I have used Geforce to install the nVidia driver, as I have always done. Since installing the Creators Update my 2014 PC, with A10-6700 CPU and 24 GB RAM is suffering a number of unwanted shutdowns, no evident reasons.
Tullio
Maybe WhoCrashed (free home edition) might be able to give some useful human-readable information of those events.
I've run all UEFI tests provided by HP on this Pavilion 152-ea PC and the hardware is simply perfect.If there is a problem. it is in the software.
Tullio
One of my older HP dual Xeon cpu machines, a Z600, had some major problems with some of the older Win10 updates and it kept trying to upgrade every day and then reverting back to the older version until I stopped it by dropping in a Win7 harddrive. Eventually after a couple of months I went back to Win10 and MS finally figured things out and it's happily crunching along now. It is not on the new Creators update though, I'm hoping MS can wait long enough to fix the problems so it's smooth this time!!
I had my second GPU-using machine go through the Windows Creators Update today. I again had Einstein trouble after the update. I paid more attention and took more notes this time. See below:
I blocked work download and suspended all Einstein tasks before beginning the update.
The Windows Update message said that an update was ready and a restart was needed, and inquired about scheduling the restart. While it mentioned version 1703 (which in fact is Creators), it did not mention Creators by name. Nor did it mention that rather than a mere restart to tidy up an already complete install, this would be a lengthy process involving at least three restarts, and on my machine very nearly an hour.
Previously I paid attention to the version number on the uninstaller shown in Windows Control Panel, Programs and Features, and as that was unchanged, asserted that the Windows Update had obeyed the status flag I had tried to set with DDU not to update my Nvidia driver. This time it was again true that the version number shown on the uninstaller was unchanged by the Windows Update. However the version number shown in Windows Control Panel|Device Manager WAS changed. I've changed my position, and now believe that the update did indeed install a different driver.
As before, BOINCMgr displayed on the individual task status lines a "GPU missing" status. Looking a little more carefully, the Event Log for the startup of BOINCMgr made it clear that it detected both the 1050 and the 1060, but thought that OpenCL support required was not present. These exact lines appeared in my event log:
6/27/2017 5:04:37 PM | | CUDA: NVIDIA GPU 0: GeForce GTX 1060 3GB (driver version 382.05, CUDA version 8.0, compute capability 6.1, 3072MB, 2513MB available, 3936 GFLOPS peak)
6/27/2017 5:04:37 PM | | CUDA: NVIDIA GPU 1: GeForce GTX 1050 (driver version 382.05, CUDA version 8.0, compute capability 6.1, 2048MB, 1672MB available, 1911 GFLOPS peak)
6/27/2017 5:04:38 PM | | App version needs OpenCL but GPU doesn't support it
6/27/2017 5:04:38 PM | Einstein@Home | Application uses missing NVIDIA GPU
The successful repair sequence I followed was this:
I downloaded the current nvidia driver for my configuration
I downloaded the current version of DDU and unpacked it
I ran the (wrong version number) Nvidia uninstaller shown in Control Panel|Programs and Features. I declined the restart offered at the end of the unstaller run.
I ran msconfig and specified under the boot tab a minimal safe boot
On reboot in safe mode I ran DDU, selecting the first option (recommended when no hardware is changed)
On reboot from DDU into regular mode Windows I ran the NVIDIA installer, selecting only the graphics driver (no optional extra components) and enabling the "clean install" option
On reboot after the nvidia installer ran, all seemed well. BOINCMgr no longer displayed the missing GPU indication. I released the suspends on the two previously active tasks, and when they seemed to work properly released all the queue tasks as well.
I won't pretend to know which parts of my procedure are needed, and which are superfluous. But this is the second Windows Creators update for which it has worked.
All Einstein GPU tasks still fail on my Window 10 PC while SETI and SETI Beta run flawlessly on the GTX 1050 Ti with nVidia driver 382.53. All Einstein GPU tasks run on my Linux box with GTX 750 Ti and driver 375.66 provided by OpenSuSE which constantly updates the kernel.
I had my second GPU-using machine go through the Windows Creators Update today. I again had Einstein trouble after the update. I paid more attention and took more notes this time. See below:
....
Thank you very much for doing this. Hopefully it will help any Win10 users that have trouble after that particular update. I don't use Windows at all so have no knowledge of the steps you have documented. The explanation is very clear so hopefully any affected volunteers can get the problem sorted by following the steps given.
