Yes. As stated here, I was using RSR 1920x1200 + AFMF2 @ Quality/High.
eeeeeeeeeee! grate!
eheh. i got my replacement keyboard today. not editing any of my afflicted posts, though, because they seemed to entertain people. alas, that era is at an end
7 Likes
I was talking to @jared_kidd who was shocked about your amazing frame rates 
I finally took care of my heat dissipation.
To remove it I disassembled the motherboard, and I used a simple hair dryer to liquefy the original paste. After a few minutes, it came off very simply.
I wanted to add some solder to the shim, but it was impossible with what I had to melt the solder at 140°C. Oh well, I cleaned everything, and put a piece of “Thermal Grizzly Phasesheet” and reassembled everything.
My config: R9-7940 without dGPU, Fedora 41
The fan curve is modified:
sudo ./ectool thermalset 3 381 381 400 315 365
sudo ./ectool thermalset 2 363 363 378 305 325
With this I already had “good” performance.
Benchmark, I use a LLM-BF16 bench (with llamafile) it uses the avx512-BF16 rather intensively. (execution over several minutes, stabilized)
./llamafile-0.9.0/bin/llamafile-bench -m Mistral-Nemo-Instruct-2407.BF16.gguf -p 8192
Result before:
- 100°C max, delta ~20° between cores, stable at 52W
Result after:
- 90°C max, delta <10° between cores, stable at 54W (limit in performance mode)
6 Likes
Well, I didn’t do a replacement shim and all that yet, but I need the computer for a work project next week, so this morning I quickly pulled the cooler and swapped the liquid metal to PTM7950. There was a pool of liquid metal in the one corner, along the foam barrier, so it’s sure good that foam was there. One side of the CPU still had good coverage, but the other side was pretty close to bare.
I didn’t have time for a full bout of testing. I didn’t even complete a full Cinebench run. But the core temps were MUCH more balanced, and it was maintaining around 43-44 watts, instead of 38. Definitely an improvement. I’m happy with the results for now. I still have the other heatsink to experiment with later. But for now, it’s much better.
6 Likes
I remember reading that you need to break in the PTM7950. Make sure to heat it up using your stress test of choice, let it cool down, and then heat it up again until you’ve done the cycle 10 times.
Finally fitted my ptm. Didn’t do the shim sandwich yet, honestly I want a spare hs before I try.
Originally:
12210 in CB R3.
Core 4 locked at 100c, lowest was core 1 at 73c, and peak power 34W.
Now 15523 and 54W sustained, peak 63W. Cores were 92-99c and Tdie 100.1c max.
That’s the first heating cycle too.
The Break in Period is for like the first 2-3 Days of normal Use and increases the Performance by additional 1-2%.
The PTM has to get liquid with enough Mounting Pressure than it presses additional Material out to get a thinner Bondline, than it has to Cool to below 45C. After about 10-20 Full Cycles the Material gets really hard and doesn’t liquifiy anymore. Then the Breakin has reached its equilibrium.
1 Like
Welp, turns out my unit is deep in the affected territory now. My cinebench score was just a hair above 12k and it’s now capping at about about 30 to 31W power draw. Hottest core at 100c, coldest at ~80c.
That is really good. I did the same thing but only get a sustained at 43w. I bought the shim and will try sandwich later.
Mine is about the same. Support asked me to wait until Framework ships out the “thermal pad kits” but I’m less than a month till my warranty runs out so I’m thinking I need to push them to make this right before that. It’s pretty frustrating they are dragging their feet on this issue.
Honestly, it needs to be a recall.
I agree with the recall, terrible performance to not get what you paid for.
I have tried to fix my thermals myself twice one using PTM7950 and another time just using thermal paste and even replaced the heatsink and had no luck.
I just hope whatever kit they have to fix the problem comes with this copper shim because it sounds like it fixes the problem so much better.
Fortunately I don’t necessarily need the performance, I mostly use mine for text editing and have a desktop for gaming. So while waiting for the kit to come out isn’t a big deal for me (provided the process to use it is feasible), it’s annoying that I feel like I have to use my desktop for gaming unless I want to watch my cpu sit at 100c the whole time.
I’m also moving towards a more minimalistic lifestyle though, and if I were to get rid of my desktop’s, this performance issue is a hiccup in that journey.
That’s partially why I have been lazy about pushing support to fix mine and during the holidays I just ignored it because I didn’t want to sit there running the same benchmarks over and over again for them, choosing to spend time with family instead. Anyway, I thought about it and came to the conclusion they need to make it right. I paid an extra $200 for a slightly faster processor and right now it’s much slower than the slowest lower end processor, maxing out to 101 degrees almost immediately and only using a max of about 32w.
If not a recall, they could at least extend the warranty for the mainboard out another year. I feel sorry for all those who haven’t realized yet and by the time they do it will be too bad, so sad.
1 Like
I have a batch 1 device, I wonder how much warranty I have left. It would be nice to see a statement from them that says warranty claims regarding this issue will be extended to at least encompass the time to arrival for the kit.
1 Like
I have a batch 1 as well. My order shipped March 6th, 2024 so I’m going by that date.
1 Like
I’m batch 5. My results are bad.
12662 in CB23 for multi core and 1657 for single core.
Realized I’ve referenced this thread a lot in research but never reported in - Batch 1, R23 Score of ~12500 and max wattage pulled of ~29.5.
I’ve more or less resolved to wait until the FW 2nd Gen event and see if they reference anything with the new thermal interface (even in passing, with an updated thermal solution on a new board or something) before I go buy a sheet of PTM7950 and resolve to do it myself. If that comes to pass, I’m planning to take detailed notes as I go to help other DIYers that may be tired of waiting as well.
6 Likes
I did more testing and compared the framework 16 with the same processor against a Razer Blade 14 (2023) with the exact same processor and iGPU.
I can barely hit 30-40W CPU package power without throttling like hell. It seems if i cover ALL of the dGPU vents i can barely just hit 45W.
Over on the razer blade im hitting 54W sustained at cool temps.
The cooling on this device just sucks, im going to see if i can try the chopper shim.
Whats the point of selling a higher sku CPU if you cant cool it and it throttles and is worse then years old laptops.
Doing a unreal build workload for my game takes an hour longer, on the FW 16 compared to the razer blade 14.