The following driver sets are bugged for me:
- 295.73 WHQL
- 295.51 Beta

The last driver set that worked for me was:
- 290.53 Beta

For me, the bug affects all 3 of my GPU projects, most of the time making tasks error out immediately:
- GPUGRID.net
- Einstein@Home
- SETI@Home

There are some reports of bugs concerning the latest NVIDIA drivers (failures when monitor goes to sleep). GPUGRID may be immune to the bug, but if it occurs to you, rollback to previous drivers.

Does anyone know if nVIDIA is aware of/working on this issue?
____________

nVidia has been informed, but there has been no response.

Claggy reported the issue formally to nVidia, on 2/1/2012, using Ref 120201-000013, as posted here:
http://forums.nvidia.com/index.php?showtopic=223426&view=findpost&p=1374585

I reported the issue in their 295.51 Beta drivers thread, on 2/16/2012, here:
http://forums.nvidia.com/index.php?showtopic=221985&view=findpost&p=1370579

I also reported the issue in their 295.73 WHQL drivers thread, on 2/24/2012, here:
http://forums.nvidia.com/index.php?showtopic=223426&view=findpost&p=1374645

We have not heard any response as of yet.
If you know of a more appropriate way to inform them of the problem, or get them to fix it, you're welcome to try it.

The message threads over at SETI seem to indicate its the windows driver that has the issue. It has been reported by people using a DVI connected monitor, not sure if a VGA connected monitor also has the problems. It depends on the card and if they are using a DVI to VGA adaptor.

I’ve rolled back to previous drivers thanks. 3 days of all error on milkyway, SETI, GPUGRID and Einstein. What mess!
Get crunchin!
BDDave

____________

I thought I'd chime in with some more information.

If you want to use 295.73 WHQL or 295.51 Beta without CUDA failures:
The workaround for the bug is that you must set the Windows Power Options to "Turn off the display: Never". You may still use a screen saver, and you may still physically turn the monitor off, but you must not let the software power the monitor down... according to my testing.

Also, for anyone trying to reproduce the problem, I have found that the problem occurs when Windows powers off the monitor first, and then BOINC tries to start or resume a CUDA task while the monitor is off. This means that, if you try to reproduce it using tasks that are already running before Windows powers down the monitor, those tasks will not fail. But any tasks that try to start or resume, while the monitor is off, will fail... according to my testing.

Finally, the best news yet, I have been contacted privately by an nVidia employee, who was having trouble recreating the problem. I assisted him, and he can now repro on demand now (it's easiest to repro with Einstein@Home, and he didn't know that the monitor has to power down before BOINC begins CUDA processing), and he will be presenting information to the developers.

I am now going to run 295.73 WHQL with a "Blank" screensaver and "Turn off the display: Never", and try to remember to physically turn the monitor off if I get up for an extended period of time.

Regards,
Jacob Klein

I did a rollback to the 285.62 driver and still no work, what do i do now?
____________

What does "still no work" mean?

Did not see anything CUDA-related in the changelog.

The 266.58 are the last drivers that seem to be problem-free, no downclocking bug and obviously no sleep mode bug. AFAIK they support everything up through the GTX 580. Unless you have a game or other software that requires the newer drivers, I would suggest rolling back to those. You will have to do a clean install though and be absolutely sure that no Nvidia software remains on your system before installing them. Otherwise certain core files will remain and you might still get the same issues.

266.58 doesn't work well on Ubuntu with Albert&Einstein and DistrRTgen tasks.

1 am running 296.10 NVidia (WHQL) drivers. Screen saver is set never to turn monitor off or "sleep" system. GPUGRID tasks yield computation errors immediately. SETI and Einstein are functioning without errors. So what's up?

The 266.58 are the last drivers that seem to be problem-free, no downclocking bug and obviously no sleep mode bug.

OK, so I went back step by step to to 285.72 drivers. All CUDA tasks have performed without errors. I have not been able to test with CPUGRID, as I am waiting for a new WU.

Matman

Win 7 64-bit (SP1)
Dual GTX 580s
Driver: 295.73
Power Control Panel -> Turn off the display: Never

Result: no errors

There are many many people out there with the 295.73 driver, and must be causing thousands of errors on the GPUGRID projects.

People in general, will be reluctant to role back drivers to earlier versions, because GPUGRID is not the primary reason to own or use a computer.


GPUGRID is wasting valuable data at this moment, because of many thousands of errors.

Therefore, shouldn't the GPUGRID team do something themselves, instead of asking every member to change drivers?

Do what?

GPUGrid has to rely on the system to deal with these issues. Users that continuously fail tasks will stop getting tasks.
If you ban users with specific drivers, unless the drivers universally fail, you will be banning users that complete tasks successfully too.

Do what?

GPUGrid has to rely on the system to deal with these issues. Users that continuously fail tasks will stop getting tasks.
If you ban users with specific drivers, unless the drivers universally fail, you will be banning users that complete tasks successfully too.

Has anyone investigated why the tasks fail with some drivers?

