Question How much does cooling the underside of the motherboard/CPU typically help ?

[bit_user]

I have an Alder Lake N97 mini-ITX motherboard which has a soldered-down BGA CPU and non-standard heatsink. Upgrading the heatsink isn't a very realistic option, at least not without some custom machining that I'm not setup to do. I have added case fans, but the CPU is still sometimes pegged at the throttling temperature of 95 C, and that's before it's even gotten hot this summer (where I live).

So, my question is how much it might help for me to try various methods of cooling the underside of the motherboard/CPU? Is there any data on this, that people are aware of? Over the years, I've read claims that it's not uncommon for chips and components to dump heat into the motherboard, via their pins.

The system sits with the board in a vertical orientation. I'm considering putting a thin copper heatsink under the CPU, electrically insulated by a thermal pad.

 

[TerryLaze]

Just a heatsink without a fan?
Easiest thing, tie wrap a fan to the heatsink, programs like 'fan control' can manage rpm to keep it from running 100% all of the time as long as there is a free fan connector on the mobo.

You might also want to check in the bios for typical overclocking settings, if it is set to run at full power all the time then it just produces heat for the heck of it.

 

[Paperdoc]

If you do mount a fan on the heatsink as TerryLaze suggests, you may not need third party software to control it automatically, The BIOS may provide all you need for that. Check the manual for tools already present in BIOS Setup. I do not know exactly which mobo you have. The manual I found for ONE such board is here, but that may or may NOT have the same BIOS Setup tools.

https://data-us.aaeon.com/DOWNLOAD/MANUAL/MIX-ALND1_UM.pdf

See its p. 3-6 to 3-8, and I would suggest what it calls the "Smart Fan IV Mode" set of settings IF that appears to be what your mobo has.

ALERT! I just added the web link to the manual I talked about but forgot to plug in!

 

[TerryLaze]

The BIOS may provide all you need for that.

The bios won't relate a random cooler with the temp of the CPU so it will seldomly ramp up or down because it will do so by case temp, with fan control you can do that.

 

[Math Geek]

never actually seen any write up or testing of such a thing. but anecdotally, those cooling pads for laptops do make a big difference when pushing a laptop for long periods.

stands to reason that something similar would help a motherboard. looking at some reviews of motherboards they often use a thermal gun to see where the hot spots are and now we get vrm heatsinks on pretty much anything but a basic model mobo.

i can't think of any negative thing that can come out of blowing some air across the back of the mobo to help remove waste heat. some thermal pads in select places and some airflow could make a worth while difference.

 

[bit_user]

Just a heatsink without a fan?

The board had an active cooler, but that was a thin heatsink with a fan in the middle.

​

I found I get slightly better cooling by removing that tiny fan and putting a 92 mm case fan blowing directly on the heatsink. I'm contemplating whether I should also "stick" a copper heatsink right where that fan was. The main issue with that is the cavity where the fan sat isn't completely flat at the bottom, but is a little concave. So, there probably wouldn't be very good heat conduction right in the center where you most want it!

...if only I had some way to grind the inner heatsink to conform it to the concavity of the outer one. I wonder how long it would take to file away about 1 mm or so of copper from the corners of a 30x30 mm square.

I already dealt with the underside of the heatsink, which originally used a thermal pad and I've now swapped in high performance heatsink compound.

Easiest thing, tie wrap a fan to the heatsink,

Yeah, the place where I mounted the fan is just about 5 mm above the heatsink. So, it's effectively what you're talking about. Also, it's a PWM fan that's connected to the CPU fan header and I've already tweaked the BIOS fan control settings to crank the fan up to 100% by 80 C.

You might also want to check in the bios for typical overclocking settings, if it is set to run at full power all the time then it just produces heat for the heck of it.

Being an industrial board, there's no overclock or over-power settings. I made sure speed step and all the usual CPU throttling options are enabled.

When I query the CPU to tell me package power, the highest I ever see it go is 17 W, but normally stays in the 11-15 W range. Given that its PL1 is set at 12 W and PL2 is set at 15 W, it doesn't seem like the CPU is using too much power, according to that.

https://ark.intel.com/content/www/u...el-processor-n97-6m-cache-up-to-3-60-ghz.html
​

 

[bit_user]

The bios won't relate a random cooler with the temp of the CPU so it will seldomly ramp up or down because it will do so by case temp, with fan control you can do that.

