Discussion Using a Cooler Master GeminII M4 on AM4

Toggle sidebar Toggle sidebar
bit_user

bit_user

Titan
Ambassador
Jan 20, 2010
19,450
10,510
79,540

Introduction​

I'm just posting this little note, in case it helps someone else. Maybe not trying to use this specific cooler on AM4 or AM5, but perhaps you've got some old Cooler Master model and are wondering if it can still be used.

Background​

Recently, I had a need to use an alternate cooler on an AM4 board/CPU. I won't go into the rationale, but another relevant fact is that I have a newer Cooler Master model I couldn't use on it, which included an AM4-compatible mounting option (the included hardware with my M4 supported only up to AM3, for AMD CPUs).

The Geminii M4 is top-down cooler I bought for $35, back in 2012. It has 4 single-ended heat pipes, using "Direct Touch technology". I had recently used one of their direct-touch tower heatsinks, in another machine, and was pleased with how well it performed. So, I got the M4 for a machine where I deemed the "down-draft" style to be more appropriate. At the time, I hadn't seen many reviews of down-draft coolers. This style of cooler rarely performs on par with the towers, and therefore tends not to be covered as well by the tech press. After I had bought it, I learned of the Scythe Big Shuriken 2 (Rev B), which is what I ended up using for that machine. So, my Geminii M4 sat in a corner, unused, until now.

www.coolermaster.com

GeminII M4 CPU Air Cooler

GeminII M4 is a super low profile CPU cooler (59mm / 2.3in height) designed for a system with limited space, such as mini-ITX, HTPC, or any small desktop. With its 4 direct contact heat pipes and unique heat pipe layout, GeminII M4 not only keeps compact size but provides amazing cooling...
www.coolermaster.com www.coolermaster.com

I wondered whether or not it had a compatible mounting system with my newer Cooler Master heatsink, and it turned out their bases are exactly the same! So, I began to wonder how it would perform, on CPUs higher-power than what it was originally designed for. This wasn't hard to guess, as you can find several comprehensive reviews of this cooler (see below). Still, I had a need, which the M4 could probably fill, and that was only amplified by my curiosity. So, here's what I learned.

Test Setup​

Here are the key specs of the test setup:
  • Case: open bench
  • Motherboard: ASRock Rack X570D4U-2L2T micro-ATX motherboard
  • CPU: Ryzen 7 5800X
  • Thermal compound: Arctic MX-6
  • BIOS settings: mostly stock (no OC), but PPT set to 142 W.
  • Fan curve: 100%, fixed speed.
  • OS: Ubuntu Linux 24.04
  • Ambient temperature: 25 C
  • Monitoring: turbostat --quiet --Summary --show=PkgWatt --interval=0.5 In addition to this, I used a custom python script to log the CCD and IOD temperatures.

wDbiQ1V.jpeg

​
Two points about installing the heatsink are probably noteworthy. First, I had to removed the motherboard's AM4 heatsink posts and backplate. Removing the backplate required the use of a heatgun. I could not budge it, either by pushing from the other side or prying it. It was only by several minutes of heating at probably 50 or 60 degrees Celsius (my heat gun has a variable temperature dial, but doesn't list exact temperatures on it) that I could start to pry it loose. I used a plastic upholstery-removal tool, that's made of fairly soft plastic and posed little risk to the motherboard. The entire backplate was covered by double-sided tape.

Second, I had to use a lot of heatsink compound! I've never used so much on a heatsink, in my life! I know such an amount was necessary, because I had a few "oops" moments of the heatsink coming off the CPU, where I got to see how it had spread (and then cleaned it off and started over). It turns out that an "X" pattern didn't spread it well enough. only an "asterisk" pattern would cover the entire CPU. For the most part, this is due to the uneven base and gaps between the heat pipes. It also didn't help that the Cooler Master mounting system restricted my range of motion enough that I had to work the cooler back-and-forth for probably 5 minutes, before I consistently felt the metal-on-metal scraping that told me it had completely "landed".

Hqpt0jy.jpeg

​
It looks like too much, but it wasn't.

