Arctic MX-5 vs Indium Foil DIY thermal pad

Also your core 1 temps dropped below your core 0 temps after the first test, and core 4 and core 6 also went down.
Core #2 is often the hottest core on a bunch of Intel lines.
 
Its great you have gone this far but you must record ambient temp each time for the figures to mean something.
Also repeat conditions that can affect heat transfer in the room. Like whether the room door or windows are open / closed.
Water loops are slow to change, it can take longer to get a viable result otherwise which might cause a result to be taken too early.

Oh trust me I do. The ambient temp is taken inches from the intake fan on each run, and I manage room air flow so that there is as little fluctuation during the run as possible. It took some trial and error because this machine is very good at heating up its section of the room during a Prime 95 run.
 
Oh trust me I do. The ambient temp is taken inches from the intake fan on each run, and I manage room air flow so that there is as little fluctuation during the run as possible. It took some trial and error because this machine is very good at heating up its section of the room during a Prime 95 run.
Cool.
Any chance we can have the ambients with the results? :)
 
Also your core 1 temps dropped below your core 0 temps after the first test, and core 4 and core 6 also went down.
Core #2 is often the hottest core on a bunch of Intel lines.

Things that make you go "hmmm." I wonder if maybe Prime 95 is not balancing the load as evenly as I would need for the resolution I'm trying to achieve.
 
Cheers.
I wouldnt worry unless the delta gets larger, for now its within margin of error and doesnt look like its going anywhere strange.
Longer term results are more likely to show larger changes.
...unless you are prepared to push the CPU to throttling temp to speed things up.
Not something I would do btw, I'm partial to my gear working for a long time, but thought I'd ask :)
 
Cheers.
I wouldnt worry unless the delta gets larger, for now its within margin of error and doesnt look like its going anywhere strange.
Longer term results are more likely to show larger changes.
...unless you are prepared to push the CPU to throttling temp to speed things up.
Not something I would do btw, I'm partial to my gear working for a long time, but thought I'd ask :)

Hahaha, nah, I like this particular chip. I've had almost a dozen different HW-E chips in this machine over the years, and this one is overall the best. Actually, the other 5960X I have would be at throttling temps running these same tests. I think the indium solder in that one has had it. I'm going to sand straight through the IHS and see how it does direct die. It's a great clocker even with the ridiculous temps it gets to.
 
Last edited:
Hahaha, nah, I like this particular chip. I've had almost a dozen different HW-E chips in this machine over the years, and this one is overall the best. Actually, the other 5960X I have would be at throttling temps running these same tests. I think the indium solder in that one has had it. I'm going to sand straight through the IHS and see how it does direct die. It's a great clocker even with the ridiculous temps it gets to.
Is there no proper way to delid it without trying to sand through the whole ihs? On soldered lga 1366 cpus I was able to torch the ihs and rotate it off after breaking the glue holding it on.
 
Is there no proper way to delid it without trying to sand through the whole ihs? On soldered lga 1366 cpus I was able to torch the ihs and rotate it off after breaking the glue holding it on.

Oh sure, there are other ways to delid one of these, although I've never personally delidded a soldered chip, I would assume you just have to keep it hot enough and try not to burn your fingers, haha. I figure that by sanding it down instead, I'll be able to use the remaining rim of the IHS to help stabilize the cold plate for direct-die and hopefully not crack the die. It's about 90% done, one of these days I'll finish it off.
 
Just a quick follow-up to this thread. Today's raw data is attached below, the table is Core Temp Delta from Water Temp. Ambient temp was 27.3C, Water temp was 35.3C. So far, everything is still within margin of error. As I look over all my data from before and after (including all the runs that I haven't posted), the average seems to be a ~2 degree improvement over MX-5, which was itself a ~1.5 degree improvement over MX-4. Hopefully this will continue to hold up long-term.

June 27June 28June 29June 30
Core 020.721.021.520.7
Core 121.720.420.820.5
Core 226.726.827.627.3
Core 324.023.923.824.3
Core 423.823.623.324.3
Core 523.423.323.423.7
Core 624.722.623.524.6
Core 721.421.821.722.2
Core Average23.322.923.323.5
 

Attachments

  • thermals 6-30-21 2130.png
    thermals 6-30-21 2130.png
    1.3 MB · Views: 0
Nice.

