Thanks for testing. Could you give us some benchmark results if that’s not a big hassle for you?
The dell rugged extreme laptops have a covert / stealth mode:
https://www.dell.com/support/manuals/en-ca/latitude-14-7424-laptop/lati_7424_sm/stealth-mode?guid=guid-3655713b-6a1b-46a8-ba69-eaa3c324b3cd&lang=en-us
Designed for stealth operation, literally, by design.
(and you can optionally turn the LCD back on)
Not exactly a fan’less’ operation…but fan-off operation.
Can you specify the type of benchmark? I may have to do some search as I haven’t done benchmark for any purpose till now…
I am wondering if you disable all the P cores and limit TDP. Pull out the fan and replace it with a copper shim inside, would that be possible?
I saw some kind of school laptop using a configuration that just uses e-cores working fanless that way. Of course, the disabled P cores do consume power and generate more heat than a setup with only e cores but I guess it works.
Maybe another way is to try for a Raspberry Pi 5 (I prefer the RK3588 but software support is bad) compute module and design a carrier board to work with the Framework chassis.
There’s a guy doing a 3D printed rugged design for the Framework. Maybe you can leverage on it. (I have been wanting to leverage on it and 3D print a glass fibre top cover and input cover, maybe even with a keys with slightly more travel/more obvious tactile bump, and even a more rounded key shape similar to thinkpads, but money has been running kinda tight recently so yea).
I have worked with this kind of pc a lot, but they are a whole different game than a laptop. Convective fins are neat if you don’t have to follow the same skin temperature constraints a laptop has to. Hell even ignoring skin temp, where would you put them on a laptop? Only place with the right orientation would be the display but then how’d you get the heat there, can’t just use a heatpipe since there is a hinge? Also the heat-sinks for chips that are more than a couple watts get heavy quick.
Good thing you can replace the fan if or when it becomes too loud at min, on the 16 you don’t even have to repaste the heatsink for that, even though you probably should at that point.
That sounds more like a pro to having a fan to me since you can use more fan to make it perform good enough for longer, I don’t see how it being passive magically gives you more performance later if you need more unless you mod a fan onto it.
My suggestion was setting the max fan speed as high as you are comfortable with and living with the performance that gives you, not just not stressing the cpu hard enough to make the fan spin up.
Imo unlikely, we are allready in a niche within a niche, this is a couple levels further down that path, especially a 16inch passive.
That sounds like something a lot more reasonable. Mainly what framework would have to provide would be access to the tdp settings, that fan control part could be done by the community.
That is probably more of a diy option
A U version of the 16 would probably be neat though I kinda doubt the performance difference between a 7840u capped to 10w and a 7840hs capped to 10w is going to be huge since they are just different bins of exactly the same silicon with different configurations so it may not be worth the extra SKU.
That sounds a lot like my “sleep-mode” script I have on my t480s, turn of fan, turn off leds, screen is turned off when I close it anyway and it literally doesn’t have speakers anymore so that parts out.
Leave the p-cores and just limit tdp, the e-cores are more about die size efficiency than power efficiency.
3 Likes
Thank you.
CPU Single Threaded 1211 is too weak. Apple M2 Macbook Air 2023 can do 3961 which is 3.27x as fast. It’s not feasible to spend money buying a system this weak, I think.
As a comparison, my 5-year-old average laptop does 2400. I wouldn’t spend money for less than that.
Airtop3 cools 200W, a laptop can’t compete with that level of sophistication, but 30W is totally doable.
But, agreed, assuming you can easily control the fans without resorting to hacks, it’s better to have it, you may run it one day in case you need it. There is still a dust downside, but that’s OK.
That’s what I’m talking about when you disable turbo & go in powersave governor (as you can see in above screenshot with btop monitoring linux kernel compilation: 1.6GHz upper limit), I’ve seen absolute no lag in day-to-day tasks, most of bottleneck is not CPU or GPU these days, but rather network & SSD or Cache performance.