Results​

I found the CPU temperature plateaued rather quickly, thus 90 seconds was adequate for a test run. I tried a variety of workloads, increasing in their amounts of power consumption. What I discovered was that the CPU would throttle at 90 degrees. So, the way I decided to quantify the results, which I think should also be most applicable to others, is by looking at the peak amount of power output the cooler could dissipate, without throttling. That should make my findings applicable to different CPU models and workloads, as long as you can find out how much package power your combination can use. However, perhaps a dual-CCD CPU will manage a bit more power than what I observed.

While the cooler could cope with brief spikes up to the 142 W package power limit, the limit of its sustained output on a strenuous lightly-threaded workload (FFT x5 threads) was about 105 W (CPU package power). That's conveniently equal to the stated TDP of the 5800X, but remember that's the point where it hits 90 degrees Celsius, which is where it throttled. So, I wouldn't recommend this specific cooler for anything above the models with 88 W PPT (which are the models advertised at 65 W TDP).

dBfmSXu.png


dlDbZKl.png

Note: This image replaces temperature data with clock speed. It was too cluttered to cram this all into a single plot.
​

With a lighter-weight, all-core workload, the cooler was able to sustain 119 W!

s5DGaqo.png


A1nzX47.png

Note: As before, this plot replaces temperature data with clock speed.
​
Also, I observed idle temps around 28 degrees. I'd wait until the IOD and CCD were both reporting below 28 C, before starting another test run.

As noted in the Test Setup section, the ambient temperature was about 25 degrees Celsius, during the above tests.

Alternate Fan​

Since the cooler included only a 15 mm low-profile fan (120 mm length & width), I tried replacing it with a Noctua NF-A12x25 PWM. This is their current 120 mm flagship CPU fan, and not exactly cheap (25 mm thick). I tested this out of curiosity and because I already had the fan, rather than because it would make sense to buy such a low-end cooler and pair it with such a nice fan. In fact, while I could feel more air coming off it, the surprising result was no improvement!

Seeking to investigate further, I discovered that I could actually touch all of the heat pipes, in a spot before they reached the fin array. One was noticeably warmer than the others, which was presumably the one hitting the CCD. Given this fact, it's not surprising that more airflow had no effect. The thermal bottleneck wasn't in the fin array. I found that a little incongruous with the fact that they'd gone to the trouble and expense of soldering the fins to the heat pipes.

Other Notes​

When installing the heatsink, I tightened the fasteners all the way to the ends of the threaded portion. Since these bolts were specific to AM4, I figured they made the threads that long for a reason, and the instructions gave no guidance on how far to go.

Subsequently, I tested reducing the mounting tension and found that I could reduce temperatures by about 1.2 degrees by backing off each of the nuts about 1.5 revolutions. This part is probably more generally interesting and relevant, so I plan to make a separate post about it. Therefore, I won't expound on it further, here.

Professional Reviews​

In chronological order:
 
Last edited:
  • Like
Reactions: thestryker and Flayed
bit_user said:

Introduction​

I'm just posting this little note, in case it helps someone else. Maybe not trying to use this specific cooler on AM4 or AM5, but perhaps you've got some old Cooler Master model and are wondering if it can still be used.

Background​

Recently, I had a need to use an alternate cooler on an AM4 board/CPU. I won't go into the rationale, but another relevant fact is that I have a newer Cooler Master model I couldn't use on it, which included an AM4-compatible mounting option (the included hardware with my M4 supported only up to AM3, for AMD CPUs).

