MP 1,1-5,1 2009 4,1 de-lidded CPUs and the plastic spacers

[designed]

I've had my cMP 4,1 dual CPU since new in 2009 but only decided to upgrade it now as a part of my "killing time during corona" project.

I purchased a duo of X5690s already de-lidded and installed them few days ago. It wasn't immediately a huge success, but "re-refitting" the heatsinks solved it mostly. My problem is that the CPU A still runs a bit hot even after re-pasting couple of times. This means I need to have more rpms in MFC than I'm comfortable with, even if my temp goals are little bit ambitious compared to maximum allowed. I know there's a considerable TDP difference between the old 2,26 GHz Nehalems and the full-fat Westmeres.

I also realise the CPU A heatsink does the Northbridge cooling. One of the NB pins was broken and while examining the other it broke off too, but I replaced them with equivalents from an old Asus motherboard. I re-pasted the NB while at it and the NB temps have been sweet ever since.

I read in an old post here that you can alternatively skip the plastic spacers used with de-lidded CPUs to better equalise the temp between CPU diode and the heatsink. Has any current user done this to any risk or effect? I'll also clean any excess silicone gunk left over from the de-lidding.

 

[tommy chen]

you can kill your CPU socket when use de-lidded CPU without the plastic spacer!

 

[h9826790]

you can kill your CPU socket when use de-lidded CPU without the plastic spacer!

You can also kill the socket when use de-lidded CPU with the plastic spacer!

TBH, use the spacer is more danger.

The heatsink is designed to work with de-lidded CPU, as long as don't intentionally over tighten the heatsink. It won't kill the socket.

However, if the user didn't clean the silicon seal properly when during de-lid. Those silicon seal will make the CPU "thicker" than normal. Which means, more pressure to the socket pins when you screw down the heatsink (with that plastic spacer). This silicon seal also increase the gap between the die and the heatsink. Which cause abnormal high CPU temperature.

Of course, I can't see anything wrong to use the spacer if everything is done properly. However, if something not work as expect (e.g. like OP's case), it's better to remove the spacer, and see if that can fix the issue.

 

[Snow Tiger]

My problem is that the CPU A still runs a bit hot even after re-pasting couple of times.

CPU A will always run hotter than CPU B in these Systems .

It is the primary processor and also because of it's physical placement .

CPU A Heatsink also receives radiated heat from it's close proximity to the northbridge heatsink , the hottest chip in your Mac . It's part of the CPU / memory air channel design .

Nehalem DP CPU Tray's air flow , from Front to Rear of System :

AIR INTAKE FAN >>> CPU B HEATSINK >>> NB HEATSINK >>> CPU A HEATSINK >>> AIR EXHAUST FAN .

 

[Snow Tiger]

I also realise the CPU A heatsink does the Northbridge cooling. One of the NB pins was broken and while examining the other it broke off too, but I replaced them with equivalents from an old Asus motherboard. I re-pasted the NB while at it and the NB temps have been sweet ever since.

I hope those replacement Northbridge heatsink fasteners are real push pins with the appropriate pressure .

The replacements should look similar to this , they are spring loaded . Do not use bolts , nuts , rivets , etc .

 

[designed]

I hope those replacement Northbridge heatsink fasteners are real push pins with the appropriate pressure .

The replacements should look similar to this , they are spring loaded . Do not use bolts , nuts , rivets , etc .

Yup, pretty close down to the color!

Thanks for the tips guys, pros and cons.
I ran the system some more. Temps seem ok but the fans creep up during time. I also noticed the heatsink and CPU diode temps have a significant gap, so there's definitely something to fix. Will disassemble it all and check again.

 

[Snow Tiger]

Yup, pretty close down to the color!

Thanks for the tips guys, pros and cons.
I ran the system some more. Temps seem ok but the fans creep up during time. I also noticed the heatsink and CPU diode temps have a significant gap, so there's definitely something to fix. Will disassemble it all and check again.

what's the difference , in Celsius please ?

difference between CPU chip and CPU heatsink temps at idle .

and

difference between CPU chip and CPU heatsink temps at full load ( 100 percent core usage ) .

