ust curious: How many out-of-house tests do you actually need to check off "beta" status as reliable?
It really depends. For technical issues / fixes like here it doesn't need much, just showing that the bug of the previous version which basically crashed each task is fixed.
To check the scientific / numerical validity of the results when we change the calculation it takes much more, recently a whole "Engineering run" set up specifically for that with tens or hundreds of thousands of tasks.
I am experiencing a wicked issue that despite dispatching FGRP1B jobs two two different (AMD) cards they are still being run on just one card. I have detailed my problem here:
There is now v. 1.22
)
There is now v. 1.22 (Windows, currently Nvidia only, Beta test).
Unfortunately our possibilities for in-house testing of Windows GPU Apps are very limited, so I'm relyng on you guys.
BM
No change in time to
)
No change in time to completion from what I can see.
Was there a different purpose for this upgrade?
Thanks!
Bernd Machenschalk
)
hmm.. I'll fire back up a MS machine on Friday morning.....
So the 1.20 app doesn't get
)
So the 1.20 app doesn't get tasks anymore? I'm back to running 1.17 on a host that doesn't do beta apps.
v1.22 is a fixed v1.21 and
)
v1.22 is a fixed v1.21 and meant to fix the same issue.
BM
Bernd Machenschalk
)
I am no "expert", but to me v. 1.22 beta on GPUs seems to be running fine.
I've tested, rather I am still testing, NVIDIAs GPUs 750ti, 960, TITAN X, TITAN V and GV100.
Hope this helps the project!
Run times are OK - as far as I'm concerned.
Just curious: How many out-of-house tests do you actually need to check off "beta" status as reliable?
San-Fernando-Valley wrote:ust
)
It really depends. For technical issues / fixes like here it doesn't need much, just showing that the bug of the previous version which basically crashed each task is fixed.
To check the scientific / numerical validity of the results when we change the calculation it takes much more, recently a whole "Engineering run" set up specifically for that with tens or hundreds of thousands of tasks.
BM
Hi Bernd,I am experiencing
)
Hi Bernd,
I am experiencing a wicked issue that despite dispatching FGRP1B jobs two two different (AMD) cards they are still being run on just one card. I have detailed my problem here:
https://einsteinathome.org/content/dual-amd-computes-first-card-only
My BOINC setup seems ok, as this worked before when I had AMD + Nvidia crunching in parallel.
Any insight would be highly appreciated!
######
edit some hours later:
The problem seems to be AMDs OpenCL implementations conflicting. It works on ROCm.
Is the project over now? No
)
Is the project over now?
No task ready to send. Work generator not running.
Still 403k WUs pending. Doubt
)
Still 403k WUs pending. Doubt the work generator will run until that drops.