The Geminii M4 is top-down cooler I bought for $35, back in 2012. It has 4 single-ended heat pipes, using "Direct Touch technology". I had recently used one of their direct-touch tower heatsinks, in another machine, and was pleased with how well it performed. So, I got the M4 for a machine where I deemed the "down-draft" style to be more appropriate. At the time, I hadn't seen many reviews of down-draft coolers. This style of cooler rarely performs on par with the towers, and therefore tends not to be covered as well by the tech press. After I had bought it, I learned of the Scythe Big Shuriken 2 (Rev B), which is what I ended up using for that machine. So, my Geminii M4 sat in a corner, unused, until now.
​
www.coolermaster.com

GeminII M4 CPU Air Cooler

GeminII M4 is a super low profile CPU cooler (59mm / 2.3in height) designed for a system with limited space, such as mini-ITX, HTPC, or any small desktop. With its 4 direct contact heat pipes and unique heat pipe layout, GeminII M4 not only keeps compact size but provides amazing cooling...
www.coolermaster.com www.coolermaster.com
​

I wondered whether or not it had a compatible mounting system with my newer Cooler Master heatsink, and it turned out their bases are exactly the same! So, I began to wonder how it would perform, on CPUs higher-power than what it was originally designed for. This wasn't hard to guess, as you can find several comprehensive reviews of this cooler (see below). Still, I had a need, which the M4 could probably fill, and that was only amplified by my curiosity. So, here's what I learned.

Test Setup​

Here are the key specs of the test setup:
  • Case: open bench
  • Motherboard: ASRock Rack X570D4U-2L2T micro-ATX motherboard
  • CPU: Ryzen 7 5800X
  • Thermal compound: Arctic MX-6
  • BIOS settings: mostly stock (no OC), but PPT set to 142 W.
  • Fan curve: 100%, fixed speed.
  • OS: Ubuntu Linux 24.04
  • Ambient temperature: 26 C
  • Monitoring: turbostat --quiet --Summary --show=PkgWatt --interval=0.5 In addition to this, I used a custom python script to log the CCD and IOD temperatures.

Two points about installing the heatsink are probably noteworthy. First, I had to removed the motherboard's AM4 heatsink posts and backplate. Removing the backplate required the use of a heatgun. I could not budge it, either by pushing from the other side or prying it. It was only by several minutes of heating at probably 50 or 60 degrees Celsius (my heat gun has a variable temperature dial, but doesn't list exact temperatures on it) that I could start to pry it loose. I used a plastic upholstery-removal tool, that's made of fairly soft plastic and posed little risk to the motherboard. The entire backplate was covered by double-sided tape.

Second, I had to use a lot of heatsink compound! I've never used so much on a heatsink, in my life! I know such an amount was necessary, because I had a few "oops" moments of the heatsink coming off the CPU, where I got to see how it had spread (and then cleaned it off and started over). It turns out that an "X" pattern didn't spread it well enough. only an "asterisk" pattern would cover the entire CPU. For the most part, this is due to the uneven base and gaps between the heat pipes. It also didn't help that the Cooler Master mounting system restricted my range of motion enough that I had to work the cooler back-and-forth for probably 5 minutes, before I consistently felt the metal-on-metal scraping that told me it had completely "landed".

I'll probably update with a pic of how much I used. It looks like too much, but it wasn't.

Results​

I found the CPU temperature plateaued rather quickly, thus 90 seconds was adequate for a test run. I tried a variety of workloads, increasing in their amounts of power consumption. What I discovered was that the CPU would throttle at 90 degrees. So, the way I decided to quantify the results, which I think should also be most applicable to others, is by looking at the peak amount of power output the cooler could dissipate, without throttling. That should make my findings applicable to different CPU models and workloads, as long as you can find out how much package power your combination can use. However, perhaps a dual-CCD CPU will manage a bit more power than what I observed.

While the cooler could cope with brief spikes up to the 142 W package power limit, the limit of its sustained output was about 105 W (CPU package power). That's conveniently equal to the stated TDP of the 5800X, but remember that's the point where it hits 90 degrees Celsius, which is where it throttled. So, I wouldn't recommend this specific cooler for anything above the models with 88 W PPT (which are the models advertised at 65 W TDP).

Also, I observed idle temps around 28 degrees. I'd wait until the IOD and CCD were both reporting below 29 C, before starting another test run.

Alternate Fan​

Since the cooler included only a 15 mm low-profile fan (120 mm length & width), I tried replacing it with a Noctua NF-A12x25 PWM. This is their current 120 mm flagship CPU fan, and not exactly cheap (25 mm thick). I tested this out of curiosity and because I already had the fan, rather than because it would make sense to buy such a low-end cooler and pair it with such a nice fan. In fact, while I could feel more air coming off it, the surprising result was no improvement!

