I didn’t run any benchmarks before I made the modification but I ran Cinebench R23 multicore and got 15.3k, and HWiNFO shows it maxed at 56.4 W and 92 C. I think generally it was around 50 W and 90 C. I’m not sure if that’s how I should be reading this stuff, please let me know if there’s a better way to report. I also did run it after rebooting and closing most other stuff if that matters.
2 Likes
That’s right in the ballpark of those with the modded heatsinks
2 Likes
Perfect. So the ai 300 Heatsink is a perfect replacement for the old Systems. And it looks right about within the modified heatsinks maybe slightly better. Maybe i get one and have a look if its possible to modify and improve too 
4 Likes
I also put the AI 300 heatsink into my 7940HS.
I couldn’t find my FW 180W charger so I just used a strange 140W charger.
I got 12900 after the first reboot with average CPU usage of 40W.
The only strange value ist \_TZ.TZ03 which always shows 100°C. Can somebody tell me which temperature sensor this is?
Here are some reasons why my old heatsink wasn’t working so well anymore.
Probably the new heatsink wasn’t necessary.
2 Likes
MAte you are still getting a 100C Package Temp and that at 40W? You definitly did something wrong while installing the new Heatsink! What Thermal Interface Material did you use?
It was the included PTM.
I’m concerned about the thermals of my DIMMs.
This is not a bench, but just heavy memory intensive workload running for tens of minutes to hours.
Generally over >70C for RAM is concerning. Anybody can share their stats in the optimized system?
The sensors in questions are SPD5118.
Interestingly package power is continuously at 50W+, all cores are at 4.4GHz - 4.5GHz continuously. I.e. it’s more intensive than s-tui stress test that only gets CPU to 45W and 82-84C.
What type of RAM are you using? 75°C shouldn’t be too problematic for DDR5 RAM, but there may be a different recommended ‘safe operation’ range depending on the specific type of RAM you’ve got installed. I’ve had a hard time finding published datasheets for a lot of types of RAM, but from what I HAVE found and read it seems like safe temps should max out somewhere between 85° and 95°.
This doesn’t really answer your question so ignore me if I’m misdirecting.
It’s another CPU temp sensor apparently.
As for why, I would guess you need to go around the heatsink screws a few more times. Also what are the fans doing in terms of duty cycle? Use framework_tool to force them to 100% and try again.
On the working assumption that you fully cleaned out the hair from the fan blades and heat exchange fins, the heat is either not leaving the CPU fast enough or not leaving the heat vapor chambers fast enough. I found I had to go around the heat sink more times than I would have assumed to get a constant clamp across all of the screws, and that would explain why you have ok-but-not-great heat transfer.
Arcadiy_Ivanov1
Conor
I also could not find any documentation officially saying those temps are a problem, but they certainly are for some DIMM configurations/manufacturers. I can personally confirm that I had problems when they were getting that hot until I used thermal transfer pads as well.
1 Like
I have 128GB of Crucial DDR5-5600.
1 Like
I found a Micron [the manufacturer of Crucial RAM] DDR5 SODIMM datasheet stating that the safe operating temperature range is between 0° and 95°C, but I would trust @Nathan_Figueroa’s experience over what I’m reading.
1 Like
I’m seeing similar results after attempting the copper shim mod… I don’t have the money to tinker around with this mod indefinitely so I’m thinking about grabbing the Ryzen AI cooler since I saw somewhere that it fixes the issue that the older coolers had, those still work as intended on the 7040 series mainboards right?
Going ryzen ai cooler can help. We only have one who tested it already. Sad to hear you had bad results after the copper shim mod. Did you get a 0.8mm shim?
Thanks for posting this!
I just ordered the new heatsink to try it out.
Yeah, I grabbed the same set that TechPriestNhyk had linked in his “Adventures in mitigating thermal throttling” thread. Although I got thermal grizzly PTM instead of the Honeywell stuff after reading in here that it shouldn’t have been that different performance wise
My best guess is that I wasn’t properly lapping the heatsink or something
Edit with additional info:
I was about to start my teardown to install the new cooler while writing this, using another Cinebench run to soften up the PTM before pulling off the hestsink, and just scored a 14958… I’d been consistently hitting the low 14000’s to high 13000’s after doing the mod with each run scoring lower and lower…. I wasn’t running HwINFO that time so I can’t look at the temp data, but I’m doing another run right now with it running to see what’s happening. But I do think I’m still going to install the new cooler anyway since I already have it here and opened, I’ll share the results afterwards.
When I was ordering my PTM I paid extra for the 7950 from LTT instead of the Thermal Grizzly because when I looked up the thermal conductivity specs of each, they were drastically different. I don’t recall the details; that’s just the impression I was left with. I’m curious if that really is the difference in the performance you’re seeing.
Either way I just wrapped up the Ryzen AI cooler installation, ran Cinebench and off the bat got a 14472 after it crashed one time…
Anyway these are from when I was trying out the Copper Shim mod:
6/28/2026 First run after installing PTM Copper Sandwich
6/28/2026: Second Run, lower score, might need to check something internal
Yeah I just realized that I’m horrible at data collection, I know I ran a few more tests between 6/28/ and 7/2, but I didn’t record them.
7/2/2026 First Run:
7/2/2026 Second run:
This was the run that made me decide to snag the AI cooler since I need this to be working by next month for classes and I didn’t have time to tinker.
1 Like
I’m almost positive the Thermal Grizzly PhaseSheet is the same as the PTM 7950, just re-branded. LTT just charges more because, well, they are LTT and people buy it regardless.
Someone compared the two a year or so back (it’s up on reddit) and their results showed the Thermal Grizzly as 1 degree lower than the PTM in stress tests, well within margin of error.
I bought both when I redid my 16, and just ended up using the Thermal Grizzly. Both materials appear identical.
2 Likes
Well then, I’m happy to be wrong. The thermal grizzly is much more convenient to buy.