And , are your fan rotationals being governed by factory SMC ? Or are you using a program like MFC ?

 

[tommy chen]

You can also kill the socket when use de-lidded CPU with the plastic spacer!

but you already know what the spacer is for?
pressure distribution!

and yes, screwing the heatsink with force does not help

 

[h9826790]

but you already know what the spacer is for?
pressure distribution!

This is exactly where the problem is, if the silicon seal still on the cMP, the spacer won't able to distribution pressure correctly.

However, it's been proved that there is no need to have that spacer for the CPU and the heatsink to work properly.

Case 1: Silicon seal not clean, with plastic spacer -> CPU may overheat, pins may damage if careless

Case 2: Silicon seal not clean, without plastic spacer -> all normal, pins may damage if careless

Case 3: Silicon seal clean, with plastic spacer -> all normal, pins may damage if careless

Case 4: Silicon seal clean, without plastic spacer -> all normal, pins may damage if careless

Best case of course is case 3. However, the conclusion should be still "use spacer create more trouble".

Anyway, I think OP have to post the screen capture for us to continue the discussion. Without knowing the CPU diode temperature and CPU heatsink temperature, it's hard for us to know if the heatsink has proper contact with the die. And should he try to remove the spacer.

 

[designed]

what's the difference , in Celsius please ?

difference between CPU chip and CPU heatsink temps at idle .

and

difference between CPU chip and CPU heatsink temps at full load ( 100 percent core usage ) .

And , are your fan rotationals being governed by factory SMC ? Or are you using a program like MFC ?

The difference at idle is pretty non-existant, less than 5°C. At load it is over 20°C. I think CPU A diode was 82°C and heatsink was 59°C. This was true for both CPUs, but as usual CPU B runs cooler.

I've been letting SMC govern the fans while sorting this out, it seems to be doing okay once it realises what's up. I have MFC set up for temp checking and I intend to do my own fan profiles once I'm happy with the heatsink setup. If you have good pointers, those would be appreciated.

Last night I tried running the system without the plastic spacers. Had some issues initially (bad contacts) but solved those. Saw no immediate improvement in heat so I installed the spacers back for now.

 

[h9826790]

The difference at idle is pretty non-existant, less than 5°C. At load it is over 20°C. I think CPU A diode was 82°C and heatsink was 59°C. This was true for both CPUs, but as usual CPU B runs cooler.

I've been letting SMC govern the fans while sorting this out, it seems to be doing okay once it realises what's up. I have MFC set up for temp checking and I intend to do my own fan profiles once I'm happy with the heatsink setup. If you have good pointers, those would be appreciated.

Last night I tried running the system without the plastic spacers. Had some issues initially (bad contacts) but solved those. Saw no immediate improvement in heat so I installed the spacers back for now.

Those number looks very normal to me.

If poor heatsink contact, the temperature difference most likely will be much bigger than that.

Of course, it can do better (e.g. You can easily make the CPU run few °C cooler by using liquid metal). However, I can't see anything wrong in your setup at this moment.

If you want to CPU run cooler, you can manually spin up the fan.

 

[Snow Tiger]

Those number looks very normal to me.

If poor heatsink contact, the temperature difference most likely will be much bigger than that.

Of course, it can do better (e.g. You can easily make the CPU run few °C cooler by using liquid metal). However, I can't see anything wrong in your setup at this moment.

If you want to CPU run cooler, you can manually spin up the fan.

He's running a X5690 at 82 degree C . That doesn't concern you ? It's a bit over the max temp of 78.5 degree C so he's already out of the safety zone . Things will only get worse going forward . Max temps also are not the same as recommended temps . Best temps are as cool as possible .
[automerge]1589195088[/automerge]
OP : Here are the thermals of a properly rebuilt Dual X5690 at continuous System load .

The difference between CPU A chip and CPU A heatsink = 8 degree C .
The difference between CPU B chip and CPU B heatsink = 6 degree C .

