Testing Folding Without Wasting a WU
Moderators: Site Moderators, FAHC Science Team
Testing Folding Without Wasting a WU
I'm dealing with a separate issue here and it got me thinking: I'd like to know my folding configuration is correct before requesting WUs. Is it possible, at any level, to test a WU? Perhaps keep one of each WU that works on each core type, not one you actually submit and one that's small enough to complete. If no official FAHClient feature, something I can tie together through FAHClient CLI, the telnet interface, or the FAH API? (I have no issue with programming/scripting.)
Re: Testing Folding Without Wasting a WU
At least part of what you'd like to do do can be done with FAHBench (which can be seeded with WUs from other projects).
When a WU fails, FAHClient makes an effort to report which WU failed (and why) so that the WU can be assigned to someone else. Returning that error-report promptly minimizes the waste of WUs, though it's not infallible. When a failed WU is repeatedly reported, it's eventually classified as a corrupt WU, but that shouldn't matter if you're getting a variety of assignments. Of course each failure does count against your bonus points ... and if your system reports too many failures, it's eventually classified as a corrupt system and it'll stop getting assignments.
A much better plan is to fix whatever is causing the failures and we're happy to work with you to accomplish that promptly so you become a constructive donor.
When a WU fails, FAHClient makes an effort to report which WU failed (and why) so that the WU can be assigned to someone else. Returning that error-report promptly minimizes the waste of WUs, though it's not infallible. When a failed WU is repeatedly reported, it's eventually classified as a corrupt WU, but that shouldn't matter if you're getting a variety of assignments. Of course each failure does count against your bonus points ... and if your system reports too many failures, it's eventually classified as a corrupt system and it'll stop getting assignments.
A much better plan is to fix whatever is causing the failures and we're happy to work with you to accomplish that promptly so you become a constructive donor.
Posting FAH's log:
How to provide enough info to get helpful support.
How to provide enough info to get helpful support.
Re: Testing Folding Without Wasting a WU
The reason I ask is because I see frequently other members concerned about faulty hardware. The concern is that you're wasting WUs. Obviously, there's different types of failure, but one can only know if their setup is working by trying to fold.
So FAHBench is the solution? My issue is I somewhat frequently switch out hardware, or may change drivers around to improve performance. So if I can fold using FAHBench, assuming no network issues are present, I should have an issue running FAHClient?
So FAHBench is the solution? My issue is I somewhat frequently switch out hardware, or may change drivers around to improve performance. So if I can fold using FAHBench, assuming no network issues are present, I should have an issue running FAHClient?
Re: Testing Folding Without Wasting a WU
FAHBench will only test your GPU. It won't test your CPU, and it will only test SOME iGPUs (for example, I don't think Intel iGPUs are supported). If a wu works in FAHBench, and you can access the internet, you should be able to fold in FAHClient. If you download a work unit to test but don't actually run it, it may cause delays to it being sent to another user.
Re: Testing Folding Without Wasting a WU
That's not something that can be proven, but it sounds plausible. Try it out.
Posting FAH's log:
How to provide enough info to get helpful support.
How to provide enough info to get helpful support.
Re: Testing Folding Without Wasting a WU
As long as your good WU count is 80% or more, you're good (bonus points will be awarded).
I regularly change drivers, update OS, swap out cards, erroneously overclock (too high) causing WUs to fail; crashes, or system hardware failure...
Anything can happen to a server.
Then again, once a setting works, I leave it running for a while.
I don't use FAH to define my hardware's overclocking parameters. I mainly use it to fold.
Overclocking is just a bonus on the side I can do.
Despite all that, I'm still at well over 97% accuracy, even when dropping more than 200WUs in total.
200 'lost' WUs may seem like a lot, but not if you consider I've uploaded over 10k of good WUs!
It's all about the good/bad WU ratio you're uploading.
If you don't overclock, occasionally dropping WUs shouldn't be a worry.
If you are overclocking, FAHBench is a good way to set initial overclocking on cards.
Once you know at what OC frequency the card fails a WU, lower the OC value a bit, until it runs stable, and then run FAH for folding.
FAHControl's log also is a great help to set Overclocking.
For instance, a situation happening to my server in the past:
"my server is outside. In the mornings, it registers more overclocking errors than during the day.
Why? Because the ambient temperature is cooler in the mornings, than during the day, and the Nvidia drivers card start overclocking to higher speeds during that time, exceeding boost speeds.
So because of that, I have to lower OC values to a point where on a cold morning, it won't register any OC errors anymore."
I've done that, and now my server will run FAH for days on end without hitting any OC errors (checking FAHControl's logs).
FAHControl will allow you to see OC errors and correct them, before a WU fails, as it'll try reprocessing the WU up to (I believe) 4x before cancelling the WU.
After a while, your system will be running autonomously, without any user interaction necessary.
And like said before, WUs don't really get lost. They just get transferred to another user folding the same WU.
I regularly change drivers, update OS, swap out cards, erroneously overclock (too high) causing WUs to fail; crashes, or system hardware failure...
Anything can happen to a server.
Then again, once a setting works, I leave it running for a while.
I don't use FAH to define my hardware's overclocking parameters. I mainly use it to fold.
Overclocking is just a bonus on the side I can do.
Despite all that, I'm still at well over 97% accuracy, even when dropping more than 200WUs in total.
200 'lost' WUs may seem like a lot, but not if you consider I've uploaded over 10k of good WUs!
It's all about the good/bad WU ratio you're uploading.
If you don't overclock, occasionally dropping WUs shouldn't be a worry.
If you are overclocking, FAHBench is a good way to set initial overclocking on cards.
Once you know at what OC frequency the card fails a WU, lower the OC value a bit, until it runs stable, and then run FAH for folding.
FAHControl's log also is a great help to set Overclocking.
For instance, a situation happening to my server in the past:
"my server is outside. In the mornings, it registers more overclocking errors than during the day.
Why? Because the ambient temperature is cooler in the mornings, than during the day, and the Nvidia drivers card start overclocking to higher speeds during that time, exceeding boost speeds.
So because of that, I have to lower OC values to a point where on a cold morning, it won't register any OC errors anymore."
I've done that, and now my server will run FAH for days on end without hitting any OC errors (checking FAHControl's logs).
FAHControl will allow you to see OC errors and correct them, before a WU fails, as it'll try reprocessing the WU up to (I believe) 4x before cancelling the WU.
After a while, your system will be running autonomously, without any user interaction necessary.
And like said before, WUs don't really get lost. They just get transferred to another user folding the same WU.