As a general comment about the true implication of the event log message,
Quote:
6/27/2017 5:04:37 PM | | CUDA: NVIDIA GPU 1: GeForce GTX 1050 (driver version 382.05, CUDA version 8.0, compute capability 6.1, 2048MB, 1672MB available, 1911 GFLOPS peak)
6/27/2017 5:04:38 PM | | App version needs OpenCL but GPU doesn't support it
the first entry (split over two lines) informs that the GPU (and its CUDA capability) are properly detected, so the GPU is not at all "missing" as the status on the tasks tab of BOINC Manager implies.
The second line also gives an incorrect 'impression' to the user. It implies that the hardware doesn't support OpenCL. The true situation is that the driver component that allows the OpenCL capability to be accessed is missing (or incorrectly installed) - ie. its a software and not a hardware problem. So the situation would appear to be that the Windows update is not installing (or incorrectly installing) the library/ies (.DLL(s)) that give access to the OpenCL hardware capabilities.
I see exactly the same message on Linux if I forget to do a simple post-driver-install fixup. In my case, the OpenCL library is installed in a place where BOINC doesn't see it. I just create a symbolic link from where BOINC will look to where the library is actually installed. That way there is never a problem. I don't use a distro version of BOINC - there is none in my distro's repository. Distros that do have BOINC in the repo would probably have tweaked things so that the library is always found.
All Einstein GPU tasks still fail on my Window 10 PC while SETI and SETI Beta run flawlessly on the GTX 1050 Ti with nVidia driver 382.53.
Hi Tullio,
Your tasks are starting and then having a computation error while crunching so that's really nothing to do with a "missing GPU" problem after a particular Win10 update.
If you click on the task ID link for one of your failed tasks you can see the error messages generated. It turns out to be exactly the same messages as reported in this particular thread earlier this year. You should probably read through that thread to see what the ideas/suggestions were about the causes of the problem. Unfortunately, the discussion stopped abruptly without any real confirmation of the ideas being tested.
In the opening message of that thread there is a reference to an even earlier example of the problem together with a comment by Christian (presumably from that earlier example) that Bernd might need to have a look at it when he returned from holidays. I don't remember seeing any further discussion so perhaps Christian might be able to give an update when he reads about your current situation.
Thanks Gary. Since I am not a Windows expert but a Linux user I have dedicated the Windows 10 PC which reboots constantly after the 15 June Creators Update to SETI and SETI Beta. My old (2008) SUN workstation with SuSE Leap 42.2 is crunching a command line BOINC on both CPU and GPU Einstein tasks with good results.
tullio wrote:I have used
)
Maybe WhoCrashed (free home edition) might be able to give some useful human-readable information of those events.
http://www.resplendence.com/download/whocrashedSetup.exe
I've run all UEFI tests
)
I've run all UEFI tests provided by HP on this Pavilion 152-ea PC and the hardware is simply perfect.If there is a problem. it is in the software.
Tullio
tullio wrote:I've run all
)
One of my older HP dual Xeon cpu machines, a Z600, had some major problems with some of the older Win10 updates and it kept trying to upgrade every day and then reverting back to the older version until I stopped it by dropping in a Win7 harddrive. Eventually after a couple of months I went back to Win10 and MS finally figured things out and it's happily crunching along now. It is not on the new Creators update though, I'm hoping MS can wait long enough to fix the problems so it's smooth this time!!
I had my second GPU-using
)
I had my second GPU-using machine go through the Windows Creators Update today. I again had Einstein trouble after the update. I paid more attention and took more notes this time. See below:
I blocked work download and suspended all Einstein tasks before beginning the update.
The Windows Update message said that an update was ready and a restart was needed, and inquired about scheduling the restart. While it mentioned version 1703 (which in fact is Creators), it did not mention Creators by name. Nor did it mention that rather than a mere restart to tidy up an already complete install, this would be a lengthy process involving at least three restarts, and on my machine very nearly an hour.
Previously I paid attention to the version number on the uninstaller shown in Windows Control Panel, Programs and Features, and as that was unchanged, asserted that the Windows Update had obeyed the status flag I had tried to set with DDU not to update my Nvidia driver. This time it was again true that the version number shown on the uninstaller was unchanged by the Windows Update. However the version number shown in Windows Control Panel|Device Manager WAS changed. I've changed my position, and now believe that the update did indeed install a different driver.