I wonder, did you take measurements of idle temps?
It would be interesting to see if/how much they vary over time, as this will give a quicker indication of whether there will be much change in load temp after the CPU sees hotter temps at times.
Though the relative temp difference will be much smaller for load than at idle.
Even worse though, the whole apparatus is under pressure so there may never be a gap and thus always a good contact, so temps will not vary.
My interest in this is to see if its worth pursuing!

What I would look for eventually under load:
There will be hot days that cause the indium to expand more, pushing some of it out the sides, 'potentially' leaving a slight gap between the core and heatsink at lower temp.
When it is run on cooler days the indium wont have quite as good a contact with the core.
The difference may not even be measurable, it may not even exist, but it could surprise us.

ie
Consider the difference in contact pressure and resulting temp when the CPU has been run at a hotter than normal temp:

Expansion/contraction is 0.1% of total volume over approx 37C.
Lets say a hot day is 10C above norm.
This will give an extra expansion of 0.1% / (37/10) = 0.027%
This assumes there is never a circumstance that allows the CPU to get a lot hotter.

The max possible gap at normal load temp with a 0.2mm sheet will become
(indium in use will be a bit thinner than that)
0.2 x 0.00027 = 5.4 x10e-5mm
This is such a small gap that a lot of the indium will still be in contact and it might not even matter.
And with it being under pressure, can there ever actually be a gap?

My feeling is there will be nothing to see other than Indium being a great thermal pad (if handled well).
I dont mind being proven wrong :)


note:
I've no idea where I read the following so please take with a pinch of salt...
Indium may cold weld itself to other metals over time, this would lead to rock solid performance all the time.
I dont know if its selective, ie some metals will not bond. Nor how long it would take.
If true, performance could improve a little over time!
But then we'd be looking for a method how to separate the CPU and heatsink easily and cleanly.

Searching for 'indium cold weld' brought up a related article, it cold welds to itself at room temp and extremely low pressure:
https://www.indium.com/blog/cold-welding-use-this-unique-indium-property-to-your-advantage.php
This is of interest because Indium will always be trying to form itself into a blob, very slowly.
It probably wont run away like many heatsink pastes that leave the core uncovered.
(Though maybe it can traverse along metals like the underside of the heatsink, what does it prefer, itself or other metals?)

So all my concerns above are probably mute lol.
Oh well, it was fun writing it.
 
Last edited:
The trend I noticed seemed to be a slight improvement over the course of the first day, which then plateaued. So far, I think it's a good alternative to paste and conventional pads, as long as you are able to apply enough mounting pressure to make it deform.
 
Update: No news is good news. (y)

I didn't see the question about idle temps - they are very good, the lowest ever on this machine. And yes, that's 80 watts at idle. :eek:
 

Attachments

  • idle temps 7-4-21-0842.png
    idle temps 7-4-21-0842.png
    1.2 MB · Views: 0
Just to update this thread, the indium foil has been installed for 2 weeks, and there has been no performance regression.

Here's a shot from the run I just did.
 

Attachments

  • thermals 7-8-21 1837.png
    thermals 7-8-21 1837.png
    1.4 MB · Views: 0
Just to update this thread, the indium foil has been installed for 2 weeks, and there has been no performance regression.

Here's a shot from the run I just did.
Cool. The one I ordered on June 27th never "arrived" from USPS and they didn't even do tracking so now they had to send out ANOTHER ONE, this time with tracking. I freaking HATE how EVERYTHING takes forever.
 
Cool. The one I ordered on June 27th never "arrived" from USPS and they didn't even do tracking so now they had to send out ANOTHER ONE, this time with tracking. I freaking HATE how EVERYTHING takes forever.

The same thing happened to me. Apparently the first one was misaddressed. Maybe someone over there needs new glasses.
 
I tried the indium foil on my GTX 1070 (MXM) laptop card. Saw something was going to go south when the idle temps were several C higher than TFX right upon turning on the system.
Ran Heaven benchmark at max fan speed,
81C in 10 seconds and rising. Very quickly shut it down and went back to Thermalright TFX.
Maybe the foil might work on a desktop CPU and IHS possibly. Definitely NOT on a video card! Glad I didn't test it on my 3090. Would not have gone well.
Even the Panasonic EYG-R graphite pad was significantly better. The indium didn't seem very soft at all (although it cut extremely easily) and the heatpipes were barely warm at all so heat wasn't getting transferred very well.
 
Thanks for testing that. This confirms that it needs a lot of mounting pressure to work. I would only use it on a CPU IHS and preferably with a cooler that you can really crank down. Might be better on HEDT/Threadripper too. Mainstream sockets and coolers that have spring-loaded screws, flimsy back plates, or those Intel plastic pegs (!) might not work as well. It would need to be tested.

