My MX4 is the “2019 edition” so I guess maybe on the older side. It does sound like the paste they provided is very good though, I wonder what it is.
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.
Which CPU?
I’m using HWInfo64 and throttlestop to measure TDP, stressing with prime95 and cinebench R23.2. I would have to check but I think I see about 15W-18W single core and 40-42W all core.
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.
It’s actually stated what it is in Marketplace. Here is a thread I made regarding that if you’re curious.
Shin-Etsu X-23-8117
1 Like
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.
2 Likes
Thanks @Usernames and @Second_Coming, I did some more testing and my i7-1165G7 only boosts up to ~38W before thermal throttling which was slightly better than stock (edit: IIRC, also now I’m starting to record temps at an ambient temp of 73F), though it’s possible that the replacement mainboard I received (that I’ve been using for several months) may have been prone to the thermal paste pump-out issue. My Arctic Silver 5 tube is years old, though comparing with other pastes, I’m pretty sure thermal performance from AS5 in this die/heatsink setup is far from the best.
I initially was going to purchase a tube from Framework and repaste with that (0.5g of Shin-Etsu X-23-8117) for $5 + $9 shipping + tax. But 0.5g for ~$15 isn’t too great of a value, so I went looking for alternatives. I settled on either Thermal Grizzly Hydronaut (not Kryonaut) or GELID GC-Extreme after reading that those are thicker pastes that are supposedly better for direct die usage (e.g. on laptops) and less prone to pump-out over time. Kingpin Cooling KPx and MX-5 as other notable mentions.
However, I came across this Reddit comment on liquid metal alternatives that sent me down a rabbit-hole on the Honeywell PTM7950 Phase Change Pads. Apparently Lenovo uses PTM7958 (“optimized from PTM7950”) in their Legion 5 Pro laptop
and they have an explanation video here.
So I scoured the interwebs for reviews, and it seems like this is a pretty recent thing. What I found were a bunch of testimonials stating that it provides amazing thermal performance. Application is also simple as it’s just a 0.25mm pad. Long-lasting and apparently used by “automotive customers” as well.
Here’s some data.
TLDR: close to liquid metal performance, without the downsides.
Notable quote from this thread comparing liquid metal to the PTM7950 in a Razer Blade 14:
At the 48W TDP, the results were nearly identical (maybe a 1-2c difference, depending on the core), which is extremely impressive for a non-conductive material.
At the 74W TDP, liquid-metal had a 5-7C lead on the PTM7950 at a slightly lower RPM, but I could not audibly hear any difference. Which, again, is still a very good result for a non-conductive paste (usually this lead is closer to 15-17C when comparing liquid metal with an aftermarket TIM).
Some other notable threads.
Honeywell also makes it in a paste version, though that requires curing time in the application process, and this user reported that
“after being a guinea pig with the paste version i would recommend of getting the pad version instead.”
So, I ended up buying some of the thermal phase change pads – I checked Amazon, eBay, etc. Unfortunately, they seem to only be available through ebuy7 and AliExpress (shipped from China). So I bought from both sites. They hopefully will arrive in ~2 weeks. When they arrive, I’m going to slap these apparently magical pads in and see how they perform in my Framework laptop.
I’m excited and have high expectations for these! If they’re mediocre, I’ll reply here. If they’re incredible, I’ll start a new thread and link to it with instructions. Stay tuned!
3 Likes
I used Kryonaut on my Framework and it was little different to the stock paste. For anything else than special personal usage I just use the newer MX-5. It just works and you can buy it in larger tubes.
Following up on this!
The person responded to your ticket didn’t take into consideration of the wording of:
“60w boost” on the 11th gen laptop product page
It wasn’t:
“Up to 60w boost”
The wording of “up to” was in relation to the heatsink & fan’s cooling capacity.
Therefore, the laptop, product as a whole, for the 11th gen was marketed to be perceived that 60w is achievable with every unit as long as operating environment (ambient temperature, humidity, atmospheric pressure) are within spec. It was NOT conditional to silicon lottery.
The keypoint 3 states:
“Dependent on workload and ambient temperature conditions”
Framework | Framework Laptop
Quite agree. The kicker is when queried support state and confirmed expectation is 45-55W not the stated 60W.
Wording is also the same on 12th gen page so I suspect that it will be similar story there too.
I’m not sure even this statement of theirs is true, at least form my testing the cooling system seems to be at its limits at 30-32W.
I ended up “accepting the silicon lottery” explanation because 1 support and I were getting nowhere and 2 the higher quality silicon being more efficient would be much more likely to hit the stated peaks.
Then I would ask…“Given the same wording, does that also mean 12th’s 60w is 'achievable only base on silicon lottery, and is not a promise of achievable performance even when ambient temperature and workload matches that of Framework’s test from nrp?”
i.e. Given the same wording, you can’t have it both ways.
Similar to this case: The law can’t have it both ways: Texas woman given traffic ticket says unborn child counts as second passenger | Texas | The Guardian
Overall, this “60w boost” and “up to 60w boost” needs some improvement on clarify. Otherwise, that’s just intentional ambiguous / deceptive marketing.
Framework 12th Gen buyers beware.
Yes.
I’m just so tired of contacting support over that last 5 months…. I just want a working laptop and this seemed like another hardware problem that could be fixed.
Nothing they say will change the performance I get, at best 47% of advertised. They don’t seem to think it will help sending replacement TIM either. That tells me they either know it wouldn’t help or they only do it for 1185G7s customers.
I have alerted them to the issue with their wording, it is up to them how to respond.
I wonder how the batteries “up to 80% capacity after 1000 cycles” will actually work out…
I’m starting to see a pattern:
- Next week - Means Friday.
- July - Means last week of July.
- Up to 60w - Silicon lottery.
- …so, up to 80% capacity after 1000 cycles - Best case scenario.
It’s about MVP…minimum viable product.
Like…I don’t know man…if a display panel says max 400 nits…they usually can go around 410-420nits in the centre…all of those that reaches a customer. So, with Framework, this MVP thing isn’t exactly what I’d call striving for excellence.
Really, you can’t sell things (well, Framework can) base on ‘best’ case scenario and place that as product specs.
e.g. When Intel, for example, says X processor has a base clock of yGHz, and boost clock of zGHz…that applies to EVERY unit of that bin / SKU. Not the best case of that bin.
When an SSD is spec-ed as up to 1000MB/s sequential write…that ‘up to’ applies to all units if placed under the same ideal condition, not just the best SSD of the batch.
Framework really need to look at that 3rd 4th screenshot response you provided (above).
The concept of silicon lottery typically applies to unspecified behaviour or characteristic beyond documented specification. So, if “up to 60w” is documented, that’s considered in spec…for every unit sold, and should not be further subjected to silicon lottery. If I want more than 60w (undocumented), then sure, park that to silicon lottery.
1 Like
I think if the language used makes it clear “this is expected in ideal conditions only” it is O.K. IMO but it’s not good when wording is misleading, intentional or otherwise.
When the deviations from stated performance in ideal conditions strays more than 10-15% from marketing materials you are just lying.
Reminded me I missed the first response before they did the 180*. Will add it now so it will be 4th SS shortly.