Seeking to investigate further, I discovered that I could actually touch all of the heat pipes, in a spot before they reached the fin array. One was noticeably warmer than the others, which was presumably the one hitting the CCD. Given this fact, it's not surprising that more airflow had no effect. The thermal bottleneck wasn't in the fin array. I found that a little incongruous with the fact that they'd gone to the trouble and expense of soldering the fins to the heat pipes.

Other Notes​

When installing the heatsink, I tightened the fasteners all the way to the ends of the threaded portion. Since these bolts were specific to AM4, I figured they made the threads that long for a reason, and the instructions gave no guidance on how far to go.

Subsequently, I tested reducing the mounting tension and found that I could reduce temperatures by about 1.2 degrees by backing off each of the nuts about 1.5 revolutions. This part is probably more generally interesting and relevant, so I plan to make a separate post about it. Therefore, I won't expound on it further, here.

Professional Reviews​

In chronological order:
Click to expand...
Any cooler should match CPU, socket match is just mechanical fit. Published TDP of a CPU rarely matches real heat output and is usually published for CPU base clock, without boost or OC. at which occasions it could rise double or more because of increased electrical power. That's why I always choose cooling solution according real electrical power requirement consumption with CPU at full boost usage and during prolonged time.
For an example, my R9 7900x has advertised TDP as 170W (AMD is more "honest" than Intel in that respect) at BIOS defaults @ base frequency 4.7GHz it draws about 120W but at full boost of 5.5GHz draws 185w with sharp temperature increase, slight OC and it hits 200W. It takes one of best performing 360 AIO coolers to keep temps under 90c for more than a minute, let alone longer 100% loads. No way any cooler that's rated for 170W TDP is going to come anywhere close.
 
Thank you for taking interest in my post!

CountMike said:
Any cooler should match CPU, socket match is just mechanical fit.
Click to expand...
Yes, I'm aware this was a weird choice of cooler, but I bought it a long time ago and so I had it on hand just when I needed a substitute cooler for that board. I had long wondered how well this cooler would work, especially if paired with a high-end thermal interface material and a better fan than the low-profile model it came with.

An interesting fact about that cooler is that Cooler Master had listed it as having compatibility with socket LGA 2011, which was a server/workstation/HEDT socket featuring CPUs of up to 130 W or more. Having seen some reviews of the cooler (see links at the bottom of my original post), I was aware that it would struggle to cope with such heat ouput.

There was a good reason I hadn't use the cooler in over a dozen years!
; )

CountMike said:
Published TDP of a CPU rarely matches real heat output and is usually published for CPU base clock, without boost or OC. at which occasions it could rise double or more because of increased electrical power. That's why I always choose cooling solution according real electrical power requirement consumption with CPU at full boost usage and during prolonged time.
Click to expand...
Believe it or not, Intel's Turbo Boost algorithm is designed specifically to be adapted to the thermodynamics of a given cooling solution. Every heatsink has the ability to absorb more heat in a burst than it it can remove in a steady state. What you're meant to do is configure the PL1 to match its steady state limit (assuming some reasonable ambient temperature) and then set the PL2 and Tau parameters so that the CPU stops boosting before it hits thermal throttling. Now, I'm aware that most people don't set it up that way, in part because thermal throttling on these CPUs isn't as drastic as it used to be .

I'm not well-versed in AMD's boosting behavior, but I know about PPT and how it's usually 35% above the stated TDP of their CPUs. I don't care about PBO, since I plan to never use it.

P.S. I added some plots to my post. I think it's interesting how much more heat the cooler can dissipate from an 8-core workload, compared with one that just keeps 5 cores busy.
 
