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.
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.
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%.
It might be a fluke but I unscrewed the 6 mainboard screws just enough to remove the holdown tension as a test.
Screwed in the coverplate and all the covers to complete the laptop.
Ran CB r23 again.
I got the highest score yet I seen my FW16 do.
Im going to do another run and see what happens.
Do note the thermals of the cores. Still very unbalanced.
It’s too soon for any conclusions, I will do more testing.
Just got my new mainboard as well. I’ll probably install it sometime this weekend when I have the time. Hopefully I get as big of an improvement others are reporting with their new boards.
I’m a Batch1 FW16 7840 owner and just found out my laptop also has problems. I checked because of this thread and i wondered because I saw the full power being used right when i got my fw16 in march.
So, here is all information i found from my testing in march.
Performance statistics are from a Manjaro linux - these screenshots are from march - so first testing out of the box.
Now when i test with Cinebench R23.200 i only get around 30 watts, also in the following image you can see the core clocks differ massivly to the core clocks in the old image. Now i get around 3500 MHz, in march i got 4200-4300 MHz on an allcore load.
Now… Because of this thread i removed my keyboard and deckplate and just slightly pressed on top of the cpu cooler. OOPS, cpu instantly jumped to 45 watts and around 3900-4050 MHz. still not the old numbers, but defintly better.
Cinebench R23 numbers are around 13500 to 14200 when thermally limited but when pressing on the cpu cooler i get around 15200 to 15600.
So it seems i have contact problems. sad I thougth i would not have to do maintenance because of the liquidmetal.
I think your CPU was somewhat screwed from the start, I have stats from my two month old laptop stressed using 7zip’s benchmark functionality and it looks similar to your system when it was healthy while using far less power on the balanced setting.
Thanks for your share. I suspected that all FW16 share this fate. Thats why i started modifying my Heatsink. You could test to loosen your 6 Mainboard Screws and just slightly torque them again. This can possibly help. Thats one Idea a User shared higher up in this Tbread Here yesterday.
The System should stay at about 50-53w on a Multicore CB23 Run. If its not its hard Throttling. And the Number 1 Reason on this System is the Heattransfer to the Heatsink.
finally got around to getting cinebench r23 going, ended up with 11.9k to 12.5k for multi-core with the power profile set to performance, with turbostat results typically looking like…
of course, being on ubuntu 24.04 lts, there’s some awkward with my current kernel (i don’t have properly functioning temp sensors, for instance), but this does look fairly below what some other folks have achieved, so maybe i’ll fiddle with screws later.
I did a little bit more experimenting.
When I screwed down the screws again, I didn’t hit the 16 anymore.
This led me to believe something is going on with the pressure and or angles of the heatpipe and mainboard.
I tested with a pad I had laying around ~25mm*30mm that I got from an nvme m.2 one of those grey rubberish slabs.
I put it on top of the heatpipe and the top plate.
still sub 16k nothing else noticeable.
Then I tried putting the slab between the bottom and the mainboard.
I didn’t hit the 16k but I did see another anomaly, the thermals for sensor 1-3 had a 3-5C lower delta. sensor 4 still hit 100c.
The slab was too small to actually have good contact or all round pressure.
I am going to order some more of these slabs and going to try different shapes and locations to see if I can find the sweet spot for improvement of the delta and all round temps.
it is a different route @PSierra117 took lapping the heatpipe and PTM7950.
My hunch is that the cpu is not getting the optimal mounting pressure for the heatpipe and ihs to get all the cores to have good enough indirect contact with the heatpipe.
Thank you for your Tests and further Tinkering. Thats the Value of a Community willing to experiment
We should inform the Framework Support of this Findings. Maybe @Destroya can Help?
I will buy a second Heatsink and 2 PTM LM Pad from Framework in mid to late October. They wrote to me in my Support Request, that the Warehouse is stocked with the LM Pad, but they will realease it for Sale when the Replacement Guide goes online and this is presumably in Early October.
When i have a second Heatsink i will try to modify my Current one further and than Test it against a "Stock Setup and i will experiment with different Mounting Pressures and maybe Shims underneath some of the Mainboard Fasteners.
I have ordered some of those thermal pads, hopefully they will get delivered tomorrow so I can proceed testing.
Depending on the results, I might end up doing what you did @PSierra117 I don’t mind tinkering, I have been lapping CPU’s since the Celeron 300 days.