The BIOS settings let you set a pair of temperature thresholds and duty cycle values: one for minimum speed and another for maximum speed. The way I have it set now is to run the fan at 20% at 40 C, which corresponds to a whisper quiet 500 RPM and is sufficient to cool the system at idle, when ambient is only about 22 C.

Given that the throttling temp is 95 C, I figured it's fair to have the fan reach 100% by 80 C, which gives it a little time to react and I don't really want the CPU riding too close to that throttle temperature. Provided the CPU reaches above 90 C anyway, lowering that maximum fan temperature won't help me very much with noise. I might go ahead and bump it up to 85 C, but I still need to improve cooling somehow.

 

[bit_user]

stands to reason that something similar would help a motherboard. looking at some reviews of motherboards they often use a thermal gun to see where the hot spots are and now we get vrm heatsinks on pretty much anything but a basic model mobo.

i can't think of any negative thing that can come out of blowing some air across the back of the mobo to help remove waste heat. some thermal pads in select places and some airflow could make a worth while difference.

I've found discussion threads of this topic on other sites, going all the way back to 20 years ago. Some people dismiss the idea, but others say backside cooling is potentially worth a few degrees C. A lot of people seem to agree with your concern about VRMs dumping heat into the motherboard.

I might put separate heatsinks on each spot, although the VRM sits next to the CPU, such that it's below it when the board is mounted vertically.

 

[Math Geek]

it's all very anecdotal for sure. i don't recall when it stopped but we used to get lots of thermal gun pics of gpu's, mobo's and the like. the back side always had a nice hot spot or 2. one is always the back side of the cpu socket. i don't know anything scientific that would suggest getting that heat away from the back of the board is a bad idea. it's why we blow air across the front of the board, so why not the back.

i'd be curious to know what the results are if you do decide to give it a shot. nothing scientific, but a fan blowing on the back showing a couple degree drop, that's good enough for our basic inquiry

and worth it overall, i'd say

 

[USAFRet]

"how much does it help"?

Lets say it give an extra 5C lowering for the CPU.

So what?
If it changes from 70C to 65C, there is actually zero performance gain.

If you're looking to reduce from 95-100C....there are other problems to tackle first.

 

[TerryLaze]

The board had an active cooler, but that was a thin heatsink with a fan in the middle.

​

You should be showing this pic, if it is this mobo, and maybe measure out the distances of the holes, maybe somebody knows of a cooler that would fit.
Also these are some nice big holes for zip ties, I mean if you have a chunk of cooler (with an attached fan would be best) anywhere in roughly that size you can get rid of the current one and just zip tie it to the CPU, CAREFULLY of course.

 

[bit_user]

"how much does it help"?

Lets say it give an extra 5C lowering for the CPU.

So what?
If it changes from 70C to 65C, there is actually zero performance gain.

Ah, but the noise gains for me are very real! In this case, I'm limited to using only a 15 mm thick fan. Even though it's Noctua, starts getting a very noticeable whine around 1000+ RPM. At that speed the noise is very tolerable, but it gets much louder the faster it spins.

If you're looking to reduce from 95-100C....there are other problems to tackle first.

Well, as I mentioned, a heavy workload can still push the CPU up to the board's throttling temperature of 95 C, and that's before it even gets hot around here. I expect ambient temps of up to +10 C, relative to my initial testing.

 

[bit_user]

You should be showing this pic, if it is this mobo,
...

That appears to be correct.

and maybe measure out the distances of the holes, maybe somebody knows of a cooler that would fit.

I've been down this road with another min-ITX (Gemini Lake-R) board. Didn't find much.

Also these are some nice big holes for zip ties, I mean if you have a chunk of cooler (with an attached fan would be best) anywhere in roughly that size you can get rid of the current one and just zip tie it to the CPU, CAREFULLY of course.

The case allows for only 30 mm coolers or below and I'd rather not switch cases. If I keep the 14 mm fan I have blowing on it, that restricts heatsink size even more.

