Air Cooling a Late 2005 G5 Quad

[mac57mac57]

Regular readers of the MacRumors' PowerPC forum will know that I have struggled mightily over the last two plus months to overhaul the failing LCS of a conventional Power Mac G5 Quad. I have been mostly successful: CPU temps are now wonderfully controlled (CPU A idles in the high 30s, CPU B idles in the low 40s) but fan speeds are still higher than I wanted, now idling in the 2800 RPM range. This still creates enough noise that the machine is not tenable (to me at least) for routine work.

Getting to this point has taken both an enormous amount of work and an enormous amount of frustration as I battled a nearly continual series of leaks that I had to hunt down and seal up. In addition, I had to tear the machine down to its most elemental pieces to get the Liquid Cooling System (LCS) back to the point where it would accomplish adequate cooling. It was highly detailed, incredibly finicky and very error prone work... and when I was done, while CPU temps are now great, fan speeds are still not, and honestly I have no hope of ever correcting that; my last gasp plan is to replace all the stock Apple CPU intake and exit fans with super quiet ones and hope that the high fan revs will no longer be an auditory issue.

HOWEVER, I started with two Quads that had failing LCS units and after the experience of the first one I have no intention of attempting an LCS overhaul on the second one. I have had it with liquid cooling... too darn finicky and unpredictable. So what is the plan for the second Quad? Air Cooling!

I will be building on the excellent work that @Doq has done and documented - see his thread:

Because I Can: Air-modding a Quad With... PC Coolers!?

I plan to attempt pretty much the same modification that he has, hopefully with similarly successful results.

As I start, here is the Quad to be modified into an "Air Quad", with the LCS/CPU assembly extracted from the machine and torn down to its elemental parts:

Here is a shot of the five new parts needed to accomplish the air cooling modification (two coolers, two copper shims and one tube of thermal paste, almost identically the parts recommended by @Doq):

The FROZN units are PC (yes, PC) air coolers and will replace the LCS. The copper shims (the small blue box in the center contains 4 such shims - I will only need two) will be used as intermediaries between the CPU chips and the FROZN A400 coolers. Finally, the Kryonaut Extreme thermal paste will mate the CPUs to the shims, and the shims to the coolers. Kryonaut Extreme is the most highly rated thermal paste I could find (thermal rating of 14.2 W/mK) but it is noted for being difficult to spread, so it may be an issue yet.

I purchased the two FROZN units on eBay, while the copper shims and the Kryonaut Extreme thermal paste were purchased at Amazon.com. I had to buy the FROZN units on eBay because they do not appear to be for sale new anymore, except for "new old stock" on eBay, having been replaced with the larger (120mm) FROZN A410. Two of these will not fit into a Quad - each one must be a maximum of 92mm to fit.

So here we go again. Like my LCS overhaul, I plan to photograph everything extensively as I go, and when done write a full blown, highly detailed and heavily illustrated guide to accomplishing air cooling of a Late 2005 Power Mac G5 Quad, so that others can replicate this result simply by following a detailed set of step-by-step instructions.

In this effort, I have to thank @Doq once more - his post/guide points the way, provided me with enough critical information to feel confident in trying this modification. Lets hope that this effort yields both cool CPUs and quiet fans. We shall see!

 

[joevt]

Regarding the copper shims, https://forums.macrumors.com/thread...-a-quad-with-pc-coolers.2427612/post-33168681 says the CPUs will thermal stop without them.

The post at https://forums.macrumors.com/thread...-a-quad-with-pc-coolers.2427612/post-33863463 says the shims might not be necessary depending on the cooler. Specifically, if the cooler has a solid copper plate then the shim is probably not necessary.

Do the FROZN A400 not have a solid copper plate? What does the bottom of those look like?

 

[ADunsmuir]

The FROZN A400 were still available on Amazon.ca this afternoon, at $C39 with free shipping. This was significantly less than current units available on eBay.ca so I ordered 2 just in case radical surgery on my Quad should become necessary in the future.

 

[Doq]

Do the FROZN A400 not have a solid copper plate? What does the bottom of those look like?

The bottom is the exposed heatpipes surrounded by aluminium. The die is too small to contact all of the heatpipes at once, thus requiring the shim.

It's best practice to use a shim regardless when working with direct-die cooling but I understand also that it may not be strictly required if the contact point is solid.

The FROZN A400 were still available on Amazon.ca this afternoon, at $C39 with free shipping.

