In case you wonder: the BRP7 app version 0.17 (released last night) is meant to improve validation with the dominant Windows app version. So far it looks quite promising (2000 results, only 6 invalid).
I restarted an out of work machine to get the new 0.17 app to see if there is an improvement in validation for Linux. It's an i3-3240 CPU with an RX 570 GPU that had a good record with FGRPB1G. As of right now (running single tasks) it has completed and returned 25 tasks. The results are:-
Pending = 10
Valid = 10
Invalid = 1
Error = 0
Inconclusive = 4
In my experience, because Linux is the minor player, a big fraction of inconclusive results in a Windows/Linux comparison eventually translate to invalid for the Linux machine.
Of course it's early days and I intend to keep monitoring this but it looks pretty disheartening.
On a side issue, the lack of an 'Inconclusive' category at the top of website task listings is really annoying. It has been asked for quite a few times over the years (shoutout to Richard Haselgrove in particular) so I'm now adding my voice as well. There seems to be enough space at the top of each page to add an extra category and my guess (I'm not a programmer) is that it should be relatively simple to do.
It would be really useful as a quick and easy to monitor indicator of looming validation issues when new app versions are released.
Inconclusive is the “3” category. So just type that in the URL and it will show all the inconclusives. The site exists, just not linked from a hyperlink.
Inconclusive is the “3” category. So just type that in the URL and it will show all the inconclusives. The site exists, just not linked from a hyperlink.
Thanks for this obvious tip, so easy to spot. But how many times have I added up the valid, invalid, pending, ... ones and subtracted them from totals to check the inconclusives...
[EDIT:] But this site is still broken. There aren't 20 tasks listed per page (as configured within websites preferences) but one task on page 1, zero at page 2, two at p. 3, two at p. 6, ... four at p. 22, lots of empty pages until empty page 27.
Still difficult to count the number of inconclusives, harder even to check each one's details. That's maybe the reason why 'inconclusives' are hidden.
If you just look at a single result, you could work out your problem.
A result at random shows run time of 21,254 with a CPU time of just 44 secs. GPU tasks are being completely starved of CPU support.
Your host shows as 4 cores. Is that 4C/4T or 2C/4T? In either case (it doesn't really matter) it looks like you are running CPU tasks on all available threads. For 27th Sep, I counted 11 CPU tasks returned with that particular return date. The average time for a CPU task is ~34.5 ksecs. So a rough ball park calculation says that 11 tasks would consume ~380 ksecs. A full day is 86.4 ksecs so the number of active cores is 380/86.4 > 4 and that is probably the problem.
Just change your prefs for that machine to allow BOINC to use 75% of the CPUs rather than all of them and I'm guessing the GPU tasks will speed up enormously.
Inconclusive is the “3” category. So just type that in the URL and it will show all the inconclusives. The site exists, just not linked from a hyperlink.
Sure, but the host I'm monitoring has around 30 pages of results so a clickable link on the top line of any page which would assemble all inconclusives into a 20 per page list (just like you can for pendings or errors, etc.) would be nice to have.
Am I wrong in assuming that should be a fairly simple job to do?
Here's a list of all the Intel GPUs (as reported by BOINC) that had at least on successful BRP7 result:
Intel(R) HD Graphics 520
Intel(R) HD Graphics 530
Intel(R) HD Graphics 610
Intel(R) HD Graphics 620
Intel(R) HD Graphics 630
Intel(R) Iris(R) Plus Graphics 655
Intel(R) UHD Graphics
Intel(R) UHD Graphics 600
Intel(R) UHD Graphics 620
Intel(R) UHD Graphics 630
Any clue what these may have in common compared to the GPUs where the tasks failed?
Unfortunately the driver version isn't recorded in the DB.
they all seem to be from similar generation of devices. are they all on similar OSs? i assume these are all windows since you only released a windows app. if they are all on say Windows 7 or windows 10, it might point to those devices on an older driver version.
can you cross reference this list with the list of devices that produced errors? same devices or different?
hmm so mostly different devices it looks like between the lists. i think a driver difference could be possible.
still strange that all these devices work for BRP4, but not BRP7. you could cross check the code to see if there was some special stuff added to the intel GPU apps for BRP4
The main difference between BRP4 and BRP7 OpenCL code (which is really the same for all GPUs) is that BRP7 requires and uses fp64 / double precision in some kernels.
