The whole point is that the CPU is only a single part of the equation. Yes, you're not going to burn out the CPU itself by unlimiting PL1/2 (although if the system vendor cheaped out on power circuitry because they'd only designed for sustained 20W draw then you might burn that out), but you're now generating more heat than the system is designed to dissipate. This may result in obvious outcomes like the chassis heating up enough to burn your legs, but it may also result in other components being operated outside their thermal limits and their lifetime being shortened as a result.
The experience of many users shows otherwise --- ThrottleStop has similar scary disclaimers, but I have yet to see any evidence of hardware damage caused by opening up the power limits and letting the CPU naturally attain its full performance.
but you're now generating more heat than the system is designed to dissipate
I believe that's more accurately rephrased as "now you have more performance than they wanted you to have", because that's how it's being used in practice.
FYI, it is painfully obvious that you're toeing some sort of company line here.
Edit: and getting flagged for pointing out the truth should itself be quite telling. ;-)
> "now you have more performance than they wanted you to have"
"An unthrottled CPU without any OS-level policy will generate enough heat that the case temperature may rise to levels that violate safety regulations" can certainly be interpreted as "now you have more performance than they wanted you to have", yes.
> FYI, it is painfully obvious that you're toeing some sort of company line here.
I work for a company that develops trucks. While thermal dissipation is something that we do need to worry about, I can promise that it is absolutely not at the scale involved here. I've also been pretty consistent in criticising Intel for refusing to document DPTF. Whose company line do you think I'm toeing here?
"An unthrottled CPU without any OS-level policy will generate enough heat that the case temperature may rise to levels that violate safety regulations" can certainly be interpreted as "now you have more performance than they wanted you to have", yes.
Despite literally zero evidence of that actually being a problem in practice?
I've also been pretty consistent in criticising Intel for refusing to document DPTF. Whose company line do you think I'm toeing here?
Not Intel, but there's plenty of evidence that DPTF just cripples performance; and why else would you jump in to an article about unlocking manufacturer-crippled products with FUD like "not the correct approach"? It's like posting "may cause processor damage" and advocating for stock speeds on every article about overclocking.
For the record, I've been overclocking (not at extreme or competitive levels) since the early 90s and haven't killed any hardware because of it. The system I'm posting this from has been on a 50% overclock for over a decade and it's still rock-solid stable.
> Despite literally zero evidence of that actually being a problem in practice?
Sustained contact with metal at a temperature as low as 45C is sufficient to cause discomfort, and it's a surprisingly small rise above that before you can actually cause burns (https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&d...). Skin temperature is a standard input to DPTF policies, and during my testing I was certainly able to get the skin temperature of my test laptop well above that level - DPTF would trigger throttling instead.
> why else would you jump in to an article about unlocking manufacturer-crippled products with FUD like "not the correct approach"?
DPTF defaults to a "safe" configuration, where "safe" is a euphemism for "almost unusably slow". If what you care about is your laptop not being almost unusably slow, it's reasonable to choose an approach that obtains that without introducing any additional thermal concerns. If what you care about is obtaining the absolute maximum performance possible from your hardware with no regard to any safety or longevity concerns, then yes, implementing the DPTF policy is probably not what you want to do. But recommending the latter without providing any information about the tradeoffs is irresponsible.
For a layman's overview of this, Linus from LinusTechTips gives a basic rundown of the relevant safety regulation and some numbers in one of his videos featuring the M1 Macbook Air and its cooling solution.[1]
While developing my DPTF implementation for Linux I caused the outer skin of my test laptop to reach a temperature that would have caused burns to my legs if I had left it sitting on them for more than a couple of minutes. Now, I have working thermoreception, so I'd probably have noticed this and put it somewhere that wasn't my legs before that happened, but there are various scenarios in which people don't and manufacturers of objects that are intended to be put in your lap are kind of reluctant to design them in such a way that they could cause injury, which is why the manufacturer policies ensure that doesn't happen.
If you're fully aware of the possible consequences of ignoring those policies then go wild. I think it's the wrong choice for most people, and as I said I think it's irresponsible to suggest people change the thermal policy in that way without ensuring they're aware of the possible consequences. If saying "You should only do this if you're ok with your laptop potentially becoming hot enough to burn human skin" is authoritarian then I guess I'm authoritarian?
Your work on reverse engineering DPTF seems abandoned. Is it that everything was upstreamed? Because there's still a performance gap between Windows and Linux.
Also, why is Intel so reluctant on fixing this themselves? It looks so simple to me too fix this by releasing the docs and fixing thermald, and still, here we are having to use throttled for those wanting a little bit more punch.
And finally, how is it that nobody it's creating a custom, patched thermald that takes all this into account? Is it that the process for reverse engineering is too complicated and that we don't have the full picture yet? It shouldn't be too difficult to read the values Windows takes and reverse engineer it, although it may be tougher than it seems at first.
I worked on it to the extent that it worked for my use case, given it was a spare time project. It's all been merged upstream, and Intel (to their credit) are maintaining it now even though still refusing to document it.
> Despite literally zero evidence of that actually being a problem in practice?
You choosing to not even look for evidence so you could make an argument in ignorance does not automatically mean it doesn't exist. This is literally the first google result for "laptop case temperature causing burns": https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4292129/
> Temperatures between 43 and 47 °C can cause this skin condition; modern laptops can generate temperatures in this range. Indeed, laptops with powerful processors can reach temperatures of 50 °C and be associated with burns.
That's a flagged and dead reply to a sibling comment which would look schizophrenic as a response to this one - on top of everything else that is already wrong with it.
Not Intel, but there's plenty of evidence that DPTF just cripples performance; and why else would you jump in to an article about unlocking manufacturer-crippled products with FUD like "not the correct approach"?
Seems like there are quite a few other reasons why someone would "jump in", including being a relative expert. Maybe you could think of a few others yourself?
Slap enough cooling on a CPU and it’s never going to thermal throttle.
They added thermal throttling because most users only max out their CPU a tiny fraction of the time so the thermal inertia can compensate for inadequate cooling. It’s infrequent enough that most users could even disable it without noticing a difference. The problem is under the wrong conditions Aka inadequate cooling plus high ambient temperatures plus extreme workload = dead CPU.
And no this isn’t theoretical, one summer working without AC I cooked a CPU before they added thermal protection. It actually happened to several people in the same heatwave.
I recall 3 different stages, thermal protection was first designed to save chips if the cooling fans failed etc. It was very conservative because the only option was to turn the system off. SpeedStep was designed to save laptop battery life, so they never needed to thermal throttle in normal operation.
Only fairly recently with Turbo Boost did Intel combine both ideas. It’s actually a fairly clever optimization as single threaded workloads benefit quite a bit from the extra thermal overhead.
The Pentium III was the last Intel CPU where shutdown was the only possible response to overheating, when the CPU asserted THERMTRIP#. The Pentium 4 has bidirectional PROCHOT# and the Thermal Control Circuit, which divides the clock down whenever the processor is hot, and continues to run.
It's also what enables some of the more modern packaging technologies.
Silicon aging is worse at higher temperatures, people are going to kill their 7/10nm chips far quicker if they keep overclocking as they've previously done. But that'll be Big Silicon's fault too, no doubt.