I just checked and A400s are also still available on Amazon US (and on sale a bit).

 

[joevt]

The bottom is the exposed heatpipes surrounded by aluminium. The die is too small to contact all of the heatpipes at once, thus requiring the shim.

It's best practice to use a shim regardless when working with direct-die cooling but I understand also that it may not be strictly required if the contact point is solid.

Oh, right. That makes sense. AMD and Intel CPUs usually have their die covered by a heat spreader. This is the function of the copper shim for the PowerPC CPU.

 

[mac57mac57]

Interesting. When I type this in to Amazon.com, I get pages and pages of A410's! I'll try again.

 

[mac57mac57]

... this time, I got ONE A400 entry, priced at $24.99! The rest were all A410 or higher.

Darn! I paid nearly $40 each for the two A400's I purchased on eBay! 😕

 

[jktwice]

We’ve seen air cooled quads happen using modified 2.3GHz coolers, but obviously this is not sustainable.

Hopefully your efforts here bear fruit because then the rest of us Quad owners can keep these machines alive and happy for just a bit longer.

 

[mac57mac57]

Thanks @jktwice. I hope so too!

We know that it CAN be done - @Doq has already done it. The question is whether it is a repeatable procedure and whether I can document it in such a way that anyone with PowerPC Mac skills can accomplish it themselves.

The effort is well worth it. Equipped with Sorbet Leopard and the latest Aquafox browser, these machines are still fully functional in today's world. They can surf the web, handle today's web-heavy emails, do your banking, balance your budget, edit photos and videos, etc. Apple released an incredible beast of a machine in the G5 Quad. 20 years later, it is still a viable machine.

 

[mac57mac57]

@Doq, in your work to air cool a Quad, do the copper shims sit directly on top of the CPU? I would assume they do, but that would place the edges of the shim in contact with a whole variety of components directly around the CPU. Granted, these are the cases of the components, not their electrical connections.

I was thinking of cutting out a shaped-for-the-purpose piece of thin non conductive material and placing that around the CPU chip just to be safe, with the entire surface of the chip exposed, but the rest of the area under the shim "insulated" with this material. Even paper should work, since its ignition point is above 200 C.

At any rate, I just wanted to check whether you took any precautions to ensure that no electrical contact occurred - copper is very conductive after all.

Thoughts?

 

[mac57mac57]

Today was not a "do the air cool work" day. It was a "figure out the mechanical alignment and connection procedures" day.

To that end, a few photos. First, here are all the parts that come in the A400 box:

There is the unit itself, its fan (provided in a separate box-wthin-the-box), and a whole box of mounting hardware, split between the left and the right of the photo above, which supports both Intel and AMD processors and motherboards. The Intel-oriented hardware seems the right fit for the G5 Quad.

Also included in the A400 box is a tube of pretty darn good thermal paste - 10.5 W/mK! I will stick with my Kryonaut Extreme, which is rated at 14.2 W/mK, but it is nice to have a fallback if needed. Both tubes are relatively small. If, like the LCS overhaul, this needs to be disassembled, fixed and reassembled a few times, having extra thermal paste on hand will be helpful.

Next, here is a photo of what the underside of the A400 looks like. Once you see this, it is clear why the copper shims are needed. They overcome the facts that (a) the copper on the underside of the A400 is not continuous, and (b) the cooling surface of the A400 is significantly larger that the 970MP chip surface. The shims overcome both. The part of the shim in contact with the CPU face picks up heat from it, then effectively spreads that heat over the larger surface area of the shim, from which the full cooling surface of the A400 can extract it.

I have placed a shim directly underneath the underside of the A400 in the photo below:

The fan must be attached to the A400 after it has been mounted and screwed down. A set of "fan clips" are provided to accomplish the attachment. The photo below shows how they are used to connect these two parts - this does require a little bit of mechanical pressure to stretch the fan clips down to fit into the slots on the sides of the A400:

By the way, which side of the fan faces out is a pure guess, based on the very small photo provided with the instruction manual.

Finally, this is a photo of the required alignment of the CPU card, the copper shim, the mounting hardware, and the A400:

The final photo below shows how the A400 will fit down onto the mounting hardware, allowing it to be screwed into place, once the mounting hardware (which is just sitting on the CPU card in this photo) is itself screwed down:

I am missing some non-conductive washers to build up a small gap between the CPU card and the mounting hardware piece closest to the CPU card's existing radiator, and so that is all I can accomplish today. Once I get these washers, more progress can occur. I actually don't see this initial assembly taking too long, as I look at how everything fits together. This will not be like putting together a few Lego bricks, but it looks SO much simpler than an LCS overhaul!!