I even did a test, where I took a copper heatsink made for a LGA 775 server CPU and plopped it right on top of the bare die (these SoCs are made for laptops, and therefore have no IHS), with the motherboard sitting on an open bench (no case). The heatsink was overwhelmed and the CPU eventually throttled back to a package power of 6 W. It's a heavy heatsink, though. Much too tall to fit in my case with a fan. It's designed to be used with a bank of 40 mm fans and ducts forcing air through it, so not too surprising that it's not awesome for truly passive cooling.

BTW, I don't agree that zip ties would be sufficient. I believe mounting pressure does matter, especially because the CPU is a bare die.

 

[Paperdoc]

Some thoughts.

1. It appears the BIOS controls for fans on this mobo are similar to those of many desktop mobos. Is has both a CPU_FAN header and a SYS_FAN header (in the manual I found). In virtually all cases the CPU_FAN header bases its actions of the temperature measured by the sensor built into the CPU chip by its maker. It is actually a TEMPERATURE control loop - that is, its focus is on maintaining the temperature at the sensor on target, and it manipulates fan speed to achieve that. A SYS_FAN header normally does the same job using for reference a different sensor built into the motherboard, so it is for control of case vent fans. OP's post above says he is using the CPU_FAN header to power and control the CPU cooler system.

2. The surfaces of the CPU chip top and the heatsink bottom must be flat and reasonably smooth over their entire mutual contact area for maximum heat transfer rate. Between them one needs a SMALL amount of thermal paste to fill microscopic voids between. Tyipcally a bead of paste the size of one grain of rice is suitable for modest-sized CPU's. Too little will not do the job. Too MUCH is bad because it acts as an insulator, so the thickness of that paste layer should be minimal. As the heatsink is mounted, you may twist it gently to help spread paste, but for the most part the paste is spread evenly when the heatsink is tightened down on the CPU top.

2. How you fasten a FAN to a HEATSINK over a CPU chip is NOT sensitive to contact pressure. Pressure IS important for the force holding the heatsink to the CPU surface. For that reason the fasteners for the heatsink are separate, as you can see from photos of typical motherboards of this type, and arranged to apply uniform pressure over the entire surface. For the FAN mounting, what is important is that it be secure so it does not move around, and so it does not vibrate against other components to cause excess noise. Zip-ties usually will suffice. If vibrational noise is suspected, you might consider iserting tiny rubber pads at the four corners where it contacts the heatsink.

3. OP, you have detailed how you have the new fan set up and its CPU_FAN header automatic controls configured. But you are still worried that is not enough cooling. What you have not told us is what tempertures you DO get now that the new cooler system is in place.

 

[bit_user]

2. The surfaces of the CPU chip top and the heatsink bottom must be flat and reasonably smooth over their entire mutual contact area for maximum heat transfer rate.

These dies aren't flat! I was pretty amazed, but that's what I observed. I'd imagine there's probably some sort of epoxy coating to provide mechanical and/or electrical protection. As this is a BGA laptop SoC, there's no integrated heat spreader (IHS) like you see on desktop CPUs. When you remove the heatsink (i.e. the aluminum block with the fins & fan), here's what's underneath:

​

The board is completely different, obviously. Mine isn't a Latte Panda, but rather what's pictured in prior posts. Also, not shown: the thermal pad I previously mentioned.

If I actually knew what I was doing, here, I might be willing try to lapping the dies. Even if I knew exactly how much to take off and had some form of precision measuring device to ensure I didn't remove too much, I'd still face the issue that the SoC is soldered down and other board components would get in the way. So, it's a moot point. I just thought it was interesting that these "bare" dies don't really seem so bare.

Between them one needs a SMALL amount of thermal paste to fill microscopic voids between. Tyipcally a bead of paste the size of one grain of rice is suitable for modest-sized CPU's. Too little will not do the job.

I used enough. The compound I used isn't very viscous and I use a mash-n-wiggle technique for mounting heatsinks, only stopping once I feel the sensation of the two hard surfaces rubbing together. The heatsink is also torqued down probably too tightly, if anything. I removed 0.5 mm spacers when I removed the thermal pad that the board originally had.

Before I upgraded the heatsink compound, the aluminum heatsink was pretty cool to the touch, even when the CPU was at 95 C. Afterwards, it would get hot, showing that more heat was indeed transferring into the heatsink, as it should.

