The 1.6ghz Sonnet Encore MDX overclocking project

[redstudio]

thanks for your work knezzen!

[redstudio]

I confirm that replacing the Sonnet heatsink is the best way to immediately lower the temperature of the MDX, be careful; I discovered that the copper heatsink, the newer tube model leaves a processor half exposed! I noticed this because once disassembled the thermal paste came out of the copper bar where the processors rest, the traditional copper one is fine. about 4-5 degrees cooler immediately. updated to 1.67 very stable... waiting to overclock it up to 1.83 as shown in the photo.. (waiting for the resistor position diagram, thanks knezzen!)
question: are the resistors ohms? like a jumper or do they have a value? sorry for the ignorance...

[indibil]

Hello! How fun this post. I recently got a Sonnet Duet 2x1.83GHz processor for MDD and after reading this post and Project Nova, I decided to go up to 2GHz. But I come to give bad news.

At 1.83GHz the Dual processor is totally stable, but when it reached 2GHz it had problems, and sometimes the diagnosis indicated problems with the RAM modules, so I blamed them first.

Once the correct RAM modules were installed, I tried to install the original MDD aluminum heatsink, but unlike Project Nova, mine did not fit well, look at the attached photo with the thermal compound marks, only part of the processor. I show it to you so you can verify it, because bad contact can burn a processor.

After reinstalling the Sonnet cooler, I replaced the 50x50 fan with an 80x80 fan. And it seems that GeekBench worked well and with maximum temperatures of 54ºC.

But I decided to do a more intensive test. I ran GeekBench tests one after another, and always at most the third test the GeekBench would freeze and when I ran it again, it would take a long time or freeze again, and sometimes appears panic kernel. The temperature was beginning to approach 56ºC.

I think that the dual processor works at 2GHz but very at the limit, probably with insufficient voltage for stability, because perhaps some installed 7447s have more stability at 2GHz than others, perhaps mine do not, others perhaps do.

I would need the combination for 1.92GHz. Does anyone know her?

I simply wanted to provide these details to those planning the overclock.

[GaryN]

Hello! How fun this post. I recently got a Sonnet Duet 2x1.83GHz processor for MDD and after reading this post and Project Nova, I decided to go up to 2GHz. But I come to give bad news.  At 1.83GHz the Dual processor is totally stable, but when it reached 2GHz it had problems,

I went through this experience a few years ago. You are correct. The Sonnet will run all day long at 1.83 but is unstable in the MDD at 2.0
I don't think it's a "Sonnet fault". Based on the kernel panic readouts after a half-dozen or so crashes, I'm more inclined to believe that the CPU and the MDD date buss fumble with data transfer somehow, but no matter.
It also seems apparent (to me, at least) that Sonnet couldn't make it work either. I'm pretty sure they would rather have been able to have a great big "2.0GHz!!" on the front of the box if they could have been confident that the system would be reliable.

Note: as far as temperature, the heat sink and fan that comes with the Sonnet is designed to fit in a Xserve rack space. It's absolutely dependent on the little fan and overall, it's a poor fit in an MDD. I got better and quieter results with the "stainless" heat sink that comes on the 2x1.25GHz MDDs. That enabled me to use a more modern, quieter main fan and eliminate the little Sonnet one entirely.
That said, it's a significant improvement in noise level but only a minor difference in temperature.
Additionally, if you dual-boot, know that there's simply NO temperature monitor for OS9 that will read the 7447/Sonnet… period. So, while it does read in OSX, in OS9 you'll need to rig some kind of sensor and readout or just spit into the wind and trust that nothing irreparable happens without you knowing until it's too late. Considering the scarcity and cost of those MDX boards, I personally lean a little conservative here.

[Knezzen]

Just as a note; mine is still running stable at 2ghz in my MDD. It’s what I have used ever since I did the overclock. No issues at all.