Another issue with using it with a graphics card is that it will create space between the memory and the VRM and the heatsink. I would only use liquid metal or high quality paste on a GPU.
 
Yeah, thank you for that info Falkentyne
I've been following this thread and wondering how it would do for direct-die in a laptop 'cause my CML-H i7 needs help but I guess liquid metal it is.
 
don't miss that! direct die + no frame + those terrifying clips on Socket A / 370 made for some nail-biting heatsink installs...
only crushed one, with one of those giant all copper heatsinks we used back then. the other prob was not jamming a screwdriver through the board, until i realized a multi-bit screwdriver with no bit works great.
 
Thanks for testing that. This confirms that it needs a lot of mounting pressure to work. I would only use it on a CPU IHS and preferably with a cooler that you can really crank down. Might be better on HEDT/Threadripper too. Mainstream sockets and coolers that have spring-loaded screws, flimsy back plates, or those Intel plastic pegs (!) might not work as well. It would need to be tested.

Another issue with using it with a graphics card is that it will create space between the memory and the VRM and the heatsink. I would only use liquid metal or high quality paste on a GPU.

Yeah, thank you for that info Falkentyne
I've been following this thread and wondering how it would do for direct-die in a laptop 'cause my CML-H i7 needs help but I guess liquid metal it is.

So,
Using the indium foil as a sandwich between liquid metal on both sides seems to work.
That is, applying a thin layer of LM to the GPU 1070 core, then very carefully wiping LM on one side of the indium foil, then quickly planting it and holding it down with a lint free swab, painting more LM on the top side of the foil quickly, before the LM has time to react to the indium and start making it soft and making it start flaking apart (I tested that on the old piece on the countertop, eventually the indium started becoming VERY soft and then started flaking apart!), then LM on the heatsink, and then attaching the entire thing together.

This seems to work, although the temps are higher than direct LM, but I'm curious as to its longevity, which is why I'm testing that. Wondering if it will get worse with time, if it will get better as the indium starts melting more, or if it will just slowly burn up.
I have kapton tape and a foam dam around the GPU so it won't go anywhere at least.

Right now it's like 1C better than Kryonaut Extreme and TFX, but it's only been running 10 minutes.
I just want to see if the indium remains hard enough to maintain its integrity or if it will degrade somehow or if it will get any better.
 
Gallium and indium form a liquid alloy that is 74.5% gallium and 25.5% indium... and then there is an unknown amount of tin in the original liquid metal as well, so the exact proportions cannot be predicted. I would imagine that some portion of the indium is combining with any "extra" gallium in the liquid metal to create additional liquid metal, but with different properties from the original, while a portion of the indium remains solid. My guess would be that the reaction takes place quickly and then stabilizes. Sounds risky, it's a good thing that you took precautions.
 
Thanks for testing that. This confirms that it needs a lot of mounting pressure to work. I would only use it on a CPU IHS and preferably with a cooler that you can really crank down. Might be better on HEDT/Threadripper too. Mainstream sockets and coolers that have spring-loaded screws, flimsy back plates, or those Intel plastic pegs (!) might not work as well. It would need to be tested.

Another issue with using it with a graphics card is that it will create space between the memory and the VRM and the heatsink. I would only use liquid metal or high quality paste on a GPU.
Hurm makes me want to try some on my 3960 that thing screws down with a lot of force....will it melt a nickel water block or the CPU or anything like that?
 
6 MONTH FOLLOW-UP

It looks like my initial concerns about mounting pressure are confirmed. As I mentioned, I was able to remove my CPU and cooler several times with no issues, but I never removed the indium foil from the CPU. Well, curiosity finally got the better of me, and I finally peeled that shit off of my IHS. Here's what I found (sorry, no pics). The Indium was indeed bonded to the IHS, but only directly over the CPU die itself - in fact, I would say that the rectangular area of the indium that was bonded to the CPU probably did not cover the entire die, but probably did cover the entire L3 cache region. OK, so what happened? The IHS and cold-plate were mostly intact (which are both bare copper), with some very slight amalgamation on the IHS where the bonding occurred. It was similar to what you'd see with liquid metal, but to much less of a degree. It was removed with about 15 seconds of lapping with 1500 grit wet/dry. Now for the problem. My CLC 280 cooler's mounting bracket has relaxed a bit and no longer provided sufficient mounting pressure to bond a new piece of indium foil to the IHS. At first cooling performance was terrible (about 5 degrees worse in Prime 26.6), but over the course of 11 days, it improved to within 2 degrees of the previous performance. It stopped improving after 2-3 days, but I left it on longer just to make sure. It apparently conformed, but it would not bond. This pissed me the fuck off. I even tried to slightly bend upwards the four corners of the bracket holding my block on to get the mounting pressure back, but there was nothing I could do short of buying the cooler again for $79 just for a new bracket, just to get those 2 degrees back. No way. I'm crazy, but I'm not that crazy.

