Test WU

[ProDigit]

I'd really appreciate it, if there was some way of getting a test WU on FAH control, to test overclocking, with feedback on error ratios.
I currently have to overclock on the fly, resulting in bad WUs in case I'm not quick enough to down-throttle vram or GPU speed.
Sometimes the overclock seems very stable, and I back off a few Mhz, only to see the end result gives a bad WU.
If we had some setting in FahControl, that allows us to play around a bit with settings, it would help. Especially if it has some sort of feedback mechanism on error ratios.

Here, room temperature changes as high as 5 degrees, affect my overclocks, they sometimes end up making a WU bad (5 degrees F increase in ambient room temperature, could result in 5 degrees C increase in the card).

[Joe_H]

It is very unlikely that will ever happen. As has been stated a number of times by PG and repeated here, overclocking is not supported by the Folding@home project.

The closest you can get is the ability to capture a copy of a WU and use that in FAHBench. I have not done that myself, but others have in the past.

[ProDigit]

Thanks!
I'll give it a try.

[bruce]

Overclocking "on-the-fly" will never work. If you're playing a game and one pixel is computed in error, you'll never notice so a low rate of failures is acceptable. With, FAH, the allowable error rate is zero since every calculation matters. Your hardware is either STABLE or UNSTABLE.

FAH used to finish a certain percentage of bad WUs, only to have them checked carefully after they were uploaded. A consideral number of client enhancements were added so that a bad WU would be aborted at the point of error whenever. (Allowing you to continue processing was a waste of resources.)

In some cases, there is a coherent checkpoint and reprocessing that portion since the last checkpoint MIGHT be successful.

[ProDigit]

I have been getting some good and stable results with overclocking. So saying it will never work is not correct.

I use ram and GPU overclocking in conjunction with monitoring GPU activity, voltage and temperature charts.
When ram is overclocked too high, there are activity dips I can see every few minutes in the chart. So I dial down ram overclock until they disappear. GPU overclocking usually dials up the voltage, resulting in hot running cards, and an immediate cancellation of the WU when the overclock is too high.
In the process of setting up, I lose a few bad WUs. But then it runs reliably.


My experience with overclocking those Pascal cards, the 1030, 1050 and 1060 for folding, is that GPU overclocking works stably up to about 100Mhz, of a Max of 175-190Mhz peak overclock.
In gaming terms that's a very mild overclock.

Both GPU and VRam overclocks are very close to one another between brands.
MSI, EVGA, Zotac, and PNY, all have very close to the same overclock ratios before failure.

The vram, some I can overclock by 1Ghz!
But I found for folding, it runs most stable at 575-600Mhz.
That is still a significant overclock, that does translate in noticeably higher PPDs.

I just don't like uploading a few bad WUs every time I install a new graphics card.
Usually 3 to 4 bad WUs are necessary, before a stable enough overclock is reached.

[foldy]

1) Install http://fahbench.github.io/
2) From a running FAH there is a work units folder e.g. C:\ProgramData\FAHClient\work\00\01
3) Inside fahbench folder there is a workunits folder
4) Create a new folder inside fahbench workunits and name it like the current running work unit e.g. 14567
5) Copy from FAH work unit the files integrator.xml, system.xml, state.xml to fahbench work unit 14567
6) Copy from fahbench\workunits\nav\wu.json to fahbench\workunits\14567
7) Start fahbench and select WU = 14567 and "Run length" = how long you want to test
8) Click start button and your work unit is running

Can give you fahbench prepared work units 9415, 11718, 13782, 14156
https://ufile.io/l8gvf

By the way the hardest GPU OC test I know is https://benchmark.unigine.com/superposition?lang=en