 

[Doq]

@Doq, in your work to air cool a Quad, do the copper shims sit directly on top of the CPU? I would assume they do, but that would place the edges of the shim in contact with a whole variety of components directly around the CPU. Granted, these are the cases of the components, not their electrical connections.

Yes, but the CPU die clears the surrounding components (it'd have to to accommodate the stock coolers, which were also copper, and have big ol' plates around it).

The fan must be attached to the A400 after it has been mounted and screwed down. A set of "fan clips" are provided to accomplish the attachment. The photo below shows how they are used to connect these two parts - this does require a little bit of mechanical pressure to stretch the fan clips down to fit into the slots on the sides of the A400:

Watch out. There's not enough width clearance for the inside clips to attach the fans without introducing stress to the processor cards and reducing cooling enough to produce a checkstop. This is why I had the center twist ties originally and later the 3D print. The outside clips still work though.

In order to make this work, you'll need to assemble and install both processor cards simultaneously as a single unit (much like with the stock cooler, but way more fiddly as it's not fully attached on all sides.

Now that I think about it, you could probably get away with using both clips by installing one of the coolers in reverse (staggered configuration), though I don't remember offhandedly if you can install the fan on either side.

By the way, which side of the fan faces out is a pure guess, based on the very small photo provided with the instruction manual.

Generally, the side with the logo is the direction that air flows from, with the power rating sticker on the reverse for the direction air flows to.

I am missing some non-conductive washers to build up a small gap between the CPU card and the mounting hardware piece closest to the CPU card's existing radiator, and so that is all I can accomplish today.

I knew this was going to happen and I was going to point it out but you beat me to it.

I do notice that some of the stock pieces are also not shown. I'm interested to see what your solution for that will be if you go with something different in that regard.

 

[mac57mac57]

OK! I think I have solved all of the "mechanical engineering" problems associated with air cooling my Quad. I can now assemble it and try it out. I have no guarantee that it will cool adequately, but the example of @Doq suggests that it will. So, now I just need an uninterrupted 4+ hours stretch of time to do the work and try it out. Up to now, I have simply been testing fastening techniques, verifying that everything will fit together spatially as expected and generally trying to anticipate possible problems before they crop up.

I am being super methodical. I want to do this once and do it right the first time. It took two months of grueling labor to accomplish the LCS overhaul on my other Quad, during which time I was tripped up by one problem after another after another. This time, I want to have everything worked out in advance, so that the only possible issues are related to how well the setup cools the Quad, not whether I can even manage to get the solution together!

 

[jktwice]

OK! I think I have solved all of the "mechanical engineering" problems associated with air cooling my Quad. I can now assemble it and try it out. I have no guarantee that it will cool adequately, but the example of @Doq suggests that it will. So, now I just need an uninterrupted 4+ hours stretch of time to do the work and try it out. Up to now, I have simply been testing fastening techniques, verifying that everything will fit together spatially as expected and generally trying to anticipate possible problems before they crop up.

I am being super methodical. I want to do this once and do it right the first time. It took two months of grueling labor to accomplish the LCS overhaul on my other Quad, during which time I was tripped up by one problem after another after another. This time, I want to have everything worked out in advance, so that the only possible issues are related to how well the setup cools the Quad, not whether I can even manage to get the solution together!

I'm very invested in a solution. I love my Quad. It edits my photos and that's that. I would love to keep this thing alive for at least another five years and air cooling imo is the way to go. Who cares about the noise it's a retro supercar in my eyes.

 

[mac57mac57]

More progress today. I did my first test assembly this afternoon (no thermal paste... just checking that everything fits together spatially as planned). I am pleased to report that it all does, but clearances from the sides of the Quad's case are miniscule! Connecting the FROZN fans to the FROZN heat sink units was challenging as a result.

A consequence of this was that I needed a new specialized tool to screw down the FROZN heat sinks onto their front mounting brackets - there is almost no vertical clearance, so a conventional screw driver would not work - but after a thorough rummage through my various tool boxes, I found what I needed.

There is one last technical detail to take care of - getting power to the FROZN fans. My kingdom for a spare MOLEX connector, but there are no (zero, nada, none) MOLEX connectors in evidence anywhere inside the Quad case.