But wait, what about that Arctic MX-5 I mentioned at the beginning of this thread? Well, folks, my 20g tube of MX-5 saved the day, but it took some patience. At first, it was worse than my current and underperforming indium TIM by about 1 degree, but after giving it 3 full days to cure (and I intentionally heat-cycled it repeatedly by blasting Linpack Extreme and then shutting the machine completely off many times during those 3 days), it matched or slightly beat the performance I was getting with the indium before I decided to fuck with it and peel it off. The improvement with a full cure was 3 degrees.

Moral of the story? There are 2 actually.

1) Indium foil can be a very good maintenance-free TIM that can match the very best paste, but it's finicky as fuck, and it probably won't work well for you unless your cooler is mounted to your CPU with the jaws of life.

2) Arctic MX-5 is nothing special when it's first applied and will probably not win any first-place trophies in any review site thermal paste roundup as a result, but the stuff is PURE GOLD once it finally cures. It takes time to get there.

Here are the results from fully-cured MX-5. Package temp of 27.0 degrees above the water temp, and 35.1 degrees above ambient with a 200 watt load (Prime 95 26.6). This is as good as it gets with this CPU sample and this cooler. If you look through the previous testing I did in this thread 6 months ago with the indium TIM, it was typically allowing for package temps of 27-28 degrees above the water temp and 35-36C above ambient with the same load.

thermals 1-13-22 1120a.png
 
Last edited:
Thank you for doing this.

You're welcome. It's been a fun side project for me, even though it came to a rather anticlimactic end. Now I'm going in a different direction, and I've been researching the compositions of various thermal pastes. Most commercial pastes have a readily-accessible MSDS that often reveals the entire formula. I noticed that MX-5 has one extremely conductive ingredient that I have not yet found in any other product, 4.5% boron nitride, which has a thermal conductivity of 751 W/mK. I'm tempted to source this stuff if I can get a small enough particle size for it to be effective, and try to make my own frankenpaste with it, first by just adding some more of it to MX-5, and if that's successful, then I'd like to experiment with making a paste from scratch. I don't need it for my 5960X obviously, with its big, slow, lazy transistors, but as thermal density increases with each generation, it might start to matter a lot. I've played with liquid metal a lot too, and I have a love-hate relationship with the stuff. I would much prefer to use a product that doesn't form alloys with other metals, causing warranty concerns, long-term performance concerns, and resale value concerns.
 
Last edited:
This was a very interesting read. Reminds me of the days I experimented with paste/HS combinations on Pentium III CPUs. Thank you for the effort.
 
So I mentioned in the thermal paste thread that I ordered some Indium foil to attempt to use like a thermal pad. Well it's here, and I had to chance to run it against my new favorite thermal paste.

The sheet I received was 100mm x 100mm x 0.2mm. It cost me $20, and it's enough to do 9 CPUs, so at ~$2 per application, it's not the most expensive TIM I've ever used, but it's definitely not the cheapest either.

First things first - 0.2mm is a little thicker than I thought it would be, but it ensures that it will be thick enough to provide consistent contact.

Test Bed:

Intel 5960X @ 4.2GHz / 1.175V manual voltage, all EIST and C-States disabled
ASUS X99-A USB 3.1 Motherboard, BIOS 3701
EVGA CLC 280mm mounted as front exhaust
4X Arctic P14 PWM rad fans in push/pull @1470RPM (90% PWM)
Coolermaster HAF-XB EVO with Noctua A20 Chromax top intake fan @ 810RPM (90% PWM)
Prime 95 26.6 Small FFTs
15 minute run time to ensure steady state
Ambient temperature measured 3 inches from top intake fan.
Temperatures reported are an average over the final 60 seconds as reported by HWINFO64.

