I ran around 250 v1.36 WUs on my various machines with Win7-64. No errors, no validation problems. Of course haven't been using 7.0.64 in quite a while. There's MANY updates and fixes in BOINC 7.1.18.
It would be interesting to see if the setting "Use at most _% CPU time" is related to this problem. Since the app is now deprecated, it can be tested on Albert@Home only.
Cheers
HBE
It has no impact for Boinc 7.0.42. And: We are talking about a GPU application.
Right, this triggered me to think about the following question: Does CPU throttling affect GPU apps in BOINC ????
And the answer is .... it depends :-)
According to a recent discussion on BOINC-dev mailing list, different BOINC versions differ in this questions, e.g the latest actually does CPU-throttle GPU apps, versions build between Oct 2008 and Jan 2013 however don't.
Meanwhile.....and independently,
A bug fix has been submitted to the BOINC API that might cure the problem discussed in this thread, so if we are confident that the fix is indeed working ok, we would have to release new app versions. Stay tuned.
As a maybe too wide question in response, how much should we (clients) worry about upgrading to the latest BOINC version? Obviously a lot of people aren't going to be following the forums here in detail, so I assume that the occasional update to the latest stable version makes sense - but equally, there is no need to update every time a new BOINC version is released? (unless you advise us specifically to do so).
We deployed BRP5 v1.37 on albert (test project) which contains another proposed upstream fix. In order to limit our deployment effort we just released the Win32-CUDA and Win64-OpenCL versions.
We deployed BRP5 v1.37 on albert (test project) which contains another proposed upstream fix. In order to limit our deployment effort we just released the Win32-CUDA and Win64-OpenCL versions.
I've got both those apps running on my i7-2600K/GTX460/HD7770 host with Boinc 7.1.18,
one anomaly, if you change the cache size (by as little as 0.1 days), the percentage progress stops (on both apps), while the app continues with unchanged GPU load and time still incrementing, changing the % back doesn't restart progress being reported,
suspending GPU usage (so the app exits) then resuming does get progress going again from the last checkpoint, (there was no progress inspite of the GPU still being loaded).
Edit: changing the % of processors allowed also stops progress, and it's the same on Boinc 7.2.1
Two questions:
1) Can you reproduce that behaviour with older BRP5 versions as well (say, on einstein)?
2) When you talk about changing the prefs, did you do that in the manager or via the web interface (followed by an "Update")?
Two questions:
1) Can you reproduce that behaviour with older BRP5 versions as well (say, on einstein)?
2) When you talk about changing the prefs, did you do that in the manager or via the web interface (followed by an "Update")?
Thanks,
Oliver
I can't reproduce it with the older BRP5 apps at Einstein (both Cuda32 and OpenCL-ATI), these continue when changing the preferences.
I was changing the preferences in the Manager only.
Edit: Changing the Web computing preferences at Seti, then updating (while Boinc is still set to use Local preferences) also stops the % progress with the Albert@home apps, but is O.K on the apps at Einstein.
We deployed BRP5 v1.38 on albert (test project) which contains another set of upstream patches that should (hopefully) finally fix the issue. In order to limit our deployment effort we just released the Win32-CUDA and Win64-OpenCL versions.
@Rechenkuenstler/Claggy: please test and report back.
We deployed BRP5 v1.38 on albert (test project) which contains another set of upstream patches that should (hopefully) finally fix the issue. In order to limit our deployment effort we just released the Win32-CUDA and Win64-OpenCL versions.
@Rechenkuenstler/Claggy: please test and report back.
Thanks,
Oliver
That looks O.K, changing Local preferences doesn't cause app's progress to stall any longer.
We deployed BRP5 v1.38 on albert (test project) which contains another set of upstream patches that should (hopefully) finally fix the issue. In order to limit our deployment effort we just released the Win32-CUDA and Win64-OpenCL versions.
@Rechenkuenstler/Claggy: please test and report back.
Thanks,
Oliver
Sorry. I have been away for a few days. Sure, I will retest and report.
I ran around 250 v1.36 WUs on
)
I ran around 250 v1.36 WUs on my various machines with Win7-64. No errors, no validation problems. Of course haven't been using 7.0.64 in quite a while. There's MANY updates and fixes in BOINC 7.1.18.
RE: RE: Hi! It would be
)
Right, this triggered me to think about the following question: Does CPU throttling affect GPU apps in BOINC ????
And the answer is .... it depends :-)
According to a recent discussion on BOINC-dev mailing list, different BOINC versions differ in this questions, e.g the latest actually does CPU-throttle GPU apps, versions build between Oct 2008 and Jan 2013 however don't.
Meanwhile.....and independently,
A bug fix has been submitted to the BOINC API that might cure the problem discussed in this thread, so if we are confident that the fix is indeed working ok, we would have to release new app versions. Stay tuned.
Cheers
HBE
As a maybe too wide question
)
As a maybe too wide question in response, how much should we (clients) worry about upgrading to the latest BOINC version? Obviously a lot of people aren't going to be following the forums here in detail, so I assume that the occasional update to the latest stable version makes sense - but equally, there is no need to update every time a new BOINC version is released? (unless you advise us specifically to do so).
We deployed BRP5 v1.37 on
)
We deployed BRP5 v1.37 on albert (test project) which contains another proposed upstream fix. In order to limit our deployment effort we just released the Win32-CUDA and Win64-OpenCL versions.
@Rechenkuenstler: please test and report back.
Thanks,
Oliver
Einstein@Home Project
RE: We deployed BRP5 v1.37
)
I've got both those apps running on my i7-2600K/GTX460/HD7770 host with Boinc 7.1.18,
one anomaly, if you change the cache size (by as little as 0.1 days), the percentage progress stops (on both apps), while the app continues with unchanged GPU load and time still incrementing, changing the % back doesn't restart progress being reported,
suspending GPU usage (so the app exits) then resuming does get progress going again from the last checkpoint, (there was no progress inspite of the GPU still being loaded).
Edit: changing the % of processors allowed also stops progress, and it's the same on Boinc 7.2.1
Claggy
Hi Claggy, Two
)
Hi Claggy,
Two questions:
1) Can you reproduce that behaviour with older BRP5 versions as well (say, on einstein)?
2) When you talk about changing the prefs, did you do that in the manager or via the web interface (followed by an "Update")?
Thanks,
Oliver
Einstein@Home Project
RE: Hi Claggy, Two
)
I can't reproduce it with the older BRP5 apps at Einstein (both Cuda32 and OpenCL-ATI), these continue when changing the preferences.
I was changing the preferences in the Manager only.
Edit: Changing the Web computing preferences at Seti, then updating (while Boinc is still set to use Local preferences) also stops the % progress with the Albert@home apps, but is O.K on the apps at Einstein.
Claggy
We deployed BRP5 v1.38 on
)
We deployed BRP5 v1.38 on albert (test project) which contains another set of upstream patches that should (hopefully) finally fix the issue. In order to limit our deployment effort we just released the Win32-CUDA and Win64-OpenCL versions.
@Rechenkuenstler/Claggy: please test and report back.
Thanks,
Oliver
Einstein@Home Project
RE: We deployed BRP5 v1.38
)
That looks O.K, changing Local preferences doesn't cause app's progress to stall any longer.
Computer 7514
Claggy
RE: We deployed BRP5 v1.38
)
Sorry. I have been away for a few days. Sure, I will retest and report.
Cheers