But all is not lost. Per @Doq, since the LCS doesn't need it any more, I can reuse the LCS pump connector, which has all the necessary voltages available. @Doq even provided a listing of the relevant pins on that connector! Thanks @Doq!

To power the fans therefore, I need to cut into the LCS pump connector and splice the Ground, +5v and +12v lines onto the matching FROZN fan power lines. As a one-time electrical engineer, this is child's play, hence the very low priority I have accorded it.

I will be checking my supply bins as I do this. I think that I have a rheostat that I can spice into the +12v line, allowing me to directly control the FROZN fan speeds by turning a simple dial!

The rest of this week is looking pretty jammed up with external commitments, so it may be the weekend before I can power the fans and try my first "production" build and test.

Air Quad Rev 1 - here we come!

 

[mac57mac57]

BTW, here is what it looks like after the first trial assembly of all the parts needed for air cooling:

You can clearly see how tight the clearances are between the two FROZN units and between the bottom FROZN unit and the case of the Quad.

Those tight clearances make attaching the fans to the FROZN units a real challenge. For the test assembly shown above, this was accomplished using a judiciously chosen combination of the clips provided for the purpose and twist ties (which are the "secret sauce" of the whole solution).

 

[Doq]

Looks like your fans are backwards-- that'll affect cooling significantly.

The tight clearances were part of the reason why I went with cooler and processor as a single unit in my build, rather than attaching the coolers later, but it looks like it's working out for you nonetheless.

 

[mac57mac57]

Generally, the side with the logo is the direction that air flows from, with the power rating sticker on the reverse for the direction air flows to

If logo'd side of the fan is the side that air flows from (i.e. out of) then I believe my chosen fan direction is correct. I want to be pushing cool air across the FROZN units, towards the back of the Quad's case, where the exit fans will guide it out of the case.

If I did it the other way, the FROZN fans would be pushing air towards the front of the case, where the intake fans would just push it back.

Am I missing something obvious?

 

[mac57mac57]

After a careful review of my photos (I am not home, so I can't look at the real thing) it appears that air flows into the logo'd side and out of the other side. Hence I have to agree... I need to reverse the orientation of the fans. Thanks for the heads up @Doq!

 

[mac57mac57]

I found some extra time today, so I made a little more progress towards Air Quad Rev 1. Today, I did my first (and hopefully only) "production" placement of the FROZN A400 units and their fans into the G5 Quad case, with Thermal Grizzly Kryonaut Extreme thermal paste used throughout.

Everything you have read about Kryonaut Extreme is more or less correct, but it does not have the consistency of clay, which has been reported by some users. It is actually rather too fluid in my opinion, but once out of the tube, it sticks together in globs and is difficult to spread evenly, particularly on wide surfaces like the 970MP chip face. For the copper heat tube lines on the bottom of the A400 it was perfect though. The tube of Kryonaut Extreme comes with a screw on spreading head that affixes to the top of the tube. By applying gentle pressure to the plunger of the tube as you move the spreading head down the copper heat tube lines on the bottom of the A400, doing each line separately, the paste spreads out evenly and nicely, because those copper tube lines just happen to be almost exactly the same width as the tube's spreading head. What is going on here? Did I just catch a break? Is Murphy "asleep at the switch" today?

So, Kryonaut Extreme is difficult to spread nicely on a large flat surface, but it lays out very easily in thin(ish) lines. The CPU temps when this is all said and done will be the real test - does it do the job as well as the Noctua NT-H1 paste I have been using to date? That paste has a thermal conductivity of 12.8 W/mK, while the Kryonaut Extreme has a thermal conductivity of 14.2 W/mK. Whether that extra 1.4 W/mK is worth it or not will be seen when the system is fired up. Certainly, the Noctua NT-H1 is a great paste, easy to spread evenly on pretty much any surface, and possessed of a very good thermal conductivity rating.

OK, Murphy wasn't entirely asleep... as I screwed down the first A400, the one closest to the bottom of the case, I forgot to add a fan clip! Then I screwed down the second A400, and there was no way to go back and add in the missing clip - thermal paste was already perfectly in place and I did not want to mess with it. What to do???

Well, when the going gets tough, the tough use duct tape! ... and that is exactly what I did. To be specific, I used a brand of construction tape called Tuck Tape, which has incredible stick. This supported the bottommost fan from the top. It already had some support from the bottom - the two fans are twist tied together on their bottom facing edges, per @Doq's approach. For the uppermost fan, I was able to clip it on as intended, but I also added a strip of Tuck Tape, just for symmetry... a little artistry perhaps!

