Are you in Maximum Performance Mode? If yes, thats an RMA
Just got my new board in today. Immediately fired up R23 on Fedora:
The results show about a 2k increase in score from the old board to the new.
Old board, it would thermal throttle and hold at 3.8ghz. This one is holding at 4.2 when it reaches 100c (I really dont trust linux temp reporting).
Ill take it as a win.
3 Likes
I’m expecting my new board to be delivered today. Will post results once I’ve swapped it over.
@PSierra117
What size pad of PTM7950 (XXmm x XXmm) is needed to replace the Liquid Metal used for heatsink cooling.
Wanting to but some PTM7950 from moddiy.com, so I need to specify sheet/pad size to buy.
Thank you
3 Likes
i got a 40x80 Pad, but thats enough to cut about 10 Sheets the Size of the Die. Better having than needing, a 40x40 should be enough to account for application Error and redoing. I also Applied PTM to my RX7700S.
4 Likes
This can’t be an issue - if you use 60W brick, your battery would be slowly discharge and performance would be 20-25W, but temps would be 50-55C (just tested on my FW16). If temps are 100C and power less than 45-50W => it’s thermal throttling. I got very similar on my first motherboard and now after replacement I have ~95-96C with continuous 56W (Example here, on different thread).
Just do few days stress-idle loops by 5-10 mins. In my case temp settled from 100C to ~95C and continuous power razed from 48W to 56W. So looks like Liquid Metal also needs some time to settle. I bought new heatsink and I have PTM7950 (60x60) which I ordered from LTT store - $14.99, applied on few FW13 (few times each) and have more than half left. However, looks like I have good temps with LM - decided to left it as is for now.
If i am not on my 180w Charger my Device is putting out 55w for 3 seconds and than limits itself to 30w. Even when i put in my 96w charger my Screen integrates.
I’m using eGPU (external Thunderbolt 3/4 box with power delivery 100W) - and there no difference between “just eGPU PD” or “eGPU + FW 180W PD” or “internal display + FW 180W PD”. However with 60W power adapter there some degradation for sure…
But looks like you have internal discrete GPU, which is also consumes a lot…
In my scenario it usually goes to 61W for 3-5 seconds and than decreased to ~56W and in ~7-10 mins decreased to ~53-54W. Note: I have Ryzen 9 (however Cinebench shows similar or even lower score compared with yours…).
When i use my eGPU which doesnt have Powerdelivery + my USBC Charger from my Screen. I also get full power. But when using it without the eGPU i get the reduced Performance even when only the iGPU is active. So to reduce Faults its easier to use the Framework Charger when testing.
1 Like
Board swapped over & it appears all sorted!
Now getting 16143 pts on Cinebench R23 & thermals are only a a delta from hottest core to ccolest core under full load of 8.3C (100c to 91.7c) along with getting a full 65W Core package power boost & a consistant ~54W under full load on my R9 7940HS.
4 Likes
amdgpu_top has a readout of per-core wattage, temperature and clock at the bottom.
I don’t know how accurate these readings are though.
interesting tool. the core clock values i’m seeing here don’t seem to mesh with what i can poll from /proc/cpuinfo, though, so i find myself uncertain if this correctly reflects the cpu.
Interesting. I haven’t performed any deeper analysis, but it seems to me that at least the clock frequency matches up pretty well (keep in mind that the update time is not perfectly synced between the two tools here)
Also, from doing some testing today, I believe I am also affected by the temperature delta… that sucks
Just wanted to post an update basically confirming everyone else’s findings that a mobo replacement fixed my uneven CPU temps. It even came with the foam pads on it so I didn’t have to install them myself lol. An increase from 12k to 15k was definitely worth the effort of swapping the board.
6 Likes
That looks perfect 
Thanks for the update.
I decided to check the thermals on mine. Been running Cinebench for a bit (Cinebench 2024, not R23 that most seem to be running. It said I needed to sign in to download that one). Package wattage around 38-40 watts. Hottest Core is 100C. The other cores vary from 97C to 78C. Clock speeds are bouncing around 3.7-3.9 Ghz.
I can’t compare my score to the 5 figure scores others get on R23. Mine is currently estimating (still running) 812 points. It bounces between about 790 and 820 while running. It’s not final yet. That puts it right around the same spot as the 10 core M1 Max and the Ryzen 7 5800 on their chart. We’ll see what it says when it completes.
Edit: Final score was 834. The hottest core was pegged at 99.9C the entire run. The coldest core was fluctuating slightly. 78.2 to 78.5C. Nearly a 22C delta. Other cores ranged from 80C to 97C.
908 is my Multicore Score in CB2024 with my 7840hs so Yep yours is also affected, by a bad Thermal Transfer
I had posted in an other thread about my uneven temperature and low results for the R9. I reported it to support and they concluded the same and send me a replacement mainboard heatpipe combo. The replacement performs better. I had 12-13k cinebench r23 multi core scores, the new replacement mainboard scores ~15900, just below 16k.
Unfortunately the big core temperatures differences remain. Core/sensor4 reaches 100c but the others in the high 60s and low 70s. That’s still a very high delta imho.
This makes me wonder if there’s an issue with the mounting of the mainboard and the cover plate. Making it warp a bit so that there’s a weak spot in the cpu/ihs mounting pressure for the heatpipes.
I’m going to experiment a bit later this week with screwing the mainboard a bit loose or putting some padding under the cover plate to see if it makes a difference.
I am happy the replacement is giving me better results than that my initial board had but I don’t think I’m seeing those 5.2ghz it could/should? potentially reach.
Not sure yet what I will do, after the experimentation I will report back with any findings.
I do appreciate the service Framework offered, by sending a replacement hassle free. As the replacements does clearly prove and indicate my original board has a problem.
1 Like
Good luck with the new board!
Please let us know how it will go.
So far, the only good and stable result was the one from @PSierra117 with the modified heatsink.
2 Likes
I sent in my CB23 scores which were around 13k-13.5k, temperature delta between cores under load 16-19°C, average cpu package power around 32-35W. I quickly received a replacement mainboard from Framework, thanks again for the quick processing of my RMA request!
Now CB23 scores about 15.5k which is almost +15%.
CPU package power is usually between 46-52W under maximum load, I am even able to improve it a bit by elevating the back of my laptop so airflow is optimized. Temperature delta is now around 10°C.
I was also able to do a benchmark that might be more relevant for real use cases.
GCC compilation (configure --disable-multilib && make -j16) is reduced from 25m24,282s to 23m16,697s, which amounts to an increase in performance of ~9.1%.
3 Likes