2. How you fasten a FAN to a HEATSINK over a CPU chip is NOT sensitive to contact pressure.

The fan is actually attached to a grille on the case. So, they're not even touching. Probably a few mm in between. Non-issue.

For the FAN mounting, what is important is that it be secure so it does not move around, and so it does not vibrate against other components to cause excess noise. Zip-ties usually will suffice. If vibrational noise is suspected, you might consider iserting tiny rubber pads at the four corners where it contacts the heatsink.

I used a Noctua fan with its rubber anti-vibration grommets. The fan is secured to the case using ties, as the grille holes are too small for their rubber nibs or conventional screws (and I didn't want to make it ugly by using a drill to widen them). I believe there's no significant vibration being conducted from the fan frame to the case.

The fan noise is coming from the fan, itself. This is partly confirmed by observing the amount & character of noise it makes, even when you're just holding it in your hand. Maybe because it's a thin fan (14 mm), the blade is wobbling on the bearings? I wonder... Other reviewers have also complained about a whining from this fan. My other thought is that it could be coil whine due to the PWM harmonics. I'd hope Noctua does some kind of waveform smoothing, but who knows?

3. OP, you have detailed how you have the new fan set up and its CPU_FAN header automatic controls configured. But you are still worried that is not enough cooling. What you have not told us is what tempertures you DO get now that the new cooler system is in place.

Not sure I follow you. I said the CPU will reach its thermal throttling temp of 95 C, under certain workloads. Weirdly, the best workloads for doing this are 2-thread ones. Not 1-thread, which are frequency-limited, or 3/4-thread, which disperse the heat over a wider area, but it seems 2-thread is perfect for quickly hitting the power limit and still concentrating the heat in a relatively small area of the die.

In GPU workloads dissipating more total power I do not get thermal throttling, probably due to the much larger silicon area over which much of the heat is being generated.

 

[bit_user]

i'd be curious to know what the results are if you do decide to give it a shot. nothing scientific, but a fan blowing on the back showing a couple degree drop, that's good enough for our basic inquiry

and worth it overall, i'd say

I measured the standoffs in my case at about 4.5 mm. So, I just ordered some copper heatsinks that are allegedly 4 mm high and I've already got a high-performance thermal pad that's 0.5mm thick. The idea is to hold the heatsinks in place (one for the VRM and one for the CPU) by squeezing them between the motherboard and case. There's not much airflow back there, nor anything I can easily do about that, so the dominant effect might be simply conducting the heat out to the solid case side panel.

If all goes to plan, I'll be trying it this weekend and will post up what I'm able to quantify.

Also, does anyone have any suggestions about how to further insulate motherboard components from shorting out? I was going to rely on the thermal pad, but I'm a little concerned about some sharp bits managing to poke through to the copper surface of the heatsink. I seem to recall I once read about people using direct die cooling with conductive heatsink compound (liquid metal?) applying clear fingernail polish to the diodes or resistors mounted on-package. Is that right? Is there any better sort of paint-on electrical insulator?

Update: I am going to try this stuff:

Amazon.com

 

[TerryLaze]

BTW, I don't agree that zip ties would be sufficient. I believe mounting pressure does matter, especially because the CPU is a bare die.

Yeah on a nakey die one would probably kill the poor CPU by trying that anyway.

Also, does anyone have any suggestions about how to further insulate motherboard components from shorting out? I was going to rely on the thermal pad, but I'm a little concerned about some sharp bits managing to poke through to the copper surface of the heatsink. I seem to recall I once read about people using direct die cooling with conductive heatsink compound (liquid metal?) applying clear fingernail polish to the diodes or resistors mounted on-package. Is that right? Is there any better sort of paint-on electrical insulator?

It's called UV curing (electrical) potting but I don't know which brand is any good.
That's what the manufacturers use when they have to protect anything.
If you have an electronics store close by you can chat them up to ask if they know of one that can take a lot of abuse.

 

[bit_user]

It's called UV curing (electrical) potting but I don't know which brand is any good.
That's what the manufacturers use when they have to protect anything.
If you have an electronics store close by you can chat them up to ask if they know of one that can take a lot of abuse.

Thanks for the info. I think I found one that's a good fit (see updated post). It's listed as being able to withstand temperatures up to 130 C.