Here is a photo of the result:

All that is left now is to power those fans and I can move into testing of this new cooling solution. Originally, I had planned to simply cut the headers off each fan's power line, cut the header off of the LCS pump control line, and splice the two together, providing power to the fans from the pump control line.

Then I got to thinking... I mentioned a rheostat in my last post, but after rummaging trough my "spare parts bin", I had nothing that I could use for this purpose. Still, I like the idea of being able to control the fan speeds directly so much that I started hunting for solutions on Amazon and eBay. I found this and ordered it for delivery on Saturday:

This may look like a strange solution, and perhaps it is, but it meets the need of allowing me to directly control the A400 fan speeds. It also eliminates all cutting and splicing of power lines. Fan power now comes from a household AC outlet, which plugs into the power plugs of each fan via the above setup. I will thread the fan header lines from the above through a spare expansion slot whose cover has been removed and simply plug the above setup in. With this, the fans are both powered and controlled in an entirely standalone way. Assuming this all works well, it will take two wall plugs to power the complete Quad, but that is such a minor "penalty" that I am not worried about it at all. The upside is that I will be able to directly control the speed of the A400 fans - such a joy!

I found one other fan speed control solution that I almost went with, and here it is:

This one is really nice - four independent speed controls for up to four fans, support for both 3-pin and 4-pin fan connectors and requires simply the space of one expansion slot of the Quad. This is really nice and really seamless, but it gets its power from an industry-standard 4 pin MOLEX connector. Since there are no MOLEX connectors to be found anywhere inside the Power Mac G5 Quad case, to use this solution I would still have to slice up the LCS fan control line and manufacture a MOLEX connector. This is not hard to do, but it is extra work vs. the solution I chose. If this whole "air cooling" solution works well, I may repeat it with my other Quad, and at that time, I may opt for the above solution instead of the one I am using this time around.

The AC solution I have ordered (above) arrives Saturday, so once again work on this project goes on hold for a few days while I wait for it to be delivered.

 

[Certificate of Excellence]

Does your quad not have a molex connector in the dvd bay? Iirc you should have a black molex cable there and can split off and run a molex lead down through where the dvd ribbon cable drops down to the lobo.

 

[mac57mac57]

Does your quad not have a molex connector in the dvd bay?

You know... I was thinking the same thing late yesterday. The DVD drive looks to be just a normal DVD drive, so it should have a MOLEX. If it does, I can fork it to power cooling solutions.

I'll look into this later today and post back.

 

[mac57mac57]

Sure enough, there was an industry standard 4-pin MOLEX connector providing power to the DVD/CD drive. I have plenty of MOLEX splitter cables, so I unplugged the DVD/CD drive's MOLEX, split it, and ran the free side of the split through the large opening at the bottom of the PCI Divider plate. This is the same opening that the DVD/CD drive's IDE cable is pulled through to get to the motherboard, so the opening is large enough to pass the head of a MOLEX connector through it easily.

The immediate area around the DVD/CD drive is a really tight space to work in, and without two different pairs of long handled pliers, I would have never been able to disconnect or reconnect any of these cables. You can see the key set of pliers at the very bottom of the photo below.

This excellent progress was amplified by a great piece of luck. Late yesterday, after more rummaging through my spare parts stock, I managed to turn up all the pieces I needed to create a MOLEX to dual-ended 4-pin fan cable. I built said cable and then hooked up the MOLEX end of this new cable to the free MOLEX connector I had just pulled through from the DVD/CD drive area. I then connected each of the 4-pin fan connector ends of the cable to each of the FROZN A400 fans.The result was a fully powered set of FROZN A400 fans!

With the entire air-cooling solution installed and powered, I tidied up the loose cables and then reassembled the Quad. Less than two weeks after starting this effort, I am now ready for the first test of my newly air-cooled G5 Quad!

The result looks like this:

The white patch on the Apple fan unit at the front of the Quad is an illusion... It is just a reflection of the high powered bench light that I use to illuminate the work area.

So... I am now ready to test... unfortunately, I am also out of time for this morning and have to stop here, which is somewhat frustrating. Hopefully later today I will be able to fire it up and see if it works.

 

[joevt]

Looks like you could add a couple fans on the rear side of the coolers. I know the G5 still has its exhaust fans. I wonder if additional fans would help.

 

[mac57mac57]

Thanks @joevt, that could be helpful, but let's see how the base system performs first!