Folding Forum

ajm wrote:In Windows, there are quite easy ways to limit the GPU usage (and thus the heat). Look at MSI Afterburner, for example:

2080S water cooled:


2070S air cooled:


But it would be crazy to try and integrate this kind of control in FAH. For one thing, FAH should do this in Windows, Linux and OSX, and no-one can, yet.

With the development of ever more performing consumers' GPUs, maybe Microsoft, Apple and some distros will implement such a control at some point, but you can't demand that from FAH. FAH is not an universal tweaking utility.

If it's just temperature you're worried about, you can decouple the temp and power regulator sliders, and just cap the temperature.
Capping the power, is great when combining with an Overclock, as long as the temperature stays within a specific range.
setting fan speed to max (or near to max) also helps out a lot!
I don't bother with temperature capping, as I know what power levels I have to set my GPU, to run at what temperatures.
However, if your PC is in an area with highly fluctuating ambient temperatures, setting the temperature to a fixed maximum, is a better solution, than modifying the power levels.

> Than why FAH doesn't have option to send data in bursts? As even amateurs could make simple utility to prevent gpu from heating down and heating up! As that was the problem! We are not talking here about anything overly complicated. And problem would be solved...

We've explored this in the past, and have been working with the hardware vendors to figure out if there is a safe and reliable way to "throttle back" GPU effort. Unfortunately, we haven't come up with anything good together yet.

If you find any examples of other scientific computing packages that have figured out a good way to do this, please point us to it! As the other folks here have said, we *did* try, but the approach didn't end up doing what we hoped.

~ John Chodera // MSKCC

GPU's 'heating and cooling' is a non issue.
The GPU die substrate, as well as the heat sink provides it with enough remnant heat for the GPU to constantly experience thermal stresses.
It's like saying that the combustion process (constant heat and cold) inside a cylinder of a car wears out the engine.
The silicon expands and contracts, but it's made for that.
What's bad is if the temperatures fluctuate from cold <45C to hot >75C on a regular cycle (like once a minute, or so).
Fluctuating wattage differences from the GPU waiting for CPU data isn't an issue, and won't destroy a GPU anytime soon.

When PCs of the 80s and 90s failed, it wasn't because the PC was turned on and off so many times, that the CPU broke.
It more than likely was the electrolytic capacitors failing (oil based caps that expand and contract with fluctuating on/off states).
Those caps are now all replaced with solid caps. They don't experience the failing, and in theory should last about as long as a memory module, or perhaps even a GPU/CPU processor..