I note you are using 7.8.3 which is a very new version and perhaps has new bugs. Did you upgrade from 7.8.2 or from the last stable 7.6 version?
I did the upgrade to 7.8.3 some hours after the beginning of my trouble after the Windows 10 update. Sadly I did not log which particular BOINC version I was on before, but it was months back, and suffered the same trouble. "same" here may be an over generalization of what I have actually observed, but for sure BOINC only mentioned noticing one GPU, and only used one, both in the considerably older version I had been running, and the shiny new one.
... but it was months back, and suffered the same trouble. "same" here may be an over generalization of what I have actually observed, but for sure BOINC only mentioned noticing one GPU, and only used one, both in the considerably older version I had been running, and the shiny new one.
OK, that's very interesting. The previous 'stable/recommended' version was 7.6.33 which was initially released and 'blessed' with that title back on June 6, 2016. It remained the recommended version until relatively recently (I believe but I don't follow this all that closely). The next change was the release of a testing version 7.7.2 on March 29, 2017. I doubt you would have been tempted to install this, but I guess you might have based on the "months back" comment. It's more likely you may have gone for one of the 7.8.x versions, perhaps. The relevant dates were (and I've included all the versions for ease of reviewing the timeline):-
June 6, 2016 - 7.6.33
March 29, 2017 - 7.7.2
August 10, 2017 - 7.8.0
August 12, 2017 - 7.8.1
August 29, 2017 - 7.8.2
October 4, 2017 - 7.8.3
If you look at how closely the transition from .0 to .1 to .2 in the BOINC versions happened, you get the sense that there was pressure to put something out (after such a long time back to 7.6.33) and what came out had lots of bugs, perhaps of the showstopper variety requiring rushed fixes. I really think that there's still more fixing to be done and that this may be having an impact on you.
If the problem with GPU detection exists in both 7.8.3 and the version that came before it, "months ago", I would be very keen to go back to last year's 7.6.33 and see if that solves the problem. I reckon it's worth considering, even if only to be able to say that it really must be a Windows problem without any likely recent BOINC bug involvement.
You're the past master at designing rigorous experiments, changing only one variable at a time - right up your alley :-).
I'm pretty sure I read somewhere that several new releases after 7.8.2 were expected in the near future due to known problems with the preview releases for upcoming versions of Windows and Mac OS. The 7.8.3 changelog appears to be focused on fixing Mac problems. What you've ran into might be on the list of known FCU issues, you should probably check in somewhere that the boinc client developers read to make sure this is something they're aware of.
Good catch! I seem to remember in the bit of reading about BOINC versions I did yesterday that 7.6.22 was a version that had some sort of issue. I could be wrong but perhaps it was a good version to move away from anyway :-).
... known problems with the preview releases for upcoming versions of Windows ...
That sounds a bit ominous - seems to imply that all previous versions of BOINC might have problems with the latest Windows stuff. I guess it might take a while to get all problems found and fixed and a stable/workable version of BOINC released, given the way that BOINC is being developed these days.
My interaction on the BOINC forum caused me to learn about CLINFO. That dumps to your screen information available to CUDA-using things. For my troubled system, that just showed one card--the one BOINC has been using. Advice there was that sounded like a driver problem.
So I tried backing down the the latest Nvidia driver that is running OK on one my my two-GPU systems, which is, however not one Nvidia advocates that Creators Fall Update system use. It seemed to function adequately for my other purposes, but gave the same single-card CUDA issue as seen by both BOINC startup messages and CLINFO. So if this in part a driver problem, possibly the driver behavior is somehow affected by this OS change.
I tried the very simple back-down of removing the 750Ti (the less capable card in the secondary slot that was the only one CLINFO reported and BOINC used.
Promptly the 1050 which remained untouched in the primary slot was visible to BOINC and CLINFO, and used successfully by BOINC.
Tomorrow I intend to try a couple of more things with two cards in the system, and if adding back the 750Ti without doing driver removal does not work, I'll try running two identical 750Ti cards, on the off chance that the cards being different in one way or another is a component of the problem.
If all multi-GPU systems on Windows 10 have my problem, I'd expect some of our heavy contributors to have trouble pretty soon, and perhaps they'll find this thread and post their experience. For the short term, I can check the top 50 Einstein host list. It currently has 17 Windows 10 machines, of which 7 run multiple Nvidia cards. All seven currently show Windows flavor 10.00.15063.00, whereas the Fall Creators Update machines I know about so far show 10.00.16299.00. More likely someone farther down the list will try it long before those three or four people, but I won't know how to spot them.
Partly in response to advice on the BOINC forum, I've flashed the motherboard UEFI BIOS, run the chipset INF updater, and tried some other configurations, including the one at this moment which has two same-model 750Ti cards.
No joy. In all cases tried to date in which two cards are in the machine, CLINFO and BOINC only report one, and BOINC only uses one.
The UEFI system browser, HWiNFO, and GPU-Z all see both cards when two are installed. MSIAfterburner asserts there are two, but is missing information on one of the two. Clinfo and boinc both seem to think there is just one.
I'm out of ideas for the moment.
I note that user Sikabonyi appears to have updated his top-50 dual GTX 1080 machine 12317842 to the Fall Creators Update. As the detail listing reports it has two GPUs, I have hope that system did not share my evil fate, which gives some hope for my two unconverted two-GPU systems. But I don't know when it was converted, nor really whether it is running well after conversion.