 

[bit_user]

Also, does anyone have any suggestions about how to further insulate motherboard components from shorting out? ... Is there any better sort of paint-on electrical insulator?

Update: I am going to try this stuff:

Amazon.com

Wow, that stuff has some fumes! Went outside and used a half-face respirator with organic vapor filters. Later on, with the board still outside, I removed the respirator and the fumes were still enough that I setup an exhaust fan in a window.

As for other board prep, I started filing down some of the capacitor leads poking through from the VRM. After nicking the board a couple times, I decided it'd be worth digging out my Dremel tool. The grinding wheel made quick work of the leads, but required a steady hand and managed to nick the board in a couple more places with it that didn't appear to cross any traces.

After that, with the board fully dusted, I had to fire it up and stress test it, to be sure it still worked. That's the point where I painted on the the insulating varnish. I've been stress testing it some more (in the room with the exhaust fan), to generate some heat and hopefully accelerate the curing process.

Next, I measured out where the heatsinks would contact the case. Upon examination, I noticed bottom of the case is painted. The Dremel tool made quick work of stripping off the paint, where the heatsinks will go. Also used the respirator for this.

During the first test run, after grinding and before painting, I did some careful and informal "touch" testing of spots on the board. The spot under the CPU is unsurprisingly super hot. The VRM much less so, but then again it is a lower-power board, so I guess that makes sense. I'm still planning on heatsinking that area, but I'm not going to bother doing it as a separate step.

BTW, here are the heatsinks I'll be using. The more squarish one is going under the CPU, while the M.2 heatsink is going under the VRM.

https://www.amazon.com/dp/B081N7LFKY

https://www.amazon.com/dp/B083FVMN7N
​

The heatsink shapes are good for my purpose, because there's not much air movement back there and not a lot I can do about it. So, the main role they'll be playing is to conduct the heat into the case (hence the part about stripping the paint), for which that chunky style is much more suitable than the sort of fins you might normally want in a heatsink.

I did a test fit and while there's still probably at least a half mm of excess height from the surface-mount components I failed to account for, the board appears to have ample flex to accommodate. This is going to happen.

 

[bit_user]

Update #2​

The patient survived the surgery, but there's still a little post-op needed before I have final results.

The final issue is that when I decided the 0.5 mm thermal pad I had should be sufficient, I failed to actually measure the surface mount components on the board. One of them under the CPU protrudes nearly an entire mm. So, you can imagine that not much of the thermal pad is making contact, with that big tent pole in the way!

A couple more random observations in case someone else tries this​

One of the heatsinks (the mSATA one, to be specific) had an adhesive thermal pad preapplied. I didn't want to use it, so I cleaned it off. In the course of doing this, I noticed it had one significantly raised edge, probably from where it was cut off a bigger sheet. I filed it down and then gave all of the edges of both heatsinks a nice bevel, just to be sure there were no other raised edges that could interfere with good contact or even damage the motherboard.

Another thing I overlooked was that I kept the BIOS battery in the board, while I was working on it. It didn't help, because the board went through a bit of a "spasm" the first time I booted it. I'm guessing I probably shorted something out, at some point, that got it into a bad state. Anyway, I didn't initially notice and the result is that my "before" testing used the default fan curve, instead of my modified one. Therefore, I'm keeping their overly-aggressive default fan curve for all of my testing, but it's not a real issue because the CPU has no trouble getting too hot, even with that fan curve.

 

[bit_user]

Okay, so I just layered on another ply of the same 0.5 mm thermal pad, and that seems to have done the trick!

There's a lot to tell about how things went. I have pics and plots, but I'm out of time for this weekend. Since I hate to be kept in suspense, I'll go ahead and give the ultra-short version: I got a 10 C 7 to 8 C drop, between peak temps of the 4-core, CPU-only workload. The bad news is that the 2-core, CPU-only and power-unlimited GPU + 4-core workloads still hit 95 C. So, there's more room for improvement.

A 10 C drop might sound hard to believe, but keep in mind there are 3 things about my setup, making it ideally suited for this sort of thing, that don't apply to the typical performance desktop.

