Recently became aware of this thread and there are far too many comments for me to read, but it seems like I may have the worst performing Framework 16 in existence! Has anyone been able to beat my high (low) score of 11,228? Best I could get was 11,959.
For the above scores, note all the following conditions:
Framework 16 with 7940HS
Have 7700S (a very loud one at that possibly due to CPU), but all test done with it removed.
x86 Utility was installed and set to Extreme profile for some tests, uninstalled and ran more tests with only Best Performance set in Windows.
Tests with 10 minute multi-core runs in Cinebench R23
Hottest core in HWiNFO 64 was 100C max, with most max in the 80’s, lowest core 72C, so pretty significant variances
CPU package power limited at around 31W in HWiNFO 64
Currently in contact with support about hopefully getting a replacement board or TIM kit.
Attached were my worst thermals and performance values.
Based off of this and the screenshots it looks like a perfect example of the issue people have been having. If it bothers you it seems like support is doing a good job taking care of the issue after getting you to the right place, and if you scroll up you’ll see people have had some luck with PTM7950/other mods and/or replacing the heatsink if you’re comfortable with that route or aren’t covered by warranty for some reason.
Yep yours is a Prime Example of Run off Liquid Metal. This bad Score is not by the Shim Bomding. I wouldn’t attempt to do anything myself on this. I would request RMA and let Framework do the Investigation on that. The more we RMA the more Framework is forced to do something.
Yeah I saw the significant gains people achieved thanks, and I have some PTM7950 already so have been considering fixing this myself, but am worried about the liquid metal spilling.
Support has been good so far and thorough. They opened immediately asking about Windows power settings, my power adapter, my battery SoC and then said to run R23 and HWiNFO 64, so for anyone else contacting support in future, coming pre-loaded with that information could help expedite things.
They’re specifically asking me now for screenshots of thermals while the test is running, so just sent some of those too. We’ll see how it goes…
You have to admire the ambition though… with the work that @PSierra117 did, up the thread slightly, this thing can best-case never thermal throttle before it throttles based on power policy. I’d never planned to see that in an x86 laptop again (though… I bet it gets more common as AMD and Intel are trying to compete with ARM).
I’m talking with support as well. So far, they asked me for
screenshots and logs from turbostat, psensor, and s-tui with stress running. (giving them, frequency, wattage, thermals)
screenshots of completed Cinebench R23 runs
BIOS reset
BIOS upgrade to 3.03
runs on AC and battery, and on different power profiles
to use a live USB so I could try out a supported Linux distro. (that was where they asked for s-tui and psensor, because the ramdisk didn’t have enough room for me to install Cinebench in Bottles like I had been.)
It’s been a marathon, but, understandably. They can’t take the general public at its word on these things, gotta try to whittle down variables.
I went through it with support as well. I did get pretty frustrated at a couple points. But try to be patient. Ultimately they sent me a mainboard which was unfortunately defective. They are now at a point of offering me a list of choices, depending on how I want to proceed. So while there have been some frustrations along the way, ultimately they have been very accommodating.
As of last night after I finished lapping the heatsink to remove leftover solder. this is what I got for score. Unfortunately the heatsink has a slight deformation where a corner of the heat spreader solder was so I think for the time being this is where I’ll keep it. With a cold boot I can get closer to 16.5k. I also haven’t tried any undervolting yet, so later today I’ll take a crack at it and see if I can get it to a stable undervolt.
These are the results from running the benchmark post lapping, but pre-undervolt:
Good Job. Lapping is something i wouldn’t do anymore. It was a Step in my Testmetric. Removing the Stock Shim and replaceing it with a Clean Coppershim 20x20x0,8 and putting it on a cleaned Vaporchamber got the Best Result by far.
The Shim is easily removable with a Razorblade. Afterwards the Vaporchamber can be cleaned and the Solder removed with Wet Sandpaper on an even Surface.
I had a lot of the solder left over on mine when I took off the stock heat spreader. I figured it would be easier to get it off that way and help get a more even heat spread.
I’ll post the new temp/power photo in a bit once I’m back home to test some more.
I used a blow dryer on the heatsink side to get the old hard liquid metal loosened up and a lot of q tips to lift it off bit by bit before cleaning with qd electrical cleaner on a microfiber towel while still warmed up.
CPU side I used a pair of tweezers to pull it off in pieces then heated it up the same way and removed with many q tips and then wiped with qd electrical cleaner… I don’t think I would do that side the same way again though.
I did end up removing the original plastic protector as well and replacing it with two layers of PTM7950 around the CPU dye itself just to add a barrier for the LM (I feel like a correctly sized thermal pad would be better here but I had no thermal pads in hand), then added a layer of vinyl cut to match the original plastic protector. I’m planning on pulling the whole thing apart in a couple of weeks and seeing how well that solution holds up short term.
Hi,
I see that hwinfo is showing per core cpu temps.
If I boot windows from a usb, I cannot see per cpu core metrics.
I also cannot see them in Linux sensors.
I have a 7840, not the 7940.
Does anyone know if anyone has a patch so I can view per cpu core metrics in linux?
There is also the whole documentation aspect that adds up to it. They need to collect data from the defective devices to probably compare it against existing internal data and try to find a pattern or a clue for what’s going wrong. It makes sense a lot of information is required because on top of not shipping hardware unnecessarily, I think they also don’t want to pollute their data sets with bad data
Somehow power limits should cause thermal throttling? Somebody needs to revise the support scripts or something or you rep just went down the wrong path.
Thing is, if the names on the emails are anything to believed, I’ve had about 5 different support people respond to me so far. But no excuse for asking the same things over and over.
