[Honeywell PTM7950 Phase Change Thermal Pads/Sheets] Application, Tips, and Results

Hey y’all :man_dancing:

So I bought the Honeywell PTM7950 Phase Change Thermal Pads/Sheets (from both ebuy7 and aliexpress) and it took 2 weeks to arrive from China. I’m not sure if I can post the link, so I’ll post it as a thread reply if someone asks. Mods please delete/let me know if necessary!


EDIT

I added links, and it seems that I received different products from aliexpress vs. ebuy7. The write-up below is for the aliexpress link which is 0.2mm. I just took the ebuy7 package from my mail and opened it up. That version looks slightly more blue (turns out the material is the same color, just the plastic film has a blue tint), feels slightly thicker, and the box says 0.25mm.

I looked around the internet and it seems that many users have been using the 0.2mm version?

So I’m not sure if I received a fake product. This Honeywell product sheet says PTM7950 only comes in 0.25mm:

image

And here on page 7.

On page 3 of this PDF specifically for PTM7950 says .25mm (which the ebuy7 link refererences), and the picture shows the material with a slightly blue-ish color (EDIT: turns out the material is the same color, just the plastic film has a blue tint):
image

Which leads me to believe that the product I bought from ebuy7 is the genuine one. EDIT: The aliexpress 0.2mm version is definitely performing well thermally. Will report back on 0.25mm ebuy7 version. EDIT2: the 0.25mm version performs seems to perform pretty much the same as the 0.2mm version.

However, this link says that it does indeed come in 0.2mm thickness:
image

aliexpress 0.2mm version (notice that the color is more gray:

ebuy 0.25mm version (notice that the color is blue-ish) EDIT: the material’s the same color, it’s just the plastic film that has a blue tint:


So perhaps both are genuine PTM7950, just different thicknesses. I can’t tell for sure, so I’m going to test the 0.25mm blue-ish version and report back. EDIT: I tested out the 0.25mm version and it seems to perform pretty much the same as the 0.2mm version. I think both are genuine per their results.


Here’s my previous post with research and findings on what the material is that led me to purchasing it:

And here’s another thread of someone asking about it:

I installed it and it performs incredibly well. Installation is very easy since it can be cut to size, but a bit tricky since both sides have a plastic sheet/film which needs to be removed from the delicate thermal pad without damaging it. It’s very thin at 0.25mm (the label on my order says 0.2mm, though I think that’s incorrect) EDIT: pretty sure the aliexpress version is 0.2mm, and the ebuy7 link is 0.25mm. Thankfully, the surface area of the CPU/GPU dies are pretty small, so a 40x80mm sheet can be used for ~7-8 applications.

I installed it three times, and each time I got better at it. Here are my installation steps and some tips.

Installation and tips

Warning: do this at your own risk!

This is what it comes in:

Google Lens translation:

The thermal material (bottom) and tape/adhesive (top) for removing the plastic film that’s on both sides:

Clean/prepped bare CPU/GPU dies and heatsink (using 91% isopropyl and microfiber):

Per the instructions, placing it in the fridge (not the freezer) will harden it and make it slightly easier to work with. Measure and cut to size:

Be sure to remove the plastic film before placing on the die. There’s plastic film on both sides. To remove the plastic film, you can use the included tape/adhesive like so:

But I found that simply using some packing tape that covers the entire pad worked better:

Be sure to gently “knead” the entire thermal pad into the die so that it sticks to it more than the plastic film. And I found that peeling the film from an entire edge rather than from a corner much easier:

My third, final, and cleanest application:

Important note:
After some heat cycles, the CPU/GPU dies seem to be glued/firmly stuck to the heatsink. I thought it was actually stuck, because if I pulled the heatsink directly up it’d pull the mainboard with it. I feared damaging something. However, I learned that the trick is to gently twist the fan/heatsink assembly to first “break up” the thermal material which will “unstick” it (annotated in the pic below – note: just a reference, as the pic shows the fan/heatsink assembly still plugged in and screws still screwed in). Then it can be pulled up and off.

This is how it looked after a few heat/stress cycles after my first application:

Notice how it solidifies when cool:
ptm

Benchmark and thermal numbers

I ran this on a fresh/new Windows 10 installation since that’s easy to standardize. Power plan set to High Performance. Ambient temp/AC was kept at 73F and laptop was flat on a wooden table.

  • i7-1165G7
  • 64GB 3200MHz RAM
  • SK Hynix 2TB P31 Gold SSD

TDP

I was able to boost momentarily to a CPU package power of 60.3W like in this thread:

Cinebench R23

Single-core:

  • Result: 1,528
  • Sustained frequency: ~4.67GHz
  • Sustained CPU package power: ~18-19W
  • Max core temp: 93C

Multi-core:

  • Result: 5,703 EDIT: 5,703 was the max I got. When the system has warmed up a bit, I’m getting ~5,550.
  • Sustained frequency: ~3.35-3.5GHz; eyeballed average of 3.42GHz
  • Sustained CPU package power: ~28W
  • Eye-balled average core temp: ~78C

These results are faster than what NotebookCheck had with their unit:

Single-core: 1,507
Multi-core: 5,477

Though it does seem that if the laptop has been stressed for a long time (e.g. after several Cinebench/Prime95 runs) and the internal components are much hotter than right after booting, Cinebench multi-core results are slightly lower around 5,550 with a sustained frequency of ~3.35GHz. Or that might be when it’s plugged into and powered by my monitor. I think it’s the former rather than the latter and I’ll update here if I double-check. EDIT: I’m confirming that it’s the former.

3DMark Time Spy Demo:


s-tui stress:

On Linux (Fedora 36) with EPP set to 0 (performance) and sudo thermald-set-pref set to PERFORMANCE

First paste:

Second paste:

Third paste:

It seems like results from pasting (or rather application? pad placing?) are pretty consistent as long as the entire die is covered since it’s a pad with equal thickness throughout. Kind of hard to screw up – if you do, you’ll likely know you did.

This is definitely significantly better than the results I had with Arctic Silver 5 and Arctic MX-2, though the AS5 tube’s a couple years old, and the MX-2 like…a decade old haha. I’m not sure if my prior saved results are using the same EPP/thermald values on Linux and my previous Windows installation’s TDP values were borked since the installation carried over TDP values from a prior laptop – so I won’t post those numbers. I also don’t have up to date numbers with Framework’s stock Shin-Etsu X-23-8117 thermal paste (which may be better than this, but I removed the stock paste and don’t want to order more :man_shrugging:). So I’ve provided a bunch of up to date numbers of my laptop with PTM7950, and that can be compared with others.

All that being said, I definitely will be keeping this in my laptop. The results are impressive (and from what I’ve read has longevity). I remembered reading these liquid metal results 1 2 – I’m going to compare against these later and report back. Got a lot of images/content saved in this post and I don’t want to lose it!

Also, since I ordered two 40x80mm sheets, still have a bunch left from the first sheet, and haven’t used the second – I can mail some out (in the US) to a few people if anyone wants to try it. And I read somewhere on the internet that it doesn’t work well for a standard CPU IHS, though PTM7950 does come in a thermal paste version (which I read anecdotally was worse than the pad/sheet).

I tried to dump most of my findings here, let me know if y’all have any questions!

EDIT 2 (vs. liquid metal and extra tips)

Here are the results compared to:

I remembered reading these liquid metal results 1 2 – I’m going to compare against these later and report back.

[1]
AIDA64 Idle:
I was seeing idle temps ~39-41C

AIDA64 CPU stress:
Ran for 10 minutes. Leveled off at ~72C

AIDA64 CPU+FPU stress:
I let the system cool for 10 minutes after CPU stress, then ran CPU+FPU. Maintained ~79-81C for the rest of the run.

[2]
EPP set to 0 / performance and thermald set to PERFORMANCE.

Left terminal:

 sudo stress-ng --matrix 0 --ignite-cpu --log-brief --metrics-brief --times --tz --verify --timeout 2m

Right terminal:

watch -n 1 sensors

Around ~76C – quite impressive! This seems to perform pretty close to liquid metal.


I forgot to mention that after my first application, I did bend the clips like mentioned here (at your own risk!):

Also, thin needle nose pliers can help with removing the film:

10 Likes

I really like the look of this stuff!

2 Likes

Why, and what do you mean you applied it three times? Like for consistency or just because you had to? Also did you clean everything off in between applications?

@Michael_Wu I had been waiting for this.

Thanks for the update!

1 Like

An interesting turn of events: so I got the package from ebuy7 and it seems like that version is slightly thicker and has a blue-ish color, leading me to believe that either one of them is fake (turns out the plastic film has a blue tint and the actual material is the same color), or they’re both genuine but of different thickness. I added an edit with some info in my original post and will be testing the thicker, blue-ish 0.25mm version today.


I applied it three times since I didn’t have the cleanest application the first two times. This was my first application where part of the material stuck to the plastic film as I was trying to remove it. I peeled the “wrinkly” portion off separately and kind of just placed it on the die. I was only able to boost to 58.6W that first application, though it might have been able to reach 60W perhaps if I tried a few more times. My Cinebench scores were about the same as my third and final time, so I don’t think the “wrinkle” affected performance too much, if at all.

So basically just to ensure that I was applying the material well and gather some data, not because I had to.

I did clean the dies and heatsinkff in between applications – I try to thoroughly clean with isopropyl and a microfiber cloth every time I apply thermal paste/material.

1 Like

Does it come off better if you shut down the laptop after the CPU has been heated (due to stress testing, benchmarking, etc.)?

Or does it change phase so quickly that by the time you get the heatsink off, it’s turned back into a solid?

Added comparison results to liquid metal in the original post for the 0.2mm version. Seems to perform very similarly, I’m quite impressed.

So I ran 3DMark to heat up the CPU/GPU, promptly shutdown the computer, then disassembled and tried removing the heatsink. It didn’t seem to help as it was still “stuck”. I was able to gently wiggle the heatsink back and forth counter and clockwise to get it to unstick, though.


It turns out that the blue-ish color is actually from the plastic film:
image

The actual material’s color seems to be the same, so I think they’re just different thicknesses.

I’ve applied the 0.25mm material and am going to test performance now!

Pics of 0.25mm version:


2 Likes

So it turns out that the 0.25mm material is performing very, very similarly. Makes sense I guess provided it’s the same material with only 0.05mm difference. Here are some numbers:

Can boost up to 60W

Cinebench R23

  • Single-core: ~1,528
  • Multi-core: ~5,562, got a max of 5,624. Can probably get ~5,700 like the 0.2mm material on a good run.

3DMark Time Spy Demo:

  • Overall: 1,679
  • Graphics score: 1,506
  • CPU score: 4,825

Idle temps: pretty much the same at around 39-41C in ambient temp/AC set to 72F.

AIDA64 CPU stress 10 minutes: leveled out at ~72C
AIDA64 CPU+FPU stress 10 minutes: leveled out at ~80C

stress-ng 2 minutes: ~77C

Geekbench:

  • Single-core: 1,678
  • Multi-core: 5,422

So yeah, from the numbers, I don’t think it matters much between the 0.2mm and 0.25mm. I’m personally going to just keep the 0.25mm in my system since that was my last application.

I think some extra takeaways from all this are:

  • The results have been very consistent across applications
  • Thermal performance is excellent (anecdotally)
1 Like

One thing I noticed that seems a bit odd is the shelf life is quoted as 12 months, you can see this in your first image. Please, if you can, keep us in the loop with regards to longevity.

I assume the peak TDP of 60.3W was in Prime95 smallest FFTs? I’m wondering as the phase change is at 45*C and starting temps are below this if the phase change itself (both the iGPU and CPU) is sapping a chunk of the heat as well as the improved thermal transfer from the chip.

Are you able share a temperature vs power graph like this:

Congrats by the way according to Framework

60W is not expected or promised on 11th gen

As you can see my peak is 45W (i5 1135G7) however I have similar Cinebench results, two best recent runs were 5766 (run after system was allowed to cool down) and 5764 (from a fresh boot). With the system still a bit warm from a previous run 5500s.

My setup is factory paste with better aligned thermal pads (under contact plate) as well as bending the brackets labeled 1-3. I remember you had worse Cinebench peaks than my original board of 38W (my original peaked at 40-42W my current board is 35-37W) so I am curious to see your thermals during a 60W boost run especially as thermals seems to always be the limiting factor, note that in the graph CPU is critical temp throughout boost window.

Yeah we get the high watt boosting but what about the sustained highest clock boost?

Nice thread. Thanks for replying to mine and letting me know you had tried this. To be honest, I was kind of waiting for someone braver than myself to test it before I tried it :sweat_smile:

Something to note, below are the temps for my i5 (stock paste) when under load with s-tui for about 5 minutes. Notice how my temperatures are much higher than yours.

Generally, it seems that the performance from the CPU with the PCM is the same, but the temperatures are lower with it compared to the stock paste. Would you agree @Michael_Wu ?
谢谢你!

What are your clock speeds in the above image? It looks like 3.8GHz from the graph but that would be all core turbo clocks for the i5.

@Usernames I’ve read on the internet from a few people saying that it should last a few years and isn’t prone to pump out issue. As opposed to that being a major issue with Kryonaut on bare dies, causing thermal performance to degrade and needing to reapply it. I’m not sure how true this is though – but yeah, I’ll try to update with longevity reports!

All Prime95 testing was done with Small FFTs, not Smallest, since it says it uses max power/heat/CPU and was used originally in the other 60W thread: Small FFTs (tests L1/L2/L3 caches, maximum power/heat/CPU stress).

Here’s an MSI Afterburner hardware monitor screenshot of Small FFTs:

And here’s after pausing the graph, waiting a while, unpausing the graph, and then finally rerunning with Smallest FFTs:

Note: I don’t know what the ambient temp is as I’m running this at the airport and on/near my lap, but it does seem fairly cool. Idle temps are around what I was seeing from my previous tests – perhaps a few degrees higher.