The use of MFC is highly recommended with this configuration .

The difference in temps between your CPU chip and CPU heatsink is too great .

Get it down as low as possible to improve the durability of your System .

For my System , MFC settings are shown in the fan window info in the picture below . 1700 RPMs for both the CPU boosters and 1200 RPM for everything else . But fan settings are customized for every System and this might not work for you .

 

[h9826790]

He's running a X5690 at 82 degree C . That doesn't concern you ? It's a bit over the max temp of 78.5 degree C so he's already out of the safety zone .

No, that 78.5°C is the max T-case, OP's 82°C is the T-diode temperature.

You can run Windows with you X5690, download and run the Intel Processor Diagnostic Tool. It will show you the xax temperature (which is the diode temperature, can be measured by the software itself, or extract from SMC key in macOS) of a X5690 is over 100°C.

82°C is still 20°C away from overheating, definitely in the normal range.

Even in your screen capture, it's clearly that the max difference between CPU A diode and heatsink is 14°C (74-60), not 8°C. For CPU B, that's 13°C (65-32), not 6°C. And that's with the much higher than normal fan speed.

When you make the screen capture, CPU A just was drawing about 47W, CPU usage on the menu bar is very low, nowhere near "under stress". Which also means, the CPU was cooling down AFTER real high stress. In fact, Terminal says "process completed". And you made a capture right after that.

Since there is a time lag in heat transfer, the heat generated from the stress test was still on the heatsink, and the CPU diode is cooling down quickly. The temperature difference is much less than your max difference. IMO, that 8°C and 6°C isn't the correct figure to be used for comparison.

If I did that on my cMP, I can actually make it shows zero delta T.

I ran Prim95 for about 10min, let the CPU stabilised at the max temperature. Then stop the process, and make this screen capture when the CPU was cooling down. CPU diode 46°C, CPU heatsink 46°C, but this doesn't mean anything.

Please don't get me wrong. You built a nice system. You can of course config the cMP to the cooling level that you want, however, there is just nothing wrong on OP's cMP.
[automerge]1589211450[/automerge]
This was captured DURING stress test (prime95) on my own cMP, as you can see, the CPU A was drawing 85.25A, that's about 99W.

And the temperature difference between diode and heatsink was 13°C, pretty much the same as your max difference captures. Of course, I am with a single processor cMP, therefore, no need to spin up the fans that much.
[automerge]1589211627[/automerge]
20°C delta T definitely not the best cooling cMP, however, normal, very normal.

If OP want to let CPU A run cooler (IMO, just a few °C cooler to stay below 80°C is enough. This already 100% guarantee the CPU can stay at max Turbo Boost clock speed all the time), then manually spin up the fan a bit is the way to go. And that's just more for self comfort, not really because his cMP is doing something wrong

 

[Snow Tiger]

Even in your screen capture, it's clearly that the max difference between CPU A diode and heatsink is 14°C (74-60), not 8°C. For CPU B, that's 13°C (65-32), not 6°C. And that's with the much higher than normal fan speed.

Oh no no no !

Look only at the current temps ( and other sensors ) being reported by Hardware Monitor , please . Those are the only ones usable for such a determination .

The minimum and maximum sensor ( not just thermal ) readings are all individual readings taken at many , separate times . They are not connected to each other . They don't represent a single snapshot in time ( which is required for such a determination ) .

Those are the absolute , individual extremes of each sensor reading during the long stress test ( 24 hours , usually ) . The time interval for the reporting was set at 1 Day .

There never was a 13 or 14 degree C difference between a CPU chip and a CPU Heatsink . I check these figures occasionally throughout the entire burn in and would have gone nuts had this been true .

I have already talked with the software publisher of Hardware Monitor on this very issue on identifying the sensors . The CPU Temperature Diode is almost , 100 percent certainly a Tcase . Apple won't say what it is . It's supposed to be a sensor reading in the very center and top of the IHS . It is not a Tjunction .

