I've had a bad experience updating one of my three active two-GPU machines to a major Windows 10 version update (the kind that makes you watch at least two counting downs of progress with cheerful initial comments along the lines of "this might take a while").
The machine is this one. The Windows Version is reported on the einstein host list as Microsoft Windows 10 Core x64 Edition, (10.00.16299.00).
After I ran the Windows Update, Einstein did not see a GPU resource it could use. We have seen this before, and installing the current Nvidia driver directly downloaded from Nvidia, possibly with extra secret sauce in the form of DDU use and specification of "clean install" for the Nvidia driver has fixed it for multiple users in the past, including me.
My screen came up after the windows 10 update in a low resolution mode (the sort I am used to seeing in a Safe Mode boot).
I did not find any Nvidia Uninstaller in Control Panel.
I ran DDU.
On reboot I ran the current Nvidia installer.
On reboot after running the Nvidia installer I was troubled to see my monitor still in a low resolution mode. I right-clicked in the desktop and changed the monitor to the hardware resolution, which worked.
BOINC was only seeing one of my two installed graphics cards, the 750 Ti into which my monitor is plugged.
In an attempt to resolve that, I uninstalled the Nvidia driver, ran DDU from a safe reboot in "install new hardware" mode, and hoped that after the subsequent full power down that installing the Nvidia driver would give me a system BOINC could see both cards in.
Not only does BOINC not see both cards, but neither does MSI Afterburner, which reports my 750 Ti as GPU1, and lists in the pick list a GPU2 with no identifier.
I've started this thread here in Cruncher's Corner rather than Problems and Bug Reports, as I currently consider myself to have a me/Microsoft/Nvidia problem, not a BOINC/Einstein problem.
I also think it would be useful to the community for other people report their 10.00.16299.00 experiences here, both good and bad.
Somewhere I recall in the past people have advocated using dummy plugs on second graphics cards to trick matters so the card is recognized and supported for use. I've never used that trick, and have been running three systems with dual dis-similar cards for a couple of years. Perhaps I need to learn and use that trick now.
I have not always plugged my monitor into the card in the physically highest PCI slot. So far it had not been a problem. Until I look I don't know if it might be an issue here.
The two graphics cards in this case are a 750Ti and a 970, so materially older than the cards in the other two systems, which are 1050/1060/1070 cards. But those systems have not seen this update yet.
Copyright © 2024 Einstein@Home. All rights reserved.
archae86 wrote:Somewhere I
)
To see if that technique would have any influence on the problem, why don't you simulate a dummy plug with the actual monitor from one of your other machines to see if it makes any difference? If not then you know you can stop thinking about a dummy plug :-).
Cheers,
Gary.
I had only one machine to
)
I had only one machine to update and it went fine. Running a single AMD R9 390 on it. Fall Creators Update kept the Radeon 17.10.1 driver and I didn't even reinstall it. Boinc and Einstein seems to accept it.
Archae86, An interesting problem you got there with those Nvidia cards. I thought maybe while waiting for other ideas to test you could try installing another driver version, from a different driver branch: GeForce Vulkan Developer Beta 383.18
Gary and Richie, Thanks for
)
Gary and Richie,
Thanks for the suggestions. I actually had the "plug in another monitor" idea myself, and with your suggesting it, intend to have a try tomorrow (I happen have an older spare monitor around, so won't have to disturb a machine, but will have to pull the affected case out from the furniture).
Richie, the machine is question is my wife's primary use machine, so I'd like reasonable hope that the Vulkan Developer Beta is not likely to cause harm, and is likely to be reversible.
When I made my post I had no idea how new Fall Creators Update actually was, but a few searches suggest it is less than 24 hours old in general public distribution. Quite possibly my main need is to wait a bit for Nvidia to get told that others have problems, and to fix them in a driver revision. While I assume Nvidia had a version to work against in house some time back, there is nothing like exposing code to tens of thousands of users to show what has been missed.
archae86 wrote:Quite possibly
)
I believe that might well be the case. Systems with multiple GPU's isn't the largest group...
About the Vulkan driver, I can't say other than I've used that version for a single GTX 960 during the recent builds (lastly with build 17017). It worked fine with Boinc, but I haven't really used that computer for 'daily' normal purposes, so I can't guarantee that driver couldn't introduce anything weird. But I haven't encountered any problems with uninstalling any Vulkan drivers ever.
Well... This is a bit offtopic, but I quickly tested how build 10.00.17017.00 would start up with two GPU's. This machine originally had a single GT 710 (installed in 'third' 8x pcie slot, closest to the side). I took the card off and moved it onto the first slot (closest to CPU). I also connected the monitor cable (VGA) on this card. Then I add another GT 710 64bit 1GB on the second slot. GPU's have identical specs but are different brand.
I started computer and Windows didn't yet proceed to login screen. It rebooted computer, but on the second run it went to login screen and let me login. Then it survived to desktop with low resolution.