Curious that you’re getting similar scores with the i5-1135G7, though that’s probably a good thing – my assumption is that there’s more than enough thermal headroom to keep the laptop running at 28W without thermal throttling/affecting performance. I was thinking that it may be possible to increase the power limits to say 45W and have the laptop running under 100F. Seems totally possible with PTM7950, though I’m not sure how to go about that. I think those limits are either set somewhere in Windows, Linux/thermald, or the firmware. If someone knows how to increase those limits please let me know!

In comparison to your graph, my thermal performance is about 9 degrees lower with the i7-1165G7 – though the Smallest FFTs run was done shortly after the Small FFTs run, so those temps may be a bit higher than running from cold.

@Jason_Dagless and @Jacob_Eva – the above two paragraphs may answer your questions, and I’m slightly embarrassed that I had to Google translate 谢谢你 but 谢谢你 for the 谢谢你 :sweat_smile:

Thanks for the info on better aligned thermal pads and bending brackets labeled 1-3. I didn’t really check the thermal pads, but I did ever so slightly bend those brackets as well. I’m not sure if it was enough to make a difference as I’m scared of damaging them haha.

1 Like

Hi Michael, thank you for your post!l :slight_smile:
I actually should have said the small FFTs before, thats what I was using to test boost as well.

I haven’t had experience with “pump-out” but if the TIM expands/contracts with heat it is logical it would “pump-out”. You would think general thermal stability would be a key parameter in TIM!

What I see is the chip hits 90 immediately but then takes some seconds to hit 100 allowing it to hit 60W. Interesting comparison to my graph where I’m at 100 the whole time.

Well in researching the chips there are circumstances where the i5 outperforms the i7 and in general the performance is quite similar. I have only bent my retention arms but I think that improved mounting pressure is significant. I also only see these scores when no monitoring software and from a fully cooled state or from a fresh boot.

I think it is closer to a difference of 5 Celsius, 88 at the end of my graph is a peak where temps were on average 82. That is just me being pedantic, and a 5+ degree difference under such load is a great improvement.

I forgot to ask, is your system noticeably quieter? As we can see the performance that extra boost you get brings is not massive but improved thermals should mean a quieter system.

Sure thing @Usernames, I’m glad I could help!

That steadily-rising-to-100C behavior is indeed interesting, and IIRC wasn’t happening with thermal paste…but my memory’s fuzzy FWIW.

Yeah, I’ve read similar things in regards to what you said about the i5 vs. the i7. There’s less than a 10% difference between the two in Cinebench R23:
https://www.notebookcheck.net/i5-1135G7-vs-i7-1165G7_12238_12118.247596.0.html

It seems that you’re getting very good performance from your i5!

Your idle temperatures are comparably quite a bit lower – I wonder if the i5 runs cooler and allows for it to catch up to the i7 in performance.

Ah I see. Hm, perhaps there’s an even greater improvement in comparison to an i7.

I’m pretty sure it has been considerably quieter. I haven’t been at my desk setup for a while as I’ve been traveling like the day after I installed PTM7950 so I can’t really compare apples to apples, and I’ve been running on battery with a power-saving profile most of the time. But I haven’t really heard or noticed the fan since during day to day usage. Earlier today, I did some compiling and being on a non-hardware-accelerated (Linux) Microsoft teams call which usually ramps up the fan, and I didn’t hear/notice it. Right now, my gut intuition is that the fan’s about one “notch” or one level lower than before, if that makes sense. I’ll have a better sense when I’m back to my regular desk in a few weeks, but for now I’d say “yes”.

Fairly certain this is correct. I really think the phase change itself (on both chips) is a part of the reason for this slower rise to peak.

It might be possible to test that by starting with CPU sitting above 45 so PCM is melted (better still if the iGPU also). That should eliminate the phase change from the equation, if you fancy the challenge hehe. If the phase change is significant it should be fairly easy to tell with changes in temperature profile alone.

Yeah this is the worst boost performance I have had from testing several boards but this is also the only one I bent the arms on. It could also be the boost wattage is lower → temperatures lower → sustained boost level higher. Or that boost is lasting less that 4.5% of a Cinebench and run doesn’t matter for scores much.

I have altered the scales on both axis (temp 30-100 power 0-60) I think this gave you that perception. We both had a min of 38 I noticed.

This is what I had hoped to hear. Again I think the PCM could be acting as a heat soak “smoothing out” the thermal load in the cooling system.

Yes I’m pretty sure that’s correct.

不用谢 :wink:

1 Like