Please recognize the extreme difficulty in determining the actual reading of a Tjunction - it's not a sensor that can be quickly , directly or accurately read . It's actually made using a bunch of points of data . Even if I had access to that sensor data , I'd be suspicious of the reading .

As a matter of fact , as far as I can tell , there is only one thermal sensor that can be read by macOS utilities for a Xeon 5600 class processor : Tcase . Tjunction sensor is not accessible . There are two other thermal signals called PROCHOT and THERMTRIP and then there's PECI , which is not synonymous with Tjunction ( I think ) .

If it were , Intel would say so in the some 240 pages of the technical datasheet of this processor series ( which I partially read ) . The term Tjunction is used only once in the entire document . And tellingly , Intel won't publicly say what the Tjunction figure is . It's like a state secret .

I will run Windows 7 on a cMP some day with these processors , because there just might be additional sensors I can access . But it's of limited value to me since these cMPs rarely get requests for Windows to be installed and I rarely stress test the Systems with this OS .

In short , in macOS , I know the Tcase max ( it's publicly reported at 78.5 C ) , the actual real time Tcase sensor data readings of the Systems I build and how far I need to keep a distance between the two in order to have decent durability .

Maybe the OP should download and run Bresink's Hardware Monitor so we can fairly compare our findings .

 

[h9826790]

Oh no no no !
Look only at the current temps ( and other sensors ) being reported by Hardware Monitor , please . Those are the only ones usable for such a determination .

That's exactly what I want to point out, you use the "current temperature" at the WRONG time.

If you want to use current temperature, please capture that DURING stress test, but not AFTER the stress test.

There never was a 13 or 14 degree C difference between a CPU chip and a CPU Heatsink . I check these figures occasionally throughout the entire burn in and would have gone nuts had this been true .

It's just impossible, the max CPU A diode temperature was 74°C. That's a fact, recorded by your screen capture.

So, at the moment that CPU A diode was 74°C, what's the heatsink temperature? 60°C or lower (That 60°C not necessary happen right at the moment that the diode was at 74°C. However, the recorded max was just 60°C, not anything higher).

What's that means? the delta T was at least 14°C, cannot be anything less than that.

I have already talked with the software publisher of Hardware Monitor on this very issue on identifying the sensors . The CPU Temperature Diode is almost , 100 percent certainly a Tcase . Apple won't say what it is . It's supposed to be a sensor reading in the very center and top of the IHS . It is not a Tjunction .

I don't care what software publisher of Hardware Monitor says, they are not Intel, they are not Apple, and they are not the person to define T-case.

You can ask how they get that temperature. I bet that's from SMC key TCAD, guess what that mean? Temperature CPU A Diode.

[N.B. there are few seconds difference between the SMC key values extraction and iStat menu capture. Therefore, the numbers are not identical, but close enough to see their relationship]

I know how to extract those SMC keys' value, and what they means (credit to Lobotomo Software)

By definition, T-case can't be measured on cMP. I pointed out this to you before. If you don't want to believe it, it's up to you.

And I already told you how that "above 100°C" from, it's form Intel. You don't need to believe me, please download that tool (directly from the Intel server) in windows, and run that by yourself. Then you will know that's the diode temperature. And no matter how you stress the CPU, the SMC will keep your CPU away from overheating. Remember, that tool is form Intel, the CPU manufacture.

Anyway, I have no interest to persuade you anything. you can keep believe that's the T-case temperature. Not my problem, and no danger anyway. I just want to make sure OP can receive the info from both sides, and I will let him to decide what he should do.

 

[Snow Tiger]

That's exactly what I want to point out, you use the "current temperature" at the WRONG time.

If you want to use current temperature, please capture that DURING stress test, but not AFTER the stress test.

You are again incorrect .

The current temps in my screen grab were during the CPU stress test - look at the menu bar where Menumeter's 24 processor core thread thermometers clearly are all solid blue ( user ) / red ( system ) , indicating 100 percent usage of each thread . This is a 12 Core @ 3.46 GHz ( 24 threads ) System .

