A happy update from me, and Framework Support came through the 2nd time around. (For some reason, I’m not able to post more than 3 replies to the other existing thread)
Framework Support saw my forum posts from before, and proactively reached out to me, case got escalated, and with them sending me new thermal paste. That got applied, and I’m now a happy camper again. Don’t know why the original thermal paste in the DIY unit wasn’t able to hit this OOTB.
Very cool. I’ve never ever heard of a company proactively reaching out to their customers and helping them resolve problems, all while eating the cost. If you asked Dell or Lenovo for some paste they’d laugh in your face.
Reading your post I decided to try and repaste with my MX-4 and it did help with overall performance (cinebench R23.2 scores are up 5%) however max power is slightly lower, core 3 will always thermal throttle the CPU, core 1 is usually close behind while core 0 and 2 can be over 10*C lower when throttling occurs.
Prior to the repaste the thermal deltas were smaller but overall performance was lower also.
Given that cores 0 and 2 are similar temps and 1 and 3 are similar temperature I believe the physical core layout would suggest this is a mounting pressure or flatness issue. I tried repasteing and remounting several times, same results.
Both right off the assembly line from the factory, as well as the tube they send you. MX-4 is a downgrade in my testing, on my unit, over multiple applications of MX-4, and from the tube sent by Framework.
(Able to go as high as 63.1w in peak load…momentarily with the tube from Framework)
Following and jumping in as I repasted with some old MX-2 and Arctic Silver 5 tubes last weekend to see if I can optimize thermal performance.
How are y’all measuring peak TDP? And what about results over a prolonged period of time?
@Usernames I’m getting similar discrepancies in core temperatures (e.g. two cores are ~5C hotter than the other two). And currently mine can only sustain 20W package TDP in a single-thread workload (like Cinebench R23 single-core) I think due to thermal throttling. Not sure if normal – I managed to max at ~1450 in single-core Cinebench R23, whereas 1500+ is supposedly normal from reviews etc.
Multi-core workloads seem okay, though (edit: I can sustain 28W package TDP). I think it’s because the heat load is spread across the chip/heatsink and there’s enough thermal headroom, whereas there isn’t for a single core. Still conjecture at this point, and day to day performance is very good – just wondering if there’s room for improvement, and how to get there.
Oh right this thread’s for the i7-1185G7, I have an i7-1165G7. Which CPU do you have?
I was comparing to Notebookcheck’s numbers. Caveat, the review’s from 7/26/2021 so things may have changed.
Thanks for the numbers – they’re very similar to what I was seeing with mine when I tested in Windows 10 with Cinebench R23 and ThrottleStop. I can confirm that I was also only able to sustain 15W-18W single core, though it seemed to improve after I repasted with the spread method instead of the pea/line method to ensure the entire die was covered. Unfortunately, I didn’t care enough to write down hard numbers.
I’ll jump into Windows later tonight to check peak TDP in single/multi-core (some curing time has passed with the Arctic Silver 5) – let me know if you’d like anything for comparison.
My curiosity’s piqued – what did you lap, the heatsink or the die? Bending the retention pins seems definitely risky, but definitely something I’m interested in
I’m thinking about ordering Framework’s tube or Kryonaut, ehh.
I have the i5 1135G7. I think the reviews are using the same CPU as yours so should be comparable.
I did some more testing 15W single core in cincebench, seems logical also considering 4 core 60W boost. I also see slightly better deltas between cores of 5-8 rather than 8-10 but otherwise unchanged.
@Second_Coming Good mods, I’m not going to do any hardware modifications till after my warranty expires to avoid potential issues but I think a little bending of the clamps would be an easy win but exposed die so some caution needed
You said you got lower performance with Kyronaut than the FW provided one, that is a high end paste IMO I’m not sure why high end thermal paste is needed to get to 60W boost, seems a little odd.
During my testing, it seems that some paste / composition is better / worse at a given pressure and / or smoothness of the contact.
Conventional wisdom seems to suggest the smoother, the better. But I’m not so sure anymore. Maybe some roughness is better(? big maybe). roughness means more surface area…for the paste to take the heat, and to pass on the heat to the heat sink.
If the other medium is a fluid (air or liquid)…then we seem to want more surface area. But with a solid, people seem to want flat…really flat / smooth. I’m not entirely sure when the two switches position. The viscosity come into play, I assume, as we’re going from a solid (die) to fluid (paste?) to solid (contact plate).
Well one thing you have reminded me is that some paste perform better under lower mounting pressures and some higher, I think viscosity and small particle size might make the difference there.
The idea is fill the gaps where the two solids don’t meet perfectly with paste right so perfectly mated parts should be the best but I have also seen Kingpin suggest when lapping to take the mirror finish off for the paste to have more to integrate into. Seems strange and could be a LN2 OC specific thing to avoid cracking, I would expect the less paste the better because they conducts less heat than the direct metal contact of the heat sink pressure being high enough.