1. The primary heatsink I'm using is probably worse than anything you could get away with, on modern x86 desktop CPUs. This means there's a lot of untapped cooling potential.
2. The CPU I have is BGA. There's not a socket in the way, providing thermal insulation. The solder joints each provide a somewhat direct path for heat to enter the motherboard.
3. The CPU I'm using is also low-TDP. Along with point #1, this means a little improvement can go a long ways

So, I would caution anyone against anticipating a similar improvement on their desktop board. Still, I think backside cooling has some potential and it might be worth at least trying to get some airflow back there.

On the flip side, because my CPU is fairly low-power, the VRM really doesn't get too hot and backside-cooling it probably didn't help much. On a desktop, that situation might be a bit different.

 

[bit_user]

Pictures & Plots​

First up, a bit more on the prep I had to do. I mentioned painting the MG Chemicals 4226A Clear Insulating Varnish on the board, after grinding down the capacitor leads and solder blobs of the VRM. However, there was also some preparation needed by the case.

The case is a Silverstone PT13, which has a thick, anodized aluminum top/front and a thin, painted steel bottom & sides piece.

https://www.silverstonetek.com/en/product/info/computer-chassis/PT13/

It's so small that it won't even accommodate the full-height mini-ITX I/O shield. Its cooler clearance is just 30 mm!

So, to achieve good thermal conduction to the case, I figured it might be necessary to strip off the paint. My Dremel tool made pretty quick work of that, although I went to the trouble of using several different sandpaper grits to try and make it as smooth as possible. I strongly recommend also wearing a face mask, for this part, since stripping the paint produced dust containing probably lots of nasty stuff. The OV (Organic Vapor) filters I had on my respirator also provide P100 particle filtration. You can buy these respirators on Amazon or at a hardware store (Home Depot, Lowes, etc.) for like $20 - $30. It's pocket change, compared to what the health care costs might be for being reckless about health & safety.

​

I should point out that the low-profile 92x14 mm Noctua fan is merely attached to the case by tying it to the vent holes. The case has no actual mounting position for any fans.

Now, I was ready to place the heatsinks, shown here with the thermal pad applied.

​

The thermal pad used was Arctic TP-3:

Amazon.com

As mentioned above, I actually had 0.5 mm thermal pad and decided this was insufficient due to the surface-mount components rising up to 1 mm off the board. So, I removed the board and added a second layer of thermal pad, taking care to smooth out any bubbles & other damage to the first layer.

Between the heatsinks and the case, I applied a dot of Arctic MX-4 thermal compound on each pillar, since I've switched to using MX-6 and have a bunch of MX-4 left over. This wasn't an option for the board side, due to all the surface mount components of various heights.

Finally, here's the before/after data for 1 thread and 4 threads. The CPU has 4 cores/threads, in case you didn't know.

​

So, we see about an 8 degree C improvement in peak temperatures, for single-threaded, and about 7 degrees C for multithreaded. As explained in the previous post, I wouldn't expect the typical performance desktop PC to benefit as much. BTW, I did take care to maintain consistent conditions between the two sets of tests. Ambient was about 22 C +/- 0.5 degrees.

Again, the CPU appears to be throttling at 95 C. So, the backside cooling mod was relevant for the single-threaded case, as well as reducing fan noise. Not shown: I tried running a 4-thread CPU job + iGPU benchmark and found that even with the backside cooling mod, that still throttles! Between that and in the pursuit of even less fan noise, my quest continues...

P.S. you might be wondering why the 4-thread benchmark runs at lower temperatures than single-threaded. I haven't done a detailed analysis of frequency behavior, but I've informally noticed that my 1-thread job runs at up to 3.3 GHz (I guess it never hits 3.6 GHz because it's a SSE/AVX-heavy job?), while the 4-thread job runs at about 2.83 GHz and won't even exceed 2.9 GHz if I significantly raise the power limits! So, basically the 4-core job is burning a similar amount of power as the 1-core job, but distributed over 4x as much area. The fact that it runs cooler tells me the problem isn't between the heatsink and the air, but rather with transferring heat from the die into the aluminum heatsink. For this reason, I plan to try inserting a copper shim between them (see below), which should serve a similar purpose as the IHS on a desktop or server CPU.

