Folding Forum

It has been explained above why FAH (and most other software that really utilise GPUs) hog the resource but you aren't happy with the answer - your 5 minute solution maybe wasn't quite a solution (Computer Scientists maybe do know what they are talking about?) … If you throttle back you GPU by basically bottlenecking your CPU from servicing the GPUs requirements it doesn't surprise me that it won't perform properly - but it was a brave try … There are technical solutions are known, but until the OS developers are prepared to incorporate them then software application vendors will for the most part have these issues … the time slicing approach was another elegant solution - which regrettably only worked until the hardware failed early.

If people are finding that GPU (or indeed CPU) folding is impairing their ability to use their kit for what they want to then much as though they will be missed then rational solution is not to fold … Many people find ways to make folding work around their other uses for their kit - One person recently worked out their own start-up scripts for when they had left the office and then a shutdown script that set the slots to finish a sensible time before they were due back in the office - this person had a fast GPU and found a happy balance - use their machine without anything using resources and potentially interfering during the day and then grind through a series of WUs overnight- but this was their choice.

Everything was perfect - until it wasn't - but then maybe it wasn't actually perfect?

HAHA btw nvidia says it is up to you to introduce gpu usage limiter. Not sure if it is true...

No it won't crash my system, but apparently wu is crashing. It is simple stress test. I don't know why it is crashing. Couldn't this be fixed ? Because it allows you to limit gpu usage like that !

Dude can't you tell joke ?

Maybe it is nvidia's fault, or your fault dunno. But you are dissuading a lot of people from crunching. Isn't there really anything that can be done ? Why it is crashing for example, if you deny resources to one cpu core ? Couldn't this be fixed ?

You said FAH tried to give work to a gpu in a bursts, but that caused gpu to cool down and than heat up repeatedly, which isn't good for longetivity. What about interface this program with gpu fan control, any 3rd party app can do that ! And based on how much gpu usage drops, fan's rpm can be tuned down. So gpu doesn't cool quickly and temps stay stable. It takes like 1 minute to cool off.

Also you said there is no way to control gpu usage, but what cpu usage. Because gpu usage is tied up to cpu usage. What about running one stubby process on a same core, which will have similar priority to eat up cpu resources. I still don't understand why wu crashes, because resources are split. My FahCore22 usage doesn't drop below 8%. Couldn't this be programmed, so FAH isn't as sensitive, when not getting resources ? Also if this program had something like process lasso. Lets say: FahCore22 needs more cpu free, priority of stubby process can be momentarily dropped down, so it doesn't crash FahCore22. It is really hard to believe something can't be done about it.

Also every pc configuration is different and each workload is different and each cpu does better, or worse in different situations. What about some benchmarking tool, which which measure performance of a pc. And than FAH decided based on that, how to run workloads. So it is used only by 50% e.g. Or give cpu work in bursts and put some slider, so user can adjust that based on cpu usage he sees, cpu doesn't have problem of cooling down and heating up by different usage on a single core. Don't know how hard it would be, so cpu work averages like between 35-70%. Or at least do that for most used cpu models, if that would be to hard to implement generally.

Few years ago when my GPU wasn't a powerful one, I got rid of lagging screen by disabling animation, disabling hardware acceleration on browsers and media players. It worked fine during that system configuration until I upgraded my system.

This won't work for me, i need at least 30%~ gpu usage free, when watching tv shows. Not sure if disabling hw acceleration would do anything, if gpu will be constantly at 99% load anyways. I am now testing new setting got to 29% without crash. Hope it won't crash, that would be nice

empleat wrote:This won't work for me, i need at least 30%~ gpu usage free, when watching tv shows. Not sure if disabling hw acceleration would do anything, if gpu will be constantly at 99% load anyways. I am now testing new setting got to 29% without crash. Hope it won't crash, that would be nice

you're asking the wrong people for a fix. In the OS, there are options to set priorities for cpu tasks. There is no such option for the gpu. End. Period. Full stop.

Go ask Microsoft.

I didn't ask about gpu priority if read what i have said. Since gpu usage determined by cpu usage, maybe something so it averages at 50%, cpu doesn't have problem of cooling off and heating up, based on change in usage on single cpu core. Maybe a program that would put cpu to sleep, based on cpu usage. And than i talked about more stuff completely something different from gpu priority...

I got yesterday to like 35% of wu, but today after i started computer it crashed, after a while. I also noticed, that even it says work bad unit and wu crashes, it still uploads results. Does this mean, even if wu crashed. Work that has been done, will be uploaded and useful ?

It is hard to believe that something can't be done, nvidia says it is up to you, but i am not sure. Anyways scientists are usually, that bunch, if something works they don't change it. Or not best at making user friendly programs.

BTW Thank you PantherX , i had freezes each x second and now, even when i was doing nothing and gpu usage sits on 92-97%, but still i can watch tv shows without a problem, after disabling a hw acceleration in media player. I am using software for frame interpolation and it was said you have to turn than on and it is also mainly cpu intensive. So i can crunch now on gpu and yet watch tv shows. NICE.

The processing uploads an error report so that a. - they can reassign the WU to someone else, and b. - if they more than a set number of errors reported on a WU, it will be suspended from being assigned. They may eventually look to see if the error is from the setup, or just a trajectory gone wrong.

The error report may get you some proportional credit based on the full base points, the WU report is not usable other than for the above.