next, but before that I have some questions which I hope someone can answer.

Which one, the upper or lower, is CPU A and which is B? I've read the machine runs on one CPU just fine, but which one?

If I switch the CPUs the other way around and later switch back, does something in the nvram reset so that thermal calibration needs to be run? I'd rather not run it as it seems to have mixed success.

What material/size are the CPU block O-rings on this system?

If I use a vacuum pump on the service valve (I gather it's of the "r134a high" type(?)), what would be a safe level of vacuum? Has anyone tried using the manual brake kits, or car cooling system refill kits that use compressed air and venturi valves, on the LCS?

I suppose I'm just going to try attaching a hose as a reservoir to the service valve and fill through that first, but if that doesn't bring satisfactory results, I'll level-up to some sort of vacuum method.

PowerMac

Posted on Sep 17, 2011 11:59 PM

252 replies

Fipps

Level 1

0 points

Mar 3, 2012 4:17 PM in response to BMaverick

Yes. I have seen this yellow Locktite stuff on the backside where the screws come out. This was one of the reasons for me not to mess too much with them. But i tend to think coolant throughput is ok right now. And blocking one or the other intake shows even flow at both blocks. So not having opened them shouldn't be too bad.

Mar 3, 2012 10:18 PM in response to Fipps

Fipps, assuming the pump doesn't contain air (sounds even), the throughput looks a bit weak. Have you tried to let the blocks soak for while in hot water (almost too hot to touch is imo good)? That is, position the blocks at the lowest point, pour some water in and let soak for say 10 min. Adding a bit of dishwashing soap probably wouldn't hurt, but it has to be thoroughly rinsed out. I suggest this because basically the inlet filters were the only thing that needed cleaning and soaking is what I did. I didn't have small enought brushes to scrape them or anything, so having the hoses attached doesn't drastically change the situation. Flushing them in reverse is of course the only direction crud will come out. Light tapping on the inlet barbs might also help while soaking.

I'd go for the Thermochill coolant: (1) it has worked well so far, (2) I'm lazy, and (3) automotive coolants would yield several litres of coolant when the project only needs a bit over 200 ml for what might well be the lifetime of the rest of the components.

The yellow threadlocker adhesive cracked open with moderate force. Threads didn't get damaged. A quick peek at Loctite 271 warns that a less favorable outcome is also possible.

By the way, is your pump a DDC-2B-VC as well?

Fipps

Level 1

0 points

Mar 5, 2012 10:48 AM in response to Heikki Lindholm

😐 So i'm actually giving the thing another few one by one block reverse flushing sessions with 2:1 vinagre-water mixture and hope it cleans up a little more...

...rather than trying to open it again. I hate having these phillipps-screws in delicate places. If they are hard to open, even a very good fitting tool tries to "turn out" at some point, most of the time seriously damaging the threads in a way that only leaves destructive opening methods in the end.

As for the pump, i don't know, if it's the 2B version. But it has VC. I have taken some pictures. Not the quality, your’s are. But they should get you some more information. One of it taken apart and a detail view of the pumpwheel in wich you can still see some remains of the white buildup in the chambers. Sticker on the PCB is 3.3. Also it seems to be 10 watts instead of 18, estimated for the quad at some postings elsewhere.

For the coolant, i think i will give Clysantin a try. I hope it to be a bit closer to Sierra / what was in there before than a usual PC-WC coolant. And having a bit more than i need might not be so bad, to be sure to get all of the air out. As i get it of your linked site, filling the LCS as "airless" as possible is ok and pre-pressuring is not neccessary? I'm asking this, because there seemed to be a little pressure in the system, when i opened it.

Mar 5, 2012 12:57 PM in response to Fipps

Much better lighting in your photos than mine. Like it says on the pump, it is the DDC 3.1 version. Looks like all DDC versions of the time got used in the G5 LCS systems. The 3.1 version is quite a bit weaker than the DDC 2, but also more reliable, I'm told. You're in luck there.

A wrong size philips screwdriver wouldn't have worked. Opening the screws did require enough force to very likely strip the heads.

As you're probably not going to preheat the coolant when filling, not pressurising the system should be fine, provided that you get all the air out. Pressurising, that is, letting the VC chamber fill a bit (watch the springs), should give a longer service interval by having some reserve for liquid evaporation. However, I wouldn't fill it up as far as it goes. I was projecting for a rather short service interval, so that evaporation wouldn't play much of a role and so wasn't really accounted for.

Fipps

Level 1

0 points

Mar 6, 2012 8:38 AM in response to Heikki Lindholm

Heikki, what made you recommend the Glysantin G48 coolant in particular? As the toxic warning on the product says, it's a usual ethylen glycol (and not propylen glycol) based mixture.

If possible interference with the old coolant is the only critical point, this should not be of any matter in my case, as i really flushed the system for hours and there should be nothing left in there. But if there's something else, that coul cause problems, using EG, i'd prefer the Thermochill too.

Mar 6, 2012 12:15 PM in response to Fipps

Glysantin G48 is recommended all over for its heavy duty corrosion inhibition package. The G5 LCS are always mixed metal loops and thus need good corrosion inhibition. That's pretty much why I mentioned it.

One of the reasons I chose Thermochill myself was that it's non-toxic and propylene glycol based. The ethylene glycol in G48 is more on the toxic side.

Thermochill was also the only PC coolant that claimed to meet automotive standards. However, I could find no corrosion test data. For automotive coolants the test data is usually available for easy comparison (for example, the G48 is much more efficient than the standards require).