[robespierre]

My memory says it was already discussed, but the posts must have been deleted:
Heatsinks are never made of "stainless". It conducts heat extremely poorly. What is really at play is chrome-plated aluminum, which has a mirror-like finish.

[indibil]

Just as a note; mine is still running stable at 2ghz in my MDD. It’s what I have used ever since I did the overclock. No issues at all.

We are not saying that it is not possible, simply that not all Sonnet Duets are capable of working stably at 2GHz. The chips are guaranteed at 1.6ghz according to the CPU text, but from there onwards not all will be capable of the same. It's only for people who try it and it doesn't work for them.

[Jubadub]

[...] perhaps some installed 7447s have more stability at 2GHz than others, perhaps mine do not, others perhaps do.

Correct. The probable reason yours and @GaryN's didn't work, and the probable reason @Knezzen's worked, is a matter of what we call Silicon Lottery. There is something else called Processor Binning. I found one of the MacRumors PPC threads I go on and on about those, and the custom 7448 dual 2.0GHz upgrades. To understand processor binning, and why it matters, refer to this comment, and about silicon lottery (plus a lot else), check this comment instead.

In short, @Knezzen won the lottery! But all of you may acquire and try more 7447 processors with the highest binning, until you win that same lottery. From experience, I will say your chances are very good with very few tries. Also keep in mind there are 3 main different MDD coolers (go for copper if you can, else the pure aluminium one. The other one reportedly freezes MDDs during stress, according to House of Moth, whose links describing this I no longer have), and they matter and differ a lot, but if you can LCS it with a silver base, that would be best, but not at all needed. I would also recommend, again, going for 7448 processors (if you find them) with the highest binning, and getting those instead, because the Sonnet MDX daughtercard can be used without requiring an interposer board, unlike with the stock MDD daughtercards. The 7448s also run cooler.

Anyway, be happy that you guys even have access to an MDD, let alone MDX daughtercards! I couldn't find the latter even when in Germany for 2 whole years of stalking eBay and the like. If you can find an MDX card, you can find A TON of 7448s! I know I did. Keep looking!

[aBc]

Here’s that House of Moth link:
https://thehouseofmoth.com/further-tweaking-the-cooling-part-1/

And an image “borrowed”’ from there as well.


House of Moth refers to the heatsink on the far right as the “metal heatsink” or “metal foil” and (perhaps like GaryN) I have always refered to them as “stainless”. Now whether or not the thin fins are chrome-plated aluminum or stainless steel, it seems that this heatsink ranks 2nd place effective for heat dispersion of the three shown above… with the copper heatsink being the best. (And there are also other variants of the aluminum (middle one) made specifically to accomodate different MDD processors.) Perhaps why indibil’s aluminum heatsink did not “seat” adequately?

A deep scratch test and a clip test of the “metal - finned” heatsink fins here revealed no visible underlying core of aluminum AND no magnet would stick (typical of both aluminum and some stainless steel). If I had to bet, I’d wager that these fins are stainless steel affixed to the thicker aluminum base. And especially after I have noticed rust on some of these fins in the past. Maybe chrome-plated aluminum also rusts?

                           

Also, see reddragon’s MacRumors posted pics of various different MDD aluminum heatsinks.
https://forums.macrumors.com/threads/mdd-heatsink-differences.2223970/?post=29952709#post-29952709

So, performance wise by the numbers, the copper heat sink still takes it home, the metal heat sink comes in second and the aluminum heat sink comes in last. Common knowledge is that the metal heat sink sucks, is useless, should be avoided etc. but these temperature readings show otherwise. What does make the metal heat sink a poor choice is the system instability it somehow causes. - House of Moth

Also from HoM:
Now keep in mind I am using graphite thermal pads which outperform Arctic Silver and even Thermal Grizzly Kryonaut by a wide margin, so that helps keep my temperatures down as well but the copper heat sink seems to be the way to go on any CPU.

Yes, heatsinks on my mind here today.