I’m still able to play 2k video without any issue, do multiple task at once… The only thing that may be affected is probably that the compilation time of the software will go high, so if you’re on Gentoo or source-only distro / doing ML on native hardware then only it’ll majorly affect you. Application opening time or bootup time is majorly affected by random reads from secondary disk, its the bottleneck, you can use ZFS or bcachefs as your rootfs to make it faster by a large extent, they use ARC (adaptive replacement cache) to keep things in RAM when possible.
Airtop is also doesn’t have to avoid burning the user which helps a lot with passive cooling.
I don’t think 30w in a laptop form-factor is doable (at least not without unusable surface temperatures) but I’d love to be proven wrong there.
Bear in mind, this test was with the 11th gen Intel i5. This is an example, but will not necessarily reflect the scores you would get with the AMD processor in the Framework 16 if you limited performance enough to get decent temps without the fans running. Until they get out in the field and people start testing them that way, we don’t really know how they would do in this scenario.
Actually, I believe it’s quite the opposite, in the past memory and storage were very slow so the CPU had to wait for them a lot. Now, thanks to improvements in memories and especially in storage, the CPU does not wait for them for long, and nowadays it’s the single-core performance that matters. Meanwhile, CPU tech did not improve so fast due to hitting the ceiling on single-core. In the past, manufacturers were just keep increasing the processor frequencies. At some point (around 5GHz) they realized they couldn’t keep increasing it, so processor gains diminished and focused on multi-core optimizations.
I have a laptop that scores 1700 on single-thread performance, with 20GB RAM and SSD, and it’s not pleasant for me to work on it.
Regardless of whether it is fast enough or not, it can’t compete with Macbook Air with this performance, not even at half the price.
It drops to 22 once heatsoaks but still pretty impressive even if the skin temperatures are pretty brutal.
I mean a fanless framework 16 could probably do it’s 45 until it’s heatsoaked too how long that will take is an entirely different question though, looking forward to someone trying it.
If you can get 20-30w out of this that should be more than plenty for most use cases. Definitely will give you more than 5276 passmark for whatever that meme benchmark is worth XD.
You won’t be beating a macbook in that ring anytime soon but it may be sufficient.
It’s not brutal, it gets 45 at peak.
Anyway, it’s possible, and doable, I guess we can agree on that. Framework does not need to beat the best, though I don’t see why it can’t, any manufacturer can beat Apple if wanted and is willing to spend resources on this, but offering a close alternative would be sufficient, like passively cooling a 20W system.
I love Framework, since I love repairability and customizability. I have another critical requirement, which is silence. In the future, I hope Framework checks that box too.
I guess that’s all from me.
That’s a lot, especially for a metal skinned laptop but right at the top of the allowed limit (weird that, allmost like it throttles based on that XD).
Maybe the 16 with the fan off already does that, who knows but I am sure thanks to this thread someone is going to try it once it’s out.
That’s not how those numbers work…
30W is the measured max power draw Notebookcheck measured for the MBA, though they noted it’s actually a constraint of the power supply Apple provides in box. When using a 100W PS, they measured up to 50W max power draw.
This still isn’t the whole picture however, since max power draw doesn’t mean anything if you can only pull that for say… a few seconds.
Their testing shows the MBA, cooled as it is can draw max power for literally a few seconds, immediately dropping power to a long term average of… 9W.
Passive cooling a performance chip is impossible even for Apple’s silicon right now. If anything, the previous poster’s results showing 22W on the FM while heatsoaked is impressive.
For additional context, the M2 Chip is running 108* in their temperature tests.
5 Likes
I think you are mistaken. ~30W is sustained not peak. Peak draw is 50W.
9W is sustained CPU draw, 30W is for the entire system.
Actually, to be precise, average power draw on load is 28.5W according to review. This is passively cooled without skin temperature exceeding 45 C.
30W load maximum is not sustained on the M2 chip for longer than a few seconds. This is the only load the cooling system manages. The screen, battery, and other power drawing components do not get cooled by the cooling system.
That the surface temp is 45 while the cpu is at 108 only shows that the MBAs cooling system is doing a poor job of transfering heat, rather than being a good point as you seem to think.
The MBA performs well for its cooling system, but it does not have a good cooling system. There’s a reason the MBPs use fans.
1 Like