Bernd Machenschalk wrote: In
)
I restarted an out of work machine to get the new 0.17 app to see if there is an improvement in validation for Linux. It's an i3-3240 CPU with an RX 570 GPU that had a good record with FGRPB1G. As of right now (running single tasks) it has completed and returned 25 tasks. The results are:-
In my experience, because Linux is the minor player, a big fraction of inconclusive results in a Windows/Linux comparison eventually translate to invalid for the Linux machine.
Of course it's early days and I intend to keep monitoring this but it looks pretty disheartening.
On a side issue, the lack of an 'Inconclusive' category at the top of website task listings is really annoying. It has been asked for quite a few times over the years (shoutout to Richard Haselgrove in particular) so I'm now adding my voice as well. There seems to be enough space at the top of each page to add an extra category and my guess (I'm not a programmer) is that it should be relatively simple to do.
It would be really useful as a quick and easy to monitor indicator of looming validation issues when new app versions are released.
Cheers,
Gary.
Inconclusive is the “3”
)
Inconclusive is the “3” category. So just type that in the URL and it will show all the inconclusives. The site exists, just not linked from a hyperlink.
_________________________________________________________________________
Ian&Steve C.
)
Thanks for this obvious tip, so easy to spot. But how many times have I added up the valid, invalid, pending, ... ones and subtracted them from totals to check the inconclusives...
[EDIT:] But this site is still broken. There aren't 20 tasks listed per page
(as configured within websites preferences)but one task on page 1, zero at page 2, two at p. 3, two at p. 6, ... four at p. 22, lots of empty pages until empty page 27.Still difficult to count the number of inconclusives, harder even to check each one's details. That's maybe the reason why 'inconclusives' are hidden.
TRAPPIST-713
)
If you just look at a single result, you could work out your problem.
A result at random shows run time of 21,254 with a CPU time of just 44 secs. GPU tasks are being completely starved of CPU support.
Your host shows as 4 cores. Is that 4C/4T or 2C/4T? In either case (it doesn't really matter) it looks like you are running CPU tasks on all available threads. For 27th Sep, I counted 11 CPU tasks returned with that particular return date. The average time for a CPU task is ~34.5 ksecs. So a rough ball park calculation says that 11 tasks would consume ~380 ksecs. A full day is 86.4 ksecs so the number of active cores is 380/86.4 > 4 and that is probably the problem.
Just change your prefs for that machine to allow BOINC to use 75% of the CPUs rather than all of them and I'm guessing the GPU tasks will speed up enormously.
Cheers,
Gary.
Ian&Steve C.
)
Sure, but the host I'm monitoring has around 30 pages of results so a clickable link on the top line of any page which would assemble all inconclusives into a 20 per page list (just like you can for pendings or errors, etc.) would be nice to have.
Am I wrong in assuming that should be a fairly simple job to do?
Cheers,
Gary.
Here's a list of all the
)
Here's a list of all the Intel GPUs (as reported by BOINC) that had at least on successful BRP7 result:
Any clue what these may have in common compared to the GPUs where the tasks failed?
Unfortunately the driver version isn't recorded in the DB.
BM
Bernd Machenschalk
)
they all seem to be from similar generation of devices. are they all on similar OSs? i assume these are all windows since you only released a windows app. if they are all on say Windows 7 or windows 10, it might point to those devices on an older driver version.
can you cross reference this list with the list of devices that produced errors? same devices or different?
_________________________________________________________________________
Ian&Steve C. wrote:can you
)
Here's the list of "error" devices:
Intel(R) Arc(TM) A380 Graphics
Intel(R) Arc(TM) A750 Graphics
Intel(R) Arc(TM) A770 Graphics
Intel(R) HD Graphics
Intel(R) HD Graphics 2500
Intel(R) HD Graphics 4000
Intel(R) HD Graphics 4400
Intel(R) HD Graphics 4600
Intel(R) HD Graphics 6000
Intel(R) Iris(R) Xe Graphics
Intel(R) UHD Graphics
Intel(R) UHD Graphics 630
Intel(R) UHD Graphics 730
Intel(R) UHD Graphics 750
Intel(R) UHD Graphics 770
The 630 occurrs in both lists. But I can't easily filter by the type of error, this might be a different one (download or whatever).
BM
hmm so mostly different
)
hmm so mostly different devices it looks like between the lists. i think a driver difference could be possible.
still strange that all these devices work for BRP4, but not BRP7. you could cross check the code to see if there was some special stuff added to the intel GPU apps for BRP4
_________________________________________________________________________
The main difference between
)
The main difference between BRP4 and BRP7 OpenCL code (which is really the same for all GPUs) is that BRP7 requires and uses fp64 / double precision in some kernels.
BM