Fipps

Level 1

0 points

Mar 6, 2012 5:29 PM in response to Heikki Lindholm

Thanks, Heikki, for confirming, i made a good choice! Also the G48 is explicitly recommended for vintage cars. So it may be good for vintage computers too.

I have refilled the system today. My setup was not as professional as your’s but seems to have done the job. I used a loop of clear pvc-hose, which could be opened at the highest point, and a lot of shaking, running the pump, releasing bubbles over and over again until there where no more. Then i quickly reattached the upper hose to the radiator. Now the pumpe runs smooth and barely audible again and there’s no sound of moving bubbles, when i shake the LCS in hands. I hope, it’s okay.

Tomorrow i’m gonna reinstall the CPU unit, after letting it sit outside the box overnight to see, there’s no leaking.

Mar 6, 2012 7:54 PM in response to Fipps

Ouch! clear PVC tubing.

Please change that grade of tubing before your loop becomes cloudy and plasticizer problems become a major problem in the LCS loop. See the link for what you are in for.

http://forums.extremeoverclocking.com/attachment.php?attachmentid=162207&d=13287 27527

A Tygon plasticizer free formula tubing works best. OR if Tygon is just too costly, Newage Technolgies has the Ether-based Superthane that is clear as well.

Mar 6, 2012 8:02 PM in response to BMaverick

MSDS Thermochill EC6 ...

http://www.thermochill.com/msds/thermochill-ec6-msds.pdf

Do not use this stuff with zinc. Otherwise it's mostly safe overall. 🙂

http://www.sidewindercomputers.com/thec6noncoco.html

Fipps

Level 1

0 points

Mar 7, 2012 2:02 AM in response to BMaverick

"Ouch! clear PVC tubing."

I only had this for refilling and getting the air out of the system to have an extended loop which could easily be opened at the highest point. The working loop are the original rubber hoses used by Apple.

Mar 7, 2012 2:12 AM in response to Fipps

The thing with plastic tubing is that once put in contact with hot enough water (or any medium), the plasticizers just jump right out and coat the whole loop with a thin film. Might not be a total disaster, but then again the G5 LCS is small system so even a small amount of contaminants might have a bigger effect.

Fipps

Level 1

0 points

Mar 7, 2012 12:36 PM in response to Heikki Lindholm

I have it all back working now!

First quick conclusion: Maybe, i should have tried a bit harder to open the CPU-blocks and have cleaned them. 😐

It’s a lot better, than it was before. Machine was barely usable then, with temps rising and fans throttling up at even the smallest tasks. But temps are still „responding“ a bit to quick to my taste and also the difference between the two CPUs, in particular under load, worries me a bit.

Here are some Hardware Monitor read outs: idling and rendering CineBench on all CPUs.

Anyway, thanks for your help so far!

@BMaverick: At least right now the pump (even with the speed-control) seems to work flawless. But for the future it could be nice to know, where to get a replacement. Do you still have some in stock or would it be better to make a quick order? 😉

Mar 7, 2012 9:39 PM in response to Fipps

The disparity between CPUs does not seem that big, but between the cores it is a bit on the high side. That could be the inside of the cooling block, but more likely a bad mount or uneven thermal grease application. The temps and fan speeds are high, both idling and loaded, though it's hard to make comparisons when the fan speeds are controlled by the OS --- which is why I put together I boot cd to measure temps.

The old DDC-1 pumps can be powered by the G5 Quad, but afaik are not really compatible, so the end result would the same as using any current Lanig DDC pump, that is, pump going full speed all the time. The old DDC-1 pumps have incompatible tacho output (6 pulses per rev) and the current DDC pumps have incompatible speed control scheme. Basically, a replacement pump for the Quad should be the same DDC-3.1 (or DDC-3.2) version that you have.

Fipps

Level 1

0 points

Mar 8, 2012 1:10 PM in response to Heikki Lindholm

After a day of working on the Quad most of the first impressions are (sadly) confirmed. Idling is ok, but at work, the machine is still too loud and fans are still too "responsive" to me.

One question concerning yor (very low temperatures, Heikki: are you still using the pump with the defective tacho signal? Can the pump running always at full speed (wich should be 3600rpm if HardwareMonitor is right) be one of the reasons for your values? In my case the pump is dramatically downthrottled when there is no load. The lowest i have seen was 1200. But temps stay in the high 30s. The lowest readings i get in the delay when the CPU load is finished before fans and pump spin down. That's when i get low 30s too.

In all, i tend to reopen the LCS, buy some very hig end exact fitting Phillipps tool and give those CPU blocks another try. I hate work only half done, and i doubt, thermal grease application could be the reason. I used Arctic Silver Ceramique, which gave me the best results, working on vid cards. and application should be as even as the mount of the 4 spring loaded screws. I tightened them by half turns in a cross pattern.

Mar 8, 2012 6:03 PM in response to Fipps

The thermal paste used on the LCS blocks is Shin-Etsu MicroSI X23-7762. The formula works well under higher thermal loads than the AS-C. The X23 was chosen since it is non-conductive.

When ordering the X23, make sure the tube is within 6 months of manufacture for the best results. To store, place in a clean coffee mug in the back of your fridge with the applicator tip facing down. This will extend the shelf life many more months.

When applying the thermal paste, do a dot-by-dot matrix application on the block surface. Do not place a glob in the center to smear outward. A typical pattern would look like this: http://img20.imageshack.us/img20/3466/pastespots.jpg Each dot must be a miniscual dab.

This thread has been closed by the system or the community team. You may vote for any posts you find helpful, or search the Community for additional answers.

Working on the G5 quad liquid cooling system