Since very first Ryzen (later on improved) microcode algorithm counts on and calculate boost according to temperature and amperage under certain load with hard limits.
PBO is not OC, PPT and other power settings are part of it, any change includes changes in PBO. Only difference between it being enabled or disabled that it adjusts algorithm to hard limits automatically but they could be changed manually.
When I had 5800x I found it to be hot by nature. It's actually higher binned 5700(X) which allowed higher power consumption with minimally higher boost frequencies.
It took me a 360 AIO to get it to full performance.
There are couple of things you can do to keep it cool and increase performance at same time. One is controlling voltages thru CO (Curve Optimizer) which scales it according to power usage. Another thing that can help is to set TDP to 65W and if you are limited by temps that can even bring performance up.
 
  • Like
Reactions: bit_user
CountMike said:
PBO is not OC,
Click to expand...
But it can void your warranty? When I go into that menu of my BIOS, there's a massive disclaimer. Anyway, I don't really care about the warranty. My priority for that machine is stability, which is one reason I don't plan to leave this little cooler on there. In fact, the thought never crossed my mind to use it, until I needed a temporary substitute. I will be making a separate post about how I do with the cooler I ultimately hope to use on it!

CountMike said:
When I had 5800x I found it to be hot by nature. It's actually higher binned 5700(X) which allowed higher power consumption with minimally higher boost frequencies.
Click to expand...
For a long time, I had wanted a 5700X. But, then I saw a good deal on the 5800X and figured I could always power-limit it to behave similarly to the 5700X, should I want that. Luckily, I waited long enough to get a B2 stepping:

www.overclock.net

Ryzen 5xxx B2 revision / stepping - post here your...

As title sais, I am interested in documenting the release dates of this B2 revision, so if you got a Ryzen B2, please post here your manufacturing date. For the ones who don't know how to spot it, it's the first four numbers fromt he second row, directly on the CPU.
www.overclock.net www.overclock.net

Mine boosts to 4.85 GHz like no problem!

CountMike said:
It took me a 360 AIO to get it to full performance.
Click to expand...
How do you define "full performance"? I'm curious to see how close I can get with my planned air cooling solution. Note that I didn't buy a top-end cooler, since this is ultimately destined for light-weight server duty, where even a 5600X would be overkill. And yes, I have ECC memory installed and the board very much supports it (ASRock Rack mATX server board).

CountMike said:
There are couple of things you can do to keep it cool and increase performance at same time. One is controlling voltages thru CO (Curve Optimizer) which scales it according to power usage.
Click to expand...
I won't be tinkering with voltages. Anything with a chance of compromising stability is a no go.

CountMike said:
Another thing that can help is to set TDP to 65W and if you are limited by temps that can even bring performance up.
Click to expand...
Yes, I saw in BIOS where I could bias it towards energy efficiency and also where I could override the package power. I just need it to stay in the low-to-mid 80's on a workload like that Ackermann stress test I ran, which is 16 threads doing simple integer computation and branching. I'm pretty confident the next cooler I try will achieve that. If not, I'll indeed dial down PPT to whatever it can sustain at < 85 degrees with a high ambient temp.
 
No, enabling PBO is not going to void warranty because it is not any kind of overclock, actually nothing except physical damage (including visible heat burn) could or at least nobody can prove it, CPU doesn't "remember" what was done with it.
Under "Full performance" I mean nothing that performs less than safe maximum possible, run benchmark programs like CineBench, Passmark Performance test and OCCT which also checks stability.
My PC is used for serious work so performance/$$$ ratio has to be as high as possible.
 
CountMike said:
No, enabling PBO is not going to void warranty because it is not any kind of overclock, actually nothing except physical damage (including visible heat burn) could or at least nobody can prove it, CPU doesn't "remember" what was done with it.
Click to expand...
That certainly doesn't apply to all AMD CPUs, though I'm not sure about mine.

CountMike said:
Under "Full performance" I mean nothing that performs less than safe maximum possible, run benchmark programs like CineBench, Passmark Performance test and OCCT which also checks stability.
Click to expand...
Can you be more specific? Like, what was the package power you saw or what all-core clocks could it sustain?