This machine is at 100 percent CPU load . Guaranteed . I'm surprised you didn't catch it . Don't you use Menumeters ?

Memtest for Mac just ended running and there was a momentary dip in the CPU power consumption report . I took a screen grab just before those power figures went right back up to ~ 95 W . Bad timing - I wasn't looking at it . I was more interested in the thermals and fan rotationals .

I'm building another 12 Core @ 3.46 GHz System right now . I'll be more careful with my screen grabs this time , if you're interested .

 

[minifridge1138]

Looking at the pic of Snow Tiger’s “Extreme Values” table, CPU A diode reached 74* and the heat sink never went above 60*. CPU B diode reaches 65* and the heat sink never went above 52*.

That is a difference of 14* between CPU A diode and CPU A Heat Sink. And a 13* difference between CPU B diode and CPU B heat sink. At a minimum.

That’s not a criticism, it’s just what the table shows.
What tool are you using to capture that temperature info over time (not the stress test tool, I can get to 100% load all by myself)?

 

[h9826790]

You are again incorrect .

The current temps in my screen grab were during the CPU stress test - look at the menu bar where Menumeter's 24 processor core thread thermometers clearly are all solid blue ( user ) / red ( system ) , indicating 100 percent usage of each thread . This is a 12 Core @ 3.46 GHz ( 24 threads ) System .

This machine is at 100 percent CPU load . Guaranteed . I'm surprised you didn't catch it . Don't you use Menumeters ?

Memtest for Mac just ended running and there was a momentary dip in the CPU power consumption report . I took a screen grab just before those power figures went right back up to ~ 95 W . Bad timing - I wasn't looking at it . I was more interested in the thermals and fan rotationals .

View attachment 914179


I'm building another 12 Core @ 3.46 GHz System right now . I'll be more careful with my screen grabs this time , if you're interested .

No, I didn't see that, and I am not a Menumeters user, I don't even know this software exist.

Anyway, I already pointed out the reason to you. 47.1W CPU A power draw, that's under stress? Terminal said "Process Completed", that's under stress?

You can believe Menumeters, but you should NOT ignore "hints". To me, there looks like a time lag in Menumeters (1-2 second lag is very normal for this kind of software). And that capture was right after the stress test completed, not during the test. Your screen capture just not supporting what you said.

Also, why use a Memtest to stress CPU? That's simply the wrong tool. If you want to stress the CPU, use Prime95, or the Yes command, etc. Memtest is really for testing memory, not CPU, or cooling ability.

If you want to show that your CPU diode and heatsink temperature difference is only 8°C, please do EXACTLY the following.

1) Boot to desktop
2) Open Hardware monitor
3) Run Prime95 Torture Test with EXACTLY this setting

4) wait for 15min
5) Capture the screen
6) Stop Prime95

Then compare the max CPU diode temperature to the max CPU heatsink temperature. And that the difference is 8°C or lower.

 

[designed]

Thanks again everyone for your input, I see that this is indeed a heated topic indeed!

Last night I swapped the CPUs around, which seemed to bring down the CPU A vs. CPU B difference down a bit, not that it was a problem in the first place. I have the plastic spacers back on for added peace of mind while tightening the heatsink screws and the silicone was cleaned off pretty nicely from the get go.

I'll try a few more things to see if I can bring the 100% load temps down a bit, but the system idles very nicely.

If anyone is using MFC, I'd be curious to know which sensors and temp ranges people have set up for 4,1/5,1 systems.

 

[h9826790]

If anyone is using MFC, I'd be curious to know which sensors and temp ranges people have set up for 4,1/5,1 systems.

For CPU, ALWAYS base on the CPU diode temperature.

e.g. Booster A base on CPU A diode, min 60, max 99 etc. (N.B. It won't really let the CPU to max at 99°C, but build a fan profile that reach max fan speed when CPU at 99°C. Since you don't need max fan speed to cool down the CPU, therefore, you can build a fan profile base on a CPU temperature that you can never reach)

If you want to let the CPU run cooler, I suggest you modify the max temperature, but not the min temperature, so that the fan won't spin up too early when the CPU still cool.

For Intake / Exhaust, you can make them base on CPU A temperature, because that's the more demanding one.

Rememeber, MFC does override the SMC protection, even your CPU is overheating, the SMC won't able to spin up the fan anymore. So, you MUST build a profile that allow you to stay on the safe side.

 

[Snow Tiger]

No, I didn't see that, and I am not a Menumeters user, I don't even know this software exist.

Anyway, I already pointed out the reason to you. 47.1W CPU A power draw, that's under stress? Terminal said "Process Completed", that's under stress?

You can believe Menumeters, but you should NOT ignore "hints". To me, there looks like a time lag in Menumeters (1-2 second lag is very normal for this kind of software). And that capture was right after the stress test completed, not during the test. Your screen capture just not supporting what you said.

Also, why use a Memtest to stress CPU? That's simply the wrong tool. If you want to stress the CPU, use Prime95, or the Yes command, etc. Memtest is really for testing memory, not CPU, or cooling ability.

If you want to show that your CPU diode and heatsink temperature difference is only 8°C, please do EXACTLY the following.

1) Boot to desktop
2) Open Hardware monitor
3) Run Prime95 Torture Test with EXACTLY this setting
View attachment 914294
4) wait for 15min
5) Capture the screen
6) Stop Prime95

Then compare the max CPU diode temperature to the max CPU heatsink temperature. And that the difference is 8°C or lower.

You need to become familiar with Hardware Monitor to understand how it performs before you comment on it .

Sensor reports sometimes dip on occasion . I grabbed the screen as this event happened with CPU power . And I already told you that . Geesh .

And I'm not using Memtest as a CPU stress test 😂 .

Here are the four stress tests I was using to place this macOS System continuously and concurrently under load :

Processors : Geekbench 3 .
Memory : Memtest for Mac .
Graphics Card : Luxmark 3.1 ( Open CL ) .
NVMe M.2 SSD : Aja System Test Lite .

These tests were run for a whole day . It's one hell of a suite and represents a greater degree of stress on a System than that of the workflow of professional editors .

 

[h9826790]

I'm not using Memtest as a CPU stress test 😂 .

Here are the four stress tests I was using to place this macOS System continuously and concurrently under load :

Processors : Geekbench 3 .
Memory : Memtest for Mac .
Graphics Card : Luxmark 3.1 ( Open CL ) .
NVMe M.2 SSD : Aja System Test Lite .

These tests were run for a whole day . It's one hell of a suite and represents a greater degree of stress on a System than that of the workflow of professional editors .

If you want to stress the PSU, that make sense.

But if you want to run stress test for a particular component, you better do them one by one.

e.g. Run Furmark to stress the GPU, but leave the CPU idle. So that Furmark can do its work as hard as possible to stress the GPU. (I prefer Furmark for stress test, no AA, full screen, that's way more stressful than Luxmark)

If you run CPU stress test at the same time, there will be less CPU resources for the GPU driver and GPU stress test software. Which may make the software less demanding to the GPU. In other words, you are just running "high stress test" to the GPU, but not "max stress test".

This may also explain why your CPU can't stay at the max stress temperature. Because it has to give out some resource to work for the GPU / NVMe / memory test, and those are less demanding job for the CPU.

I also wonder why you run stress test on NVMe for a whole day. That burn the NVMe's life span with little to no benefit. If you want to know if the NVMe may overheat or not, a short stress test is good enough.

Let's say it's a 250GB 970Evo, and can only achieve 1500MB/s sequential write on the cMP. Stress test it for 24 hours means 129600000MB written, that's ~129600GB, or 129.6TB. The warranty only cover up to 150TB written.

If you put this 970 Evo on a Highpoint 7101A card, and run stress test for a whole day with ~3000MB/s. You can basically burn the warranty away in a single day.

Of course, the disc speed test will do write -> read -> write -> read... Therefore, not keep writing 24 hours, however, IMO, still bad enough to burn the life span (and warranty).