Gary Roberts wrote:I note you
)
I did the upgrade to 7.8.3 some hours after the beginning of my trouble after the Windows 10 update. Sadly I did not log which particular BOINC version I was on before, but it was months back, and suffered the same trouble. "same" here may be an over generalization of what I have actually observed, but for sure BOINC only mentioned noticing one GPU, and only used one, both in the considerably older version I had been running, and the shiny new one.
archae86 wrote:... but it was
)
OK, that's very interesting. The previous 'stable/recommended' version was 7.6.33 which was initially released and 'blessed' with that title back on June 6, 2016. It remained the recommended version until relatively recently (I believe but I don't follow this all that closely). The next change was the release of a testing version 7.7.2 on March 29, 2017. I doubt you would have been tempted to install this, but I guess you might have based on the "months back" comment. It's more likely you may have gone for one of the 7.8.x versions, perhaps. The relevant dates were (and I've included all the versions for ease of reviewing the timeline):-
If you look at how closely the transition from .0 to .1 to .2 in the BOINC versions happened, you get the sense that there was pressure to put something out (after such a long time back to 7.6.33) and what came out had lots of bugs, perhaps of the showstopper variety requiring rushed fixes. I really think that there's still more fixing to be done and that this may be having an impact on you.
If the problem with GPU detection exists in both 7.8.3 and the version that came before it, "months ago", I would be very keen to go back to last year's 7.6.33 and see if that solves the problem. I reckon it's worth considering, even if only to be able to say that it really must be a Windows problem without any likely recent BOINC bug involvement.
You're the past master at designing rigorous experiments, changing only one variable at a time - right up your alley :-).
Cheers,
Gary.
archae86 wrote:Sadly I did
)
I believe you had that host running 7.6.22. That's what the completed tasks from a couple of days back show:
https://einsteinathome.org/task/688524538
I'm pretty sure I read
)
I'm pretty sure I read somewhere that several new releases after 7.8.2 were expected in the near future due to known problems with the preview releases for upcoming versions of Windows and Mac OS. The 7.8.3 changelog appears to be focused on fixing Mac problems. What you've ran into might be on the list of known FCU issues, you should probably check in somewhere that the boinc client developers read to make sure this is something they're aware of.
Richie_9 wrote:I believe you
)
Good catch! I seem to remember in the bit of reading about BOINC versions I did yesterday that 7.6.22 was a version that had some sort of issue. I could be wrong but perhaps it was a good version to move away from anyway :-).
Cheers,
Gary.
DanNeely wrote:... known
)
That sounds a bit ominous - seems to imply that all previous versions of BOINC might have problems with the latest Windows stuff. I guess it might take a while to get all problems found and fixed and a stable/workable version of BOINC released, given the way that BOINC is being developed these days.
Cheers,
Gary.
My interaction on the BOINC
)
My interaction on the BOINC forum caused me to learn about CLINFO. That dumps to your screen information available to CUDA-using things. For my troubled system, that just showed one card--the one BOINC has been using. Advice there was that sounded like a driver problem.
So I tried backing down the the latest Nvidia driver that is running OK on one my my two-GPU systems, which is, however not one Nvidia advocates that Creators Fall Update system use. It seemed to function adequately for my other purposes, but gave the same single-card CUDA issue as seen by both BOINC startup messages and CLINFO. So if this in part a driver problem, possibly the driver behavior is somehow affected by this OS change.
I tried the very simple back-down of removing the 750Ti (the less capable card in the secondary slot that was the only one CLINFO reported and BOINC used.
Promptly the 1050 which remained untouched in the primary slot was visible to BOINC and CLINFO, and used successfully by BOINC.
Tomorrow I intend to try a couple of more things with two cards in the system, and if adding back the 750Ti without doing driver removal does not work, I'll try running two identical 750Ti cards, on the off chance that the cards being different in one way or another is a component of the problem.
If all multi-GPU systems on Windows 10 have my problem, I'd expect some of our heavy contributors to have trouble pretty soon, and perhaps they'll find this thread and post their experience. For the short term, I can check the top 50 Einstein host list. It currently has 17 Windows 10 machines, of which 7 run multiple Nvidia cards. All seven currently show Windows flavor 10.00.15063.00, whereas the Fall Creators Update machines I know about so far show 10.00.16299.00. More likely someone farther down the list will try it long before those three or four people, but I won't know how to spot them.
Partly in response to advice
)
Partly in response to advice on the BOINC forum, I've flashed the motherboard UEFI BIOS, run the chipset INF updater, and tried some other configurations, including the one at this moment which has two same-model 750Ti cards.
No joy. In all cases tried to date in which two cards are in the machine, CLINFO and BOINC only report one, and BOINC only uses one.
The UEFI system browser, HWiNFO, and GPU-Z all see both cards when two are installed. MSIAfterburner asserts there are two, but is missing information on one of the two. Clinfo and boinc both seem to think there is just one.
I'm out of ideas for the moment.
I note that user Sikabonyi appears to have updated his top-50 dual GTX 1080 machine 12317842 to the Fall Creators Update. As the detail listing reports it has two GPUs, I have hope that system did not share my evil fate, which gives some hope for my two unconverted two-GPU systems. But I don't know when it was converted, nor really whether it is running well after conversion.
I believe this is a gfx
)
I believe this is a gfx driver problem.
-----
There is some reports about
)
There are some reports about opencl.dll as supplied by Microsoft as being at least part of the problem.
There are some workarounds to try on this thread here
if that fails, there is always the nuclear penguin option...