 

[bit_user]

Next Steps​

In case anyone is curious, here's what's next on my list of things to try:

1. Upgrade the fan to NF-A9x14 HS-PWM, which has 28.7% more static pressure and a 13.6% higher maximum speed.
2. Block some of the vents on the side & top, to create more pressure in front of the heatsink, thus hopefully forcing more airflow over it. Note that the fan doesn't entirely cover the heatsink, so I think the potential exists for achieving more airflow across it.
3. The store that sold me the motherboard (MITXPC.com) managed to get a passive heatsink for me, which has fins all the way across and no circular carveout for the embedded fan. They're kindly sending it free of charge!
4. I bought a couple 1 mm copper shims to put on top of the dies (1 each), which should help mitigate thermal hot-spotting. I read that the IHS of Rocket Lake is only 1.2 mm thick, so it seems like even 1 mm of copper is enough to provide some real benefit. Thanks to @thestryker for suggesting this.

So far, I've already completed #1. I haven't done the same degree of data analysis, but it seemed to provide a solid 1 C temperature drop. Yes, it's also louder, but the top priority is to minimize throttling. Anything I do to improve cooling efficiency should help limit fan speeds and mitigate the noise.

Thanks for your interest. I don't plan to post any further updates to this thread, since the original question has now been answered to my satisfaction and hopefully that of everyone who's followed or contributed. Feel free to PM me, if you have any questions not related to the core topic of backside cooling, or if this thread gets locked.

P.S. The support team at MITXPC.com also engaged the manufacturer (Jetway) to get me a BIOS image for the board which enables in-band ECC! I'm waiting to install it, until after I finish tweaking with the thermals, since I want to limit the number of variables. In case anyone wonders why I'm going to so much trouble to make this board work better, there aren't a lot of Alder Lake N boards/systems which support in-band ECC and most of them have soldered-down LPDDR5 or DDR4 memory. So, that makes this board quite a rare bird! Plus, I do really like that case, in spite of all the grief it's giving me.

 

[bit_user]

Backside Case Temperatures​

There's one more relevant piece of information I collected but forgot to include, above. I measured the actual backside case temperatures!

I have an old IR thermometer "gun" (operating range: -29 C to 760 C) on a slightly low 9V battery, so I wouldn't trust the absolute measurements from it. However, it does show across-the-board increases of the hot spot temperature, especially under load.

Run​	Idle (C)​	4T (C)​	Delta (C)​
Baseline	25.1​	29.0​	3.9​
Experimental	28.2​	38.0​	9.8​
Increase	3.1​	9.0​	5.9​

Idle increased 3.1 C and case temperature under load increased almost triple that: 9.0 C! This further confirms that a significant amount of heat is being dumped to the case. As the design of the case doesn't promote a lot of airflow under the board, this result is unsurprising.

Here's where I should probably add that, for all of my tests, the case was kept in the vertical orientation, with about 11 mm of clearance maintained below the bottom air vents. I was not using the included plastic stand, since it doesn't permit those vents to breathe freely (this could be mitigated by attaching large rubber feet, underneath it).

 

[mac_angel]

I have an Alder Lake N97 mini-ITX motherboard which has a soldered-down BGA CPU and non-standard heatsink. Upgrading the heatsink isn't a very realistic option, at least not without some custom machining that I'm not setup to do. I have added case fans, but the CPU is still sometimes pegged at the throttling temperature of 95 C, and that's before it's even gotten hot this summer (where I live).

So, my question is how much it might help for me to try various methods of cooling the underside of the motherboard/CPU? Is there any data on this, that people are aware of? Over the years, I've read claims that it's not uncommon for chips and components to dump heat into the motherboard, via their pins.

The system sits with the board in a vertical orientation. I'm considering putting a thin copper heatsink under the CPU, electrically insulated by a thermal pad.

I've done a bit of it in the past and find it helps. It really depends on the motherboard though, and what they have on the backside of the CPU socket. When I have been able to, I've placed a 10cm x 10 cm aluminum heat sink on it with double sided thermal tape. The motherboards that I've seen lately that have the metal backplate have 4 rivets sort of welding it on, and they aren't very smooth, so it's very hard to make good contact. I have looked at replacing the back plate with something from AlphaCool or another brand that's escaping my memory at the moment. The thing is, they still stand somewhat away from the motherboard and you'd need a pretty big, non-conductive thermal pad to go along with it. It seems rather expensive to complete for very diminished returns compared to a few generations ago.