Anyway, ready for the challenge? (follow the exact steps in post #18 and shows us your CPU diode temperature only 8°C higher than the heatsink temperature)

A side topic, from memory, you own 7,1 with 1.5TB RAM. May I know what's the boot time (from press the power button to wallpaper shows up). This is purely for fun. Since so many people care about boot time, and we know that more RAM on the cMP will need more time to POST. I just suddenly wonder how much time the 7,1 need to POST with 1.5TB RAM. It will be funny if the 7,1 with 1.5TB RAM still boot faster than a cMP that only has 128GB RAM. Of course, I assume both Mac Pro have a relatively clean OS on a SSD. [Please feel free to reject this request, or provide relavant data, I really just want to know this for fun. Million thanks in advance!]

 

[Snow Tiger]

Thanks again everyone for your input, I see that this is indeed a heated topic indeed!

Last night I swapped the CPUs around, which seemed to bring down the CPU A vs. CPU B difference down a bit, not that it was a problem in the first place. I have the plastic spacers back on for added peace of mind while tightening the heatsink screws and the silicone was cleaned off pretty nicely from the get go.

I'll try a few more things to see if I can bring the 100% load temps down a bit, but the system idles very nicely.

If anyone is using MFC, I'd be curious to know which sensors and temp ranges people have set up for 4,1/5,1 systems.

It can be dangerous to get caught in the crossfire of two stubborn people here ...😗

But we do try to help others on the forums .

It is extremely unlikely you will damage your Xeons by overtemping them . They are quite robust components and have thermal control circuitry that permits the chip to throttle its performance and even shut down if it gets too hot . Then , all you have to do usually is cool it properly and the System fires up again . The cMP is a great machine to learn how to repair and upgrade workstations with .

With regards to constant MFC settings , for my System above , 1700 RPMs for both the CPU boosters and 1200 RPM for everything else . But fan settings are customized for every System and this might not work for you .

 

[designed]

It is extremely unlikely you will damage your Xeons by overtemping them . They are quite robust components and have thermal control circuitry that permits the chip to throttle its performance and even shut down if it gets too hot . Then , all you have to do usually is cool it properly and the System fires up again . The cMP is a great machine to learn how to repair and upgrade workstations with .

Thank you! I agree. I built a few PCs back in the day but I'll say this to others: cMP really is a great platform to learn how to wrench a Mac, just do some research beforehand (mini-6-pin power cables, 2009 dual de-lidded CPUs etc.) and you'll
avoid pesky part order times. And try not to break the CPU tray.

As for the original purpose of this thread: you can run the CPUs without the plastic spacers, so it's not the end of the world if you loose/break them. And as pointed out, this might desireable if the CPUs has been de-lidded with a lot of the silicone still there.

As for my system, I'm pretty sure I haven't been tightening the CPU heatsinks enough. Even if the CPU tray red lights go away, thermal performance seems to improve with extra turns of the screws. System was stable during a 40 minute Folding run, but the CPU A red light morse codes at me, probably because PCECI temp insists on being 98-99°C. Diode is at 85°C or less. PMSET shows no record of thermal or throttling events. CPU B is less than 10°C behind (with less rpm), which is nicely balanced. Northbridge diode hangs around 59°C.

I've ordered some better quality thermal paste and chipset cleaning agent and I have some more RAM on the way, so I'll do one last pass while replacing the memory.

 

[flyproductions]

I hope those replacement Northbridge heatsink fasteners are real push pins with the appropriate pressure .

The replacements should look similar to this , they are spring loaded . Do not use bolts , nuts , rivets , etc .

View attachment 913881

Why not use bolts and nuts? I am using bolts and nuts with the stock springs and the space between the bolt's head and the (self saving) nut exactly equal to the stock pin's length. All works perfectly fine. Only difference to the pin solution is, that it is very unlikely to break.

As far as i see, the pic shows a common pin like it has been used on many older GPUs. I wouldn't recommend these as their springs are much weaker than what is used on the cMP’s genuine northbridge heatsink pins.