Help.
While I have been running GPU BOINC (almost entirely Einstein) for a couple of years, I've been doing so on machines with only one user (me). However two of my three machines are intended for family use with two users, and just yesterday I learned that all of the three share the failing that when a different user logs in, BOINC shuts down the GPU jobs.
The BOINC message log entry reads"
[pre]GPUs have become unusable; disabling tasks[/pre]
and when the BOINC-launching user logs back in the recovery message reads:
[pre]GPUs have become usable; enabling tasks[/pre]
This old thread over at SETI suggests that the problem is fundamental, given the way Windows handles graphics drivers on a user switch, and does not seem to offer a true two-user solution.
I know some folks reading here are GPU-crunchers on Windows 7. Do you just suffer the only when installing user is logged-in (or ctrl-L locked) but not another user logged in? Or is there a manner of installing BOINC or other fiddle which avoids this problem?
I can provide much more detail, but doubt most are relevant. All three hosts are Windows 7, the boinc versions are 7.0.25, 7.0.64, and 7.2.42.
All three seem to show the problem to some degree, although the 7.2.42 machine seemed to wait longer after the "wrong" account login to disable the GPU, and the message on that one was "Remote desktop in use: disabling GPU tasks" with the reciprocal message when I switched back to my account.
The different message in 7.4.42 seems just to be a text tweak by David, based on this boinc_cvs report
Lastly, I realize this is probably properly a BOINC issue, but I think myself among knowledgeable friends here, and imagine that with more Einstein participants considering lower barrier to entry GPU options, this issue might be of interest to others.
Copyright © 2024 Einstein@Home. All rights reserved.
GPU BOINC on Windows 7 with multiple users
)
Somebody called?
There are two problems here: one probably insurmountable, the other being actively worked on.
The active one first. Starting (I think) with Windows 7, the BOINC installation program fails to set up the users and groups security system properly so that all users of the computer can (if so permitted by the installing administrator) run BOINC Manager and generally control BOINC. I bought host 8864187 last September with the specific purpose of exploring the problem and testing solutions once provided.
It's taken six months (largely because the developers were buried under an Android Avalanche, I suspect), but in the last couple of weeks there have been a flurry of alpha builds in the v7.3.8/.9/.10 range: a little more to do, but the Manager problem is well on the way to being fixed.
The disappearing GPUs are the insurmountable problem. That's caused by a change in the device driver security model in Windows (starting with Vista). Before this change, it was possible for applications installed as system services (is that what's known as 'ring 0'?) to interact with device drivers.
Under the new model (called WDDM), only applications running in the user space of the logged-in user can interact with the graphics driver. If you log in and start BOINC under your account, the client will be running with your account credentials: if your other family member then uses "switch user", and becomes the foreground logged in user, BOINC no longer has access to the graphics driver.
I only know of two ways round this.
1) Instead of using 'switch user', do a full log out and log in again. This will close down BOINC, and restart it in the user space of the other family member.
2) Revert to Windows XP, and install BOINC v6.12.34 or earlier "as a service". That is capable of running GPUs even with nobody logged in to the computer.
RE: Instead of using
)
While slightly annoying, and requiring some family "training", this can work for us. When my "replace XP machines" campaign reaches close there will be three desktop PCs here. "His", exclusively, "Hers" almost always--save that I need to be able to do maintenance and have backup access, and "Scan/backup/alternate" which neither of us should normally stay on for dozens of hours a week.
Should I interpret your full post as meaning that the availability of this log/out--login solution awaits a hoped to be forthcoming BOINC build that fixes the Manager problem, then perhaps do a full uninstall and on installing the new boinc version select the option labelled something like "any user can control"? I think I've been avoiding that option because of previous bad behavior.
If a new boinc release combined with the login/logout method is pretty likely to give a long-term viable solution for our household within a few months, this issue is looking much less daunting than it did this morning. Thanks.
While XP has served me long and well, getting it out of my household within weeks is an express goal of my current round of build activity, so for my house the second option is not attractive.
Thank you very much, Richard. Both for this reply and for your active efforts to address this and other problems.
RE: Should I interpret your
)
Yes, pretty much.
I'd say that most of the bad behaviour has been eliminated already in the available (by alpha download) v7.3.10 - in fact, all the gross misbehaviour that you are likely to come across in daily use. There are bugfixes in more obscure areas pending (logging Manager errors - but client event log reporting is fine: opening a second copy of the Manager), but the basic functions are OK. I have no idea how long they expect v7.3 to remain in testing: given that we've only just had a flurry of v7.2 'recommended' (but buggy) releases, your guess is as good as mine.
Given the length of time [trac]#1025[/trac] has been given 'blocker' priority but no attention, I will be pressing for either a quick release of v7.4, or a backport to v7.2 - but don't hold your breath.
Installing over the top of an existing installation is no bother - you don't need an uninstall first. And the current setting of the 'Allow all users' is visible about three screens in: you just have to click the 'Advanced' button to reach an edit screen.
I did not know where to look
)
I did not know where to look for trial version downloads until mikey gave a link on another forum.
I have since installed 7.3.11 on one machine.
My initial experience was poor, suggesting that my personal configuration on this machine posed problems for the software:
1. On repeated trials, I could not log off from the installing account (which had been running BOINC jobs successfully on that boot each of these times).
2. when I booted straight to another account, BOINC started up and CPU jobs ran, but not GPU jobs.
3. when logging out of that second account and into the primary (and installing) account, BOINC ran CPU jobs, but showed something like "GPU not available" against the non-running GPU jobs.
I assumed the problems arose from non-standard elements of my configuration of which prominently there were:
1. I had deleted the registry entry BOINC installs to launch boincmgr, and instead launched it using an entry in a batch file triggered by user login.
2. also in that delayed launch batch file were launches for other applications, some of them there because I had observed or suspected bad behavior dependent on launch order and timing.
So, I disabled ALL the delayed launch entries, and uninstalled and re-installed boinc 7.3.11, this time leaving the registry "run" entry as built.
This greatly improved matters.
1. I was (usually) able to log out.
2. Booting to either user account gave me a running copy of boinc, including running GPU jobs.
3. on a (very) few trials, logging out of one user account and into the other gave me a fresh boinc start with all the same jobs running (including GPU).
I (very tentatively) identified Speedfan as my prime suspect for at least some of the blocked logout behavior. I have used and mostly liked it for a long time, but will do without for the time being.
I (tentatively) identify my deletion of the installed registry launch run command and substitution of a batch file line as related to the failure to see GPU work under both user accounts, though I have no idea why.
So at the moment I'm back to advising my wife it is good to try out the new machine, as it is very much like the new one that will arrive at her usual seat in a few weeks. I intend to do more testing, and if most seems well, to add back the monitoring and control programs I disabled en bloc.
Bottom line, for my situation the multi-user GPU capability as of version 7.3.11 works but may be more configuration-sensitive than 7.2.42 was.
RE: Bottom line, for my
)
Which could be helpful to Dr. A and his team of programmers if you can fill out an error log of what doesn't work for you:
"Please report test results using the BOINC Alpha Test web site:
http://isaac.ssl.berkeley.edu/alpha/"
Meaning the rest of us could benefit from your problems when they fix them in the release version.