Arctic MX-5

Ambient Temp: 23.6C
Water Temp: 31.6C
Package Temp: 61.7C
Cores: 54.4, 53.6, 61.7, 56.9, 58.0, 57.3, 58.6, 54.6
Delta from ambient temp: 38.1C
Delta from water Temp: 30.1C
Average Power: 198.0 watts
Thermal Resistance (system): 0.19242 C/watt

Indium Foil [0.2mm thickness]

Ambient Temp: 23.8C
Water Temp: 31.6C
Package Temp: 60.5C
Cores: 53.3, 54.1, 60.5, 58.0, 57.7, 57.8, 57.5, 55.1
Delta from ambient temp: 36.7C
Delta from water temp: 28.9C
Average Power: 197.5 watts
Thermal Resistance (system): 0.18582 C/watt

The improvement over good paste was 1.4C - this is about the same improvement I would have expected from liquid metal based on my previous testing.

I'm gonna chalk this one up in the WIN column.


That's pretty cool. I'm going to have to try this at some point.

I also wonder if a thinner foil would be more effective.

New ideas like this are great. Reminds me of the early days of the hobby before all the solutions were commoditized.

I also wonder if this would be effective on a bare chip surface, like in a fullcover GPU block, or if it is really only suitable for heatspreader use. (I wouldn't want to risk cracking the die with the force of squishing metal, even a soft metal like Indium.
 
Its great you have gone this far but you must record ambient temp each time for the figures to mean something.
Also repeat conditions that can affect heat transfer in the room. Like whether the room door or windows are open / closed.
Water loops are slow to change, it can take longer to get a viable result otherwise which might cause a result to be taken too early.

I don't really think the ambient is the most important temp here.

I'm most interested in the delta between the core temp and the coolant temp.

Assuming load is the same power output and thus produces the same amount of waste heat, we only really need to know the Delta T to the coolant to know if anything is changing or if it is stable.
 
The trend I noticed seemed to be a slight improvement over the course of the first day, which then plateaued. So far, I think it's a good alternative to paste and conventional pads, as long as you are able to apply enough mounting pressure to make it deform.

I imagine this could be caused by the clamping force of the block mount continuing to deform the soft Indium to fill out the cracks and crevices better over time.
 
I tried the indium foil on my GTX 1070 (MXM) laptop card. Saw something was going to go south when the idle temps were several C higher than TFX right upon turning on the system.
Ran Heaven benchmark at max fan speed,
81C in 10 seconds and rising. Very quickly shut it down and went back to Thermalright TFX.
Maybe the foil might work on a desktop CPU and IHS possibly. Definitely NOT on a video card! Glad I didn't test it on my 3090. Would not have gone well.
Even the Panasonic EYG-R graphite pad was significantly better. The indium didn't seem very soft at all (although it cut extremely easily) and the heatpipes were barely warm at all so heat wasn't getting transferred very well.

I think it is likely highly dependent on the clamping pressure of the cooler mount.

I'm going to go out on a limb and guess that the cooler isn't clamped down as hard on that MXM GPU in your laptop, as a water block tends to be bolted down on a desktop CPU.
 
oh god. i just had flashbacks to the early athlon days...
Tell me about it. I crushed three top end Athlon XP's with my Vapochill back then.

To my defense, the tube was too short, and mounting it to the Athlon CPU location on the board was near impossible without doing damage. That thing worked much better on intel chips.
 
That's pretty cool. I'm going to have to try this at some point.

I also wonder if a thinner foil would be more effective.

New ideas like this are great. Reminds me of the early days of the hobby before all the solutions were commoditized.

I also wonder if this would be effective on a bare chip surface, like in a fullcover GPU block, or if it is really only suitable for heatspreader use. (I wouldn't want to risk cracking the die with the force of squishing metal, even a soft metal like Indium.

Thinner would be better if the surfaces mated extremely well. One thing that would help would be to lap the CPU while it's actually mounted in a socket. I guess you would have to a buy a broken motherboard with the same socket (ideally the same board), cut out the section around the CPU, and break off any components that stood proud of the CPU, and then get to lapping. Something like this...

x99a.jpg
 
Thinner would be better if the surfaces mated extremely well. One thing that would help would be to lap the CPU while it's actually mounted in a socket. I guess you would have to a buy a broken motherboard with the same socket (ideally the same board), cut out the section around the CPU, and break off any components that stood proud of the CPU, and then get to lapping. Something like this...

View attachment 432159
Its pretty easy to lap a CPU well without doing anything drastic like cutting a mobo socket off.
 
Back
Top