[GaryN]

My memory says it was already discussed, but the posts must have been deleted:
Heatsinks are never made of "stainless". It conducts heat extremely poorly. What is really at play is chrome-plated aluminum, which has a mirror-like finish.

Well, that's why the word is in "quotes", isn't it? It's only to differentiate it from the thick-finned aluminum one.

So, performance wise by the numbers, the copper heat sink still takes it home, the metal heat sink comes in second and the aluminum heat sink comes in last. Common knowledge is that the metal heat sink sucks, is useless, should be avoided etc. but these temperature readings show otherwise. What does make the metal heat sink a poor choice is the system instability it somehow causes. - House of Moth

It's been said before, but stuff gets lost or just unseen in a gazillion posts, so I'll repeat:

The copper sink with the heat pipes, although it's the most efficient of the three with original Apple 7450s, cannot, repeat cannot be used with the Sonnet MDX card!
The 7447's are NOT located in exactly the same place and only one of them will cool properly. The other one will overheat and die.
I too, tried all of the sinks and settled on the "stainless" for that reason. It was disappointing for sure, as I had previously had a 2x1.42 Xserve CPU card in there with the copper sink and it was clearly the coolest one…   It's a trap waiting to spring since you can't actually see they don't align properly when installing the sink. You can only tell if you pull it up again and notice the compound is partly undisturbed over one CPU.

As for the Moth saying the stainless causes some kind of mystery "instability", that is not my experience at all. I haven't had a crash or the smallest hiccup even once. I suspect that a lot of people with little or no knowledge of thermodynamics are also doing amateur fan changes and mods at the same time. The MDD was called the "wind tunnel" because that was how it was designed. It needed to move a LOT of air through a cluttered, tight space with the main fan and it ended up generating probably a LOT more noise than they would have liked but they were married to the case design and it was the last G4 built, using Xserve stuff, to fill the gap behind the G5 that was taking forever to finish… so long that they soon even had to add another stopgap model for LESS money, the FW800. It's never a good thing when you have to offer an entire new PSU and fan to your customers FOR FREE because you've sold them the noisiest home computer ever made.

What's equally as remarkable is that you would think after all that, the G5 would have been whisper-quiet. Oops…

[Jubadub]

@aBc Thanks for those links!

I really doubt House of Moth's claim on graphite thermal pads beating Thermal Grizzly Kryonaut (let alone the superior Kryonaut Extreme), but I can't say I tried them personally, so... *shrug* Who knows.

@GaryN That's a good point with regards to the processor height difference in the MDX, in terms of getting the copper heatsink to do its job properly. However, you should still be able to use it properly for both CPUs if you add copper shims (one or multiple). Using 7448 and 7457 processors, and/or interposer boards, instead of the usual 7447 or 7455 with MDX and stock daughtercards also leads to those kinds of height concerns. The opposite is also true: Washers/spacers also help with when the CPUs are too high for the heatsink. Quoting one of my linked comments:

If using any of the 3 stock heatsinks, use copper shim(s) for the 7457 to take into account its slightly lower height compared to the stock 7455, or, in the case of the 7448 that has additional height due to the interposer board, use washers/spacers between the posts and heatsink to raise the height at which the heatsink stands.

If you ever give this a try, let us know how it worked out with your MDX card.

[DieHard]

What's equally as remarkable is that you would think after all that, the G5 would have been whisper-quiet. Oops…

Yeah, not only "noisy" but some of them were liquid cooled and to add to Apple's incredible track-record of "Think different" "Design Different" and sometimes "Screw the customer over with a different design" the liquid-cooled G5s were dying left and right in music studios due to leaks... from the web...

    "I have a lab of 17 Dual 2.7Ghz G5s that are all now leaking their coolant- all at once. A significant number are dead in the water and the rest are just holding on right now. Symptoms are wide-ranged but will include fans spinning wildly, machines shutting down when they heat up, greenish liquid leaking from the case, and if you are able to look, crystalized liquid forming where the CPU meets the heatsink as well as corrosion of all the metal surrounding the CPU module. Eventually, the machines just stop working altogether necessitating a replacement of CPU, Logic Board, Power Supply, and two smaller parts. In one case the power supply started to shoot off black smoke and then died.

    "As of yet, Apple has been fairly unresponsive in fixing or replacing them and I am now working on making this more public. I am curious to know if any other people are having this problem. I know of 3 other cases outside our own lab here. The machines in our lab were the first of the dual 2.7s- they were bought right when the came out."

I can tell you as an Apple-tech in the field in 2005 thru 2007, it was mind blowing to see $3500+ music studio and graphics studio machines with blown logic boards, and for added insult, the power supply was at the bottom, so, 4 out of 5 dead G5s had blown power supplies also; I can tell you this was NOT a "one-off"... I am just one guy, and I personally inspected a few dozen that leaked !  Man it was ugly... white liquid trails running into CPU sockets, blown logic boards, and then a small trail going directly into the Power supply, I almost didn't have the heart to show the client that forked over big bucks, the absolute destruction of their workhorse from just being used like a computer should be.  Many were clean as a whistle with no dust and very little use and BAMMM, death.  As irony and Apple go hand in hand, the slower and cheaper 2.3 was still air cooled and there are probably a lot of them still alive today, and also the single CPU models, but the most of the dual 2.5 or 2.7 units died in epic fashion.

This is probably Apple's worst "screw-over" to date, by never officially making a re-call... and just saying in 2006/2007 at the very height of the number of leak failures, "Hey sweet client, this new intel Mac pro is better than those shitty G5s in about 10 ways, so order now"

Talk about tangling a carrot, and wanting everyone to move on... I can say, that if you weigh acoustics, expand ability, reliability, and power, they finally got it right with the cheese-grater Mac Pro; it's just a shame so many valued customers had to pay a ton of cash for Apple's own G5 mistakes before the Mac Pro.

[GaryN]

@GaryN That's a good point with regards to the processor height difference in the MDX, in terms of getting the copper heatsink to do its job properly. However, you should still be able to use it properly for both CPUs if you add copper shims (one or multiple). Using 7448 and 7457 processors, and/or interposer boards, instead of the usual 7447 or 7455 with MDX and stock daughtercards also leads to those kinds of height concerns. The opposite is also true: Washers/spacers also help with when the CPUs are too high for the heatsink. Quoting one of my linked comments:

I'm afraid you misunderstand the problem. Height has nothing to do with it.

The procs on the MDX board are in a different position than the Apple boards. If you look at the pic, you can see the contact surface is just large enought to cover the procs and transfer the heat up the pipes.
The white rectangles are the location of the original 7450s on the Apple boards. The red are the 7447s on the MDX. The problem is obvious.

(Nitpickers: Don't hold me to absolute exact-ness here. I'm not 100% certain that this is exactly where the procs align. This is only from memory but it clearly makes the point. I would have to pull the MDX out of my MDD to check precisely where they land and I'm not going to do that.)

[Jubadub]

@GaryN I see, so you meant horizontal alignment, and I assumed you meant vertical. Yeah, now that is a problem. I suspect even Sonnet didn't realize the copper heatsinks were a thing early enough in time or, if they were aware, they assumed the alignment and all would be the same for all the heatsinks. In short, what most likely happened is that Sonnet got Apple'd!

I keep thinking there must be something we could do about this, but I guess that would be ugly, nasty and convoluted...

Y'know, it might be worth a look to see if you can perhaps find a 3rd party copper heatsink that will fit and also not have this issue. Some people tried that, and I don't know if they succeeded. I know I tried years ago, then gave up and went for the Apple one (as I wasn't affected by the MDX problem since I never got one). Or, probably better, it also seems like liquid-cooling might be your best friend to address that very issue you pointed out.

Then again, a dual 7447 @1.83GHz is more than good enough for just about anything.

@DieHard I honestly think the pre-Quad-core liquid-cooled G5s were engineered precisely to lead people into accepting the upcoming switch to the generic IBM-PC-compatible Intel-based architecture. Even the MDD's power supply wasn't on the bottom. I don't think their engineers were incapable of seeing the issue with that design from the get-go.

I wonder if the power supply of a G5 could be externalized... But that'd be a different topic.

[redstudio]

I confirm GaryN's observations.. in fact I had made them too (see at the beginning of the conversation of this post). visibly seen after placing and observing thermal paste on copper heatsink. Half CPU not covered.

[Knezzen]

Yeah, I can confirm as well. Although, I use a copper heatsink with my MDD and it has worked like this since 2018 or so without any issue together with my MDX.

Covering half of the CPU core seems to be enough after all. I don't recommend anyone to do this, but it does work just fine

[Jubadub]

Covering half of the CPU core seems to be enough after all. I don't recommend anyone to do this, but it does work just fine

That's somewhat surprising! I'm glad it can work like that.

[GaryN]

Yeah, I can confirm as well. Although, I use a copper heatsink with my MDD and it has worked like this since 2018 or so without any issue together with my MDX.

Covering half of the CPU core seems to be enough after all. I don't recommend anyone to do this, but it does work just fine

That's fascinating. After the time it took for me to find an MDX and the obscene amount I paid for it, I was not going to try that.

Two observations:
1) Even without doing any real scientific testing, I would surmise that the copper sink's small improvement in cooling over the "stainless" would likely be offset by the 25% loss of contact area, making the risk/reward ratio rather poor.
2) Again, no testing done but considering that OS9 mostly exercises only one proc, it's possible that no overheating risk would occur if that proc is the fully-covered one and the other is just basically idling. I would still be risk-adverse to trust the arrangement doing any processor-intense tasks in OSX however.    Knez: Do you run only OS9 on your MDD?

The reality may be that the 7447 just runs a little cooler than the 7450 anyway and so up to 1.8Ghz at least, they don't really require that fancy-pants copper cowboy anyway. For that matter, unless you're obsessed about noise, the MDX is perfectly happy with the tiny little Xserve–compatible sink with its little built-in fan that comes with it. One could even make the argument that's better overall in an MDD since it offers less obstruction to the airflow from the main fan and the tiny MDX fan is barely audible anyway. Of course, then we'd be into aerodynamics and how the original heatsinks sit right in front of the main fan directing the airflow through the fins and out the back but now we have a bunch of turbulence and inefficiency instead.

Of course, pushing to 2.0Ghz or beyond, all of the above goes right out the window and maximum cooling becomes paramount, along with the often–observed instability and crash–proneness (proneness?) at ludicrous speed.

Perhaps if we all just sit back and wait, one of our resident OCD's will try to TIG–weld an extension to the copper sink to cover that protruding proc……

                  I'll be here all week folks… don't forget to tip your waitress…   

[refinery]

Couldn't one just use some copper shims and some washers to adjust the height of the heatsink to account for the shims?
I had to do something similar when putting a wx4150 into a 2010 iMac.

(Edit) something like this:
https://www.ebay.com/itm/274335443085?mkcid=16&mkevt=1&mkrid=711-127632-2357-0&ssspo=vE7WmKCSTMq&sssrc=4429486&ssuid=SJ9KN-eSRDy&var=&widget_ver=artemis&media=COPY

[indibil]

Yeah, I can confirm as well. Although, I use a copper heatsink with my MDD and it has worked like this since 2018 or so without any issue together with my MDX.

Covering half of the CPU core seems to be enough after all. I don't recommend anyone to do this, but it does work just fine

That's fascinating. After the time it took for me to find an MDX and the obscene amount I paid for it, I was not going to try that.

Two observations:

I agree with your contribution