I have clockspeed data that I didn't include in my plots, because they just became too cluttered. The FFT test case settled at 4.59 GHz, for the 5 cores which were active. The Ackermann test settled at 4.48 GHz for all 8 cores (package power: ~119 W). I only mention this, because it provides a sort of baseline for comparing against the next cooler I'm going to use (and hopefully the one I'm going to leave in place). For completeness, I just did a long-running test of 16 FFT threads, and found the all-core clock settled at 4.22 GHz @ ~121 W of self-reported package power. That was with an ambient temp of 25 C (and, if you've read my original post, you'll know the CPU temp was a hair over 90 C, because that's where my setup is thermal-throttling).

As an aside, I'd say it's not so far-fetched to use this thing on a 130 W LGA 2011 CPU, if it can dissipate 121 W from the far smaller die area of the single-CCD 5800X!

CountMike said:
My PC is used for serious work so performance/$$$ ratio has to be as high as possible.
Click to expand...
At my job, I did some performance tweaks to that machine, since I just plain hate waiting for compiles. Not overclocking, but tweaking with the power limits, cooling, and frequency governor. It's an Intel CPU, though.
 
Last edited:
I just wanted to update this thread with a final addendum, showing the distribution of heatsink compound, like I applied in the original post.

N8pXwzs.jpeg

​

The texture shows where there's more and less heatsink compound. You can clearly see that the spacer block wasn't thick enough to avoid leaving gaps between the heatpipes. Furthermore, I think it's pretty clear that the heatpipes themselves weren't completely flat.

Lastly, as for the amount that I applied, it was almost perfect, but didn't spread quite out to the corners. In hindsight, I think it would've worked better if the "X" part went corner-to-corner. However, I think it wouldn't have made much difference in the outcome, since the only part potentially touching the corners is that aluminum spacer block. The corners are neither where most of the heat is, nor would bridging to that aluminum block change much.

If I really wanted more performance out of this heatsink, I think it's clear that it needs to be lapped. However, I've satisfied my curiosity and it's certainly never going to be sufficient to keep this CPU from throttling at full load (or run at the kind of temperatures I'd like, under more moderate loads). And so, I guess I'll put it away until maybe I want to revive an old Sandybridge or some similar system with a 65W CPU.

fLu2Z8D.jpeg

​
 
Last edited:
bit_user said:
I just wanted to update this thread with a final addendum, showing the distribution of heatsink compound, like I applied in the original post.
​
N8pXwzs.jpeg
​

The texture shows where there's more and less heatsink compound. You can clearly see that the spacer block wasn't thick enough to avoid leaving gaps between the heatpipes. Furthermore, I think it's pretty clear that the heatpipes themselves weren't completely flat.

Lastly, as for the amount that I applied, it was almost perfect, but didn't spread quite out to the corners. In hindsight, I think it would've worked better if the "X" part went corner-to-corner. However, I think it wouldn't have made much difference in the outcome, since the only part potentially touching the corners is that aluminum spacer block. The corners are neither where most of the heat is, nor would bridging to that aluminum block change much.

If I really wanted more performance out of this heatsink, I think it's clear that it needs to be lapped. However, I've satisfied my curiosity and it's certainly never going to be sufficient to keep this CPU from throttling at full load (or at the kind of temperatures I'd like, under more moderate loads). And so, I guess I'll put it away until maybe I want to revive an old Sandybridge or some similar system with a 65W CPU.
​
fLu2Z8D.jpeg
​
Click to expand...
With direct contact pipes it's better to apply paste to cooler base making sure the gaps are filled up without any air bubbles or at least prefill those first.
 
It's always interesting to see how off the old coolers were and how little it mattered for the era they existed. I have two Hyper 212s with the same sort of heatpipe/cold plate design and the older one (circa 2011) is the same as this one, but the newer one (circa 2018) is much flatter.
 
  • Like
Reactions: bit_user
CountMike said:
With direct contact pipes it's better to apply paste to cooler base making sure the gaps are filled up without any air bubbles or at least prefill those first.
Click to expand...
I guess, if you're not first going to do a test application to calibrate the amount, that would be one way to go. You'd want to "spatula" it, so you get the amount right and neither under- nor over- apply it. In general, I'm not a big fan of the spatula method.

Without modifying the base of this particular heatsink, I think I really couldn't have achieved a better result by any means.
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom