A colleague of mine just picked one of these up, he is VERY happy and excited about it! More than he should be perhaps, it makes me want one too
We both work at a simulation company and it got us wondering, have these been put through any sort of analyses to check they don’t overheat?
Interested to know
The CPU dynamically turbos to the highest performance point it can reach as defined by software. There’s a LOT of stuff that goes into defining this that I’ve only ever scratched the surface of, so the simple version of the story is, it wants to go as fast as possible until software tells it “Whoa, hey, you’re getting too hot” at which point it’ll slow down. I’m assuming you’re an engineer at this point, so it won’t surprise you that with that simplified description the problem now boils down into basic thermodynamics.
The CPU will balance its input power with the thermal dissipation ability of the built-in thermal system. This basically means the heat transfer coefficient of the system (fan, heat sink, heat pipes, and thermal interface material between all that and the CPU die) defined as a function of the die temperature. What all this means is, the CPU will always balance its input power and fan speed dynamically with package temperature, CPU demand, task type, GPU activity, and a whole host of other variables, to ensure that it doesn’t overheat.
The Framework team has published that the system cooling solution is tuned for a 28W sustained load (meaning longer computations like you’d see if you’re doing simulations), but given how messy the math I described above is, I’d consider this to be a conservative estimate. You can pump that estimate up significantly higher if you’re in a cold room, or are on a laptop cooling pad, etc., and you can drop it significantly lower if your fans are disabled for quiet running, or your system is clogged with dust, or what have you. Overall though, to answer your question, in all those situations the one guarantee is, the CPU will limit its performance to ensure it doesn’t ever approach the point of overheating. Modern CPUs die (no pun intended) when running with temperatures in excess of 100C, for extended periods of time; additionally modern generations of laptop CPUs have been known to last for years at those temperatures. With all that being said I’ve not seen my framework laptop come anywhere close to 100, though I do keep my laptop on a cooling pad when running stressful tasks.
Really interesting, thanks for sharing!
I didn’t really think about it this way around. We’re always trying to optimise airflow for more efficient cooling, but if you work it the other way around I guess you might have a lower performance but at least you guarantee a longer lifespan.
(And yeah, I have a masters in thermodynamics/fluid mechanics for my sins )
As a fellow mechanical engineer who did lots of multiphysics modeling for his last role… I assumed you were immersed in that field when your first thought was CFD
Yep! During our engineering design phases the engineering team did thermal simulations before we build the first hardware.
After we had actual hardware, we did validation using temperature probes and IR cameras to tune the firmware and validate the hardware design.