As before, BOINCMgr displayed on the individual task status lines a "GPU missing" status. Looking a little more carefully, the Event Log for the startup of BOINCMgr made it clear that it detected both the 1050 and the 1060, but thought that OpenCL support required was not present. These exact lines appeared in my event log:
6/27/2017 5:04:37 PM | | CUDA: NVIDIA GPU 0: GeForce GTX 1060 3GB (driver version 382.05, CUDA version 8.0, compute capability 6.1, 3072MB, 2513MB available, 3936 GFLOPS peak)
6/27/2017 5:04:37 PM | | CUDA: NVIDIA GPU 1: GeForce GTX 1050 (driver version 382.05, CUDA version 8.0, compute capability 6.1, 2048MB, 1672MB available, 1911 GFLOPS peak)
6/27/2017 5:04:38 PM | | App version needs OpenCL but GPU doesn't support it
6/27/2017 5:04:38 PM | Einstein@Home | Application uses missing NVIDIA GPU
The successful repair sequence I followed was this:
I downloaded the current nvidia driver for my configuration
I downloaded the current version of DDU and unpacked it
I ran the (wrong version number) Nvidia uninstaller shown in Control Panel|Programs and Features. I declined the restart offered at the end of the unstaller run.
I ran msconfig and specified under the boot tab a minimal safe boot
On reboot in safe mode I ran DDU, selecting the first option (recommended when no hardware is changed)
On reboot from DDU into regular mode Windows I ran the NVIDIA installer, selecting only the graphics driver (no optional extra components) and enabling the "clean install" option
On reboot after the nvidia installer ran, all seemed well. BOINCMgr no longer displayed the missing GPU indication. I released the suspends on the two previously active tasks, and when they seemed to work properly released all the queue tasks as well.
I won't pretend to know which parts of my procedure are needed, and which are superfluous. But this is the second Windows Creators update for which it has worked.
I say you should voice your
)
I say you should voice your concerns to M$ about their incompetence, they can't get better if they don't know.
All Einstein GPU tasks still
)
All Einstein GPU tasks still fail on my Window 10 PC while SETI and SETI Beta run flawlessly on the GTX 1050 Ti with nVidia driver 382.53. All Einstein GPU tasks run on my Linux box with GTX 750 Ti and driver 375.66 provided by OpenSuSE which constantly updates the kernel.
Tullio
archae86 wrote:I had my
)
Thank you very much for doing this. Hopefully it will help any Win10 users that have trouble after that particular update. I don't use Windows at all so have no knowledge of the steps you have documented. The explanation is very clear so hopefully any affected volunteers can get the problem sorted by following the steps given.
As a general comment about the true implication of the event log message,
the first entry (split over two lines) informs that the GPU (and its CUDA capability) are properly detected, so the GPU is not at all "missing" as the status on the tasks tab of BOINC Manager implies.
The second line also gives an incorrect 'impression' to the user. It implies that the hardware doesn't support OpenCL. The true situation is that the driver component that allows the OpenCL capability to be accessed is missing (or incorrectly installed) - ie. its a software and not a hardware problem. So the situation would appear to be that the Windows update is not installing (or incorrectly installing) the library/ies (.DLL(s)) that give access to the OpenCL hardware capabilities.
I see exactly the same message on Linux if I forget to do a simple post-driver-install fixup. In my case, the OpenCL library is installed in a place where BOINC doesn't see it. I just create a symbolic link from where BOINC will look to where the library is actually installed. That way there is never a problem. I don't use a distro version of BOINC - there is none in my distro's repository. Distros that do have BOINC in the repo would probably have tweaked things so that the library is always found.
Cheers,
Gary.
tullio wrote:All Einstein GPU
)
Hi Tullio,
Your tasks are starting and then having a computation error while crunching so that's really nothing to do with a "missing GPU" problem after a particular Win10 update.
If you click on the task ID link for one of your failed tasks you can see the error messages generated. It turns out to be exactly the same messages as reported in this particular thread earlier this year. You should probably read through that thread to see what the ideas/suggestions were about the causes of the problem. Unfortunately, the discussion stopped abruptly without any real confirmation of the ideas being tested.
In the opening message of that thread there is a reference to an even earlier example of the problem together with a comment by Christian (presumably from that earlier example) that Bernd might need to have a look at it when he returned from holidays. I don't remember seeing any further discussion so perhaps Christian might be able to give an update when he reads about your current situation.
Cheers,
Gary.
Thanks Gary. Since I am not a
)
Thanks Gary. Since I am not a Windows expert but a Linux user I have dedicated the Windows 10 PC which reboots constantly after the 15 June Creators Update to SETI and SETI Beta. My old (2008) SUN workstation with SuSE Leap 42.2 is crunching a command line BOINC on both CPU and GPU Einstein tasks with good results.
Tullio