I looked at the Device Manager. At this point it didn't list any Nvidia GPU's, but there were two lines of unidentified HDMI Audio hardware or something like that
I installed Nvidia driver 387.98 (every component and 'clean install'). Then I reboot the computer. Windows proceeded to login screen and started fine, now with high resolution. Device Manager said there were two GT 710's. Also GPU-Z displays them correctly. Boinc does find both cards.
This machine has Boinc 7.8.3 , but that shouldn't of course change how Windows will identify GPU's.
I got the Creators Fall
)
I got the Creators Fall update and had to reload the nVidia driver for my GTX 1050 Ti. My CPU, an AMD A10-6700 went to 4.18 GHz while previously it reached 4.14 GHz. I can now see the GPU on the Task Manager which I newer saw before. I am running LHC@home with VirtualBox 5.1.28 and SETI@home GPU tasks. I am running both SETI@home and Einstein@home both CPU and GPU tasks on my main Linux box with SuSE Leap 42.2.
Tullio
Note, that the GPU usage
)
Note, that the GPU usage reported in RS3 is just the Direct3D layer, so it won't cover computing, etc.
For example it reports my RX Vega usage as 2-3% which actually it runs at ~100% (OpenCL).
-----
I must considerably revise my
)
I must considerably revise my description of the details of my problem, but after some invasion of the hardware, and multiple software installations and tests, the basic fact remains that on my offending system, BOINC only says it sees one GPU, and only runs Einstein work on that one GPU. Oddly it is using the less capable GPU, which is not installed in the top PCIe slot, and is not the GPU connected to the system monitor.
I did try the Vulkan Beta developer's driver, and saw no difference in behavior, so I have reverted to the most recent production Nvidia driver release.
HOWEVER, I was over-generalizing in claiming that the two GPU cards were not seen by Windows, Nvidia or both.
In fact all of HWINFO, GPU-Z, and the Windows Control Panel Device Manager Display Adapter list both cards I currently have installed. As of now (not yesterday) that is an GTX 1050 in the top PCIe slot, which is driving the system monitor, and which is not noticed by BOINC, and not used for Einstein work, and a 750 Ti which is in a lower PCIe slot, and is not currently connected to anything.
During my fiddling around, I did boot the system with separate monitors attached to these two cards, and pixels were indeed sent to both of them. So it seems the system including Windows has much awareness that they both exist.
I updated the BOINCmgr to the current recommended version, without change in my situation. I checked that my cc_config.xml had the line to instruct that all GPUs be used (it already did).
The reason for swapping in the 1050 for the 970 is that I had always planned that as the final configuration for this machine, and while I was fiddling around I preferred to get to that final configuration sooner.
I still have hope that a new release of something, or new wisdom on something to try, will get both cards doing something useful. If not, I should pull out the 750Ti, as I am getting less BOINC work done and consuming more power than I think would the the case were only the 1050 in the box. I'm quite afraid that when the Windows Fall Creators Update is installed on my other two GPU systems it will ruin their ability to use both cards, also.
The only idea I have at the moment is to install up-to-date versions of Afterburner and Nvidia Inspector, which might perhaps remove a monitoring anomaly, but are unlikely to correct my inability to have BOINC see both cards and use them.
I'm not much of a BOINC forum participant. Do you think I should attempt to start a thread there on this problem?
archae86 wrote:As of now (not
)
I know you will have already scrutinised these multiple times, but perhaps if you just post a copy of the startup messages where BOINC lists exactly what it detects, there might be some clue that somebody might recognise.
It can't hurt. The wider the audience, the better. Perhaps you might find someone else with a similar experience after applying this update that is running one of the other GPU using projects and so wouldn't know about your situation. Worth a try.
Cheers,
Gary.
Gary, I've created a thread
)
Gary, I've created a thread on the BOINC user forum "Questions and Problems"
Here is the beginning of the event log for the system in question in the current (undesirable) state:
OK, thanks for posting the
)
OK, thanks for posting the messages. I've perused what you posted at boinc.berkeley.edu - looks good. Hopefully you might get some responses.
From the startup messages, BOINC is seeing the lesser GPU as GPU 0 which seems to confirm the problem is with BOINC's detection mechanism, seeing as everything else is seeing both of them. The use of the 'use all coprocessors' option is also confirmed. I don't really see anything else unusual, not that I would consider myself particularly well informed in these matters.
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? It may be a long shot but it's very tempting to postulate that an apparent change in BOINC behaviour may have come from the change in version in combination with the Windows update. Did you have successful 2 GPU operation with 7.8.3 before the Windows update?
In any case, and particularly if there is any doubt about what was working and when things changed, I'd be tempted to go back to the last stable 7.6 version of BOINC and see if that makes any difference. From what I've been told, lots of things have changed between 7.6.x and 7.8.x and that there may not yet have been enough testing to know there aren't still problems lurking.
I shall continue to follow your adventures with interest! Sorry I don't know anything about Windows and can't be of much use :-).
Cheers,
Gary.