Uneven CPU thermals!

I didn’t have any major issues with performance (15,015 in cinebench R23) on my batch 5 7840HS. However, I could not match the sustained 10min performance of @PSierra117’s modified cooling solution.

I lapped my heatsink and bought some unverified PTM7950 on Amazon like @PSierra117. With the new cooling solution I have gotten above 16k in single cinebench runs and the fans take noticeably longer to ramp up. I am attaching some images of the original liquid metal and lapped heatsink.




9 Likes

Thank you for trusting my Worklog and verifying my findings. Its a big leap of faith, doing something like this to an expensive Machine just on the Words of a Stranger in the Internet. As you said its alot more pleasant device after the Thermals are in Order :wink:
What did you use for lapping the surface?

2 Likes

I used a 600 and 1200 grit diamond stone after covering the surrounding area with tape.

4 Likes

How is the performance so far?
Is there any degradation compared to the liquid metal?

I’ll receive a new mainboard this week, but, I also have an extra heatsink that I plan to modify just in case.

2 Likes

I switched a little over 4 weeks ago and have not experienced any degradation.

3 Likes

For those who haven’t seen, I posted my experience over at FW16 Batch 20 Guild - #361 by lxke

Today, I ran another test, getting about 14,400 which is probably within margin of error of my previous score, considering that I had been using my laptop all day for work and only had powered it down to drive home, then run the test, so it likely wasn’t a ‘cold start’ as previous runs had been.

I had an idea today to go about replacing the liquid metal TIM with regular thermal paste and I’m happy to say that this has improved the situation somewhat.

Before, I would be hitting roughly 14.5K in Cinebench R23 at about 40w over a 10 minute run, and now I’m getting 15,359 which initially started at about 51w, dropping to 45w by the end of the 10 minute run.

Thermal-wise, I was previously seeing an almost 20c delta between the hottest and lowest core. Following the re-paste, the now have a delta of about half that, at 9c. When the CPU was hitting 52-ish watts package power, my hottest core was now core 5 instead of core 4 previously.

I’ll be update this thread to see if the performance degrades at all

5 Likes

Did you modify the heatsink?
Also, can you please add the name of the thermal paste? It could help us keep track of what is working.

1 Like

No heatsink modifications (yet)

I used some old Gelid GC-Extreme from 2018 I had lying around, and have ordered some genuine PTM 7950 from moddiy which I’ll test out once it arrives.

3 Likes

Just curious - how hard was the LM removal?

1 Like

Removing the liquid metal with qtips and rubbing alcohol was pretty easy for me. Removing the heatsink was the most difficult part for me - I had to use a plastic spudger to separate the heatsink from the CPU.

2 Likes

I used GC Extreme some Years ago. Great Stuff, but degrades it Performance over the Course of a Year.

2 Likes

You can rub it off with a paper towel but it was proving more effort than I wanted it to be.

Ultimately I used the plastic end of the included screwdriver to scrape it off with ease, it almost came off as a single piece, then cleaned it up with a tiny bit of rubbing alcohol.

I didn’t take any pictures of the LM TIM but it looked quite sparse, like, a wafer thin amount that in all fairness did cover the whole CPU die.

This morning I ran a cold run of R23 before heading to the office, managing a score of 15,640 - My highest yet. I’ve attached a screenshot which I took a few moments after the fact, so you can see hwinfo stats. CPU Package Power started at about 54w on the first run, dropping and stabilising to 48w by the end of the 10 minute sprint.

2 Likes

I received my new mainboard and, because after a day I could already see a decrease in performance, I decided to join the club and make the same changes to the heatsink.

Heatsink

Heatsink removed to check the pad


3 minutes run

10 minutes run after the 3 minutes run

Now, the downside is the fact that I can hear the electrical noise from the GPU as the fans won’t run as hard as before.

@PSierra117 thanks again for trying this.

4 Likes

Great Work, what did you use for Lapping and what Thermal Interface Material? I will buy another Heatsink in October and after mine arrived i will fully lap the Plate so the Vapor Chamber touches the Die directly.

Thanks.
I used Knife Sharpening Whetstone.

  • started with 400
  • continued with 1000 and 3000/8000
  • finished with “wet sanding” (99% alcohol instead of water), 3000

I can’t wait for your new test :smiley:

3 Likes

Sharing my experience here. I brought a batch 1 f16, has been running fine, but I always found the temperature to be a little bit too high. I could not watch youtube with hardware accel turned on without the fans kicking in at high blast.

‘Edge’ sensor would report 60c on idle, and instantly jump to 90-100c on any bit of usage, which would kick the fans to max.

Decided to finally replace the liquid metal, this is what it looked like:

Images


I’ve replaced the liquid metal with thermal paste (arctic mx-6 since it’s all i had on hand). The difference is night and day. I do wish I did a little bit more testing before I replaced the liquid metal, but I’m very happy with the results.

Before temp on idle 5m

After on idle 5m

It still looks like the core-to-core delta is pretty large (84.9 on Core1 vs 100 on Core6), so it would seem that there’s still something funny going on under the hood.

Yup. I’m not exactly sure why.
It could be anything right now. Fake PTM7950, bad modification of the heatsink or, bad application.
What is strange is that it looked like it was applicated evenly, once I removed the heatsink.

During normal use, the temp is more even.

1 Like

I suspect, the attachment of the riser Plate to the Vapor Chamber. You can have a great plane from the die to the plate. But its still possible that the Transfer into the Vaporchamber is still borked.

I wonder if that attachment is just glued to the plate.

While playing a game, I can see that the power used for the CPU hit 82.58W while the max CPU temp was 83.5, so I can’t complain about this.

Power usage