On a global scale, Nvidia will not change their drivers to pander to a relatively small group.

Therefore, shouldn't GPUGRID be looking at re-writing the code required to untertake the tasks under the newer drivers?

Do what?

On a global scale, Nvidia will not change their drivers to pander to a relatively small group.

This bug is considered as release critical (show-stopper) for the next NVIDIA driver release that's due in 2-4 weeks. Thus a fix will be available by that time.

The latest version of the BOINC client software + manager, that being version 7.0.25 , has taken it upon themselves not to recognize the GPU anymore, if the user has one of those 'incompatible drivers'.

I think that this represents a mistake, because now that I've programmed my own Windows 7 Pro, x64 computer never to switch off the monitor, I am handing in work units successfully again, and I do think it's unrealistic thinking from BOINC, that users will downgrade their drivers, for the sake of BOINC.

I'm using driver version 296.10 successfully now.

Actually, one reason I had for upgrading from my outdated drivers, was my concern that 'the old method' of implementing "PhysX", would have used a discrete "PPU" (Physics Processing Unit) on my graphics card, and I wanted to /make sure/ that since this approach has been abandoned by nVidia, in favor of using the "GPGPU" itself, my own graphics card should also be using the GPGPU.

Especially since the instructions for downgrading, now tell us to remove ALL nVidia software from our computers, this has become a totally infeasible thing for me to do, with PhysX and "CUDA" SDKs all installed and working.

I have to add something to the advice, for how to prevent the monitor from sleeping though. Well enough, one would set the general Power Settings, accessible through Screensaver preferences. But then it can happen that some other process tries to give the command anyway, to put the monitor to sleep, especially since /some of us/ have sundry programs installed.

The stronger setting I would recommend would be (in addition to the standard setting):

Start Menu
Type in "Edit Group Policy" into the search field and hit Enter
Computer Configuration
Administrative Templates
System
Power Management
Video And Display Settings
Turn Off Display (Plugged In AND On Battery)
--> Disabled

What this does on Windows 7 Pro at least, is take away the privileges processes would have, which we might not have kept rack of, to put the monitor to sleep.

I think that by simply banning all up-to-date device drivers, the new version of BOINC client software will kill off one major source of contributed work for you. The main reason my own GC did crash at one point, was simply the fact that I had not researched the subject (in the forums), and it's not likely to happen to me again.

Dirk

Sorry to double post, but it seems important:

(not my words)
You'll be happy to know that 301.24 fixes the sleeping monitor Bug, althrough i haven't tried it on PrimeGrid yet, only Einstein & Seti so far,

Claggy (Some user on PG)

First post
I did 7 Setiathome offline Benches last night and couldn't get it to fail (But i'm using a different monitor to when i could get to fail with 295.xx drivers),
Before i upgraded i grabbed some BRP4Cuda work and have done some of it this morning no problem,
In a little while i'll downgrade to 295.73 and check i can get offline benches to fail on this monitor.

Second:

I downgraded my i7-2600K/GTX460/HD5770 host to 295.73, ran a setiathome offline bench, proved that the cuda apps do fail with this monitor, then upgraded back up to 301.24,

EDIT: Decided to check for myself using 301.10, and after letting monitor sleep for awhile, I resumed and saw that GPU usage remained steady. Still have to wait for validation from wingman

Doesn't it even seem to matter to you, that BOINC's new client and manager package, is overriding the individual projects' policies on the subject?

Dirk

It does, but since many people rely on auto update, or always get latest driver. This may become a moot point after awhile. From what I can tell, only people who know what they're doing didn't use those drivers anyway, and since BOINC blacklisted them it won't matter when NVIDIA releases WHQL. I mean it prevents failed WU, and even Einstein quit allowing people to use those drivers (they blacklisted WU from going to hosts w/ those drivers anyways. Even if you fixed it yourself it wouldn't work. Should actually help projects in the long run, even if its rude to the users.

EDIT: by allowing BOINC to blacklist, it would have allowed me to use my 680 on Einstein, since they would know that even though mines higher than there 290 limit (301), they would have been able to send me WU. Arrogant, kinda yea, but MANY WU are failing everywhere b/c of it. Especially here I do believe

The latest version of the BOINC client software + manager, that being version 7.0.25 , has taken it upon themselves not to recognize the GPU anymore, if the user has one of those 'incompatible drivers'.

@Michael Goetz:

It's possible that I misread the issues with the newer BOINC Manager and Client.
What they wrote, is that when we install BOINC as a Service, OR in Protected Execution, GPU detection won't work anymore.

http://boinc.berkeley.edu/wiki/Release_Notes#Protected_Application_Execution_.28Service.29_Installation.2C_GPU_detection_and_Windows_XP

I was under the impression that 'as a service' is the opposite of 'in protected execution'.

If in fact they are one and the same thing, then I got it wrong. In that case, BOINC installed 'in User Mode' will still recognize the GPUs (without problem)... If that's so, you might want to make the text just a tad more clear about it. How does it address the malfunctions?

It does, but since many people rely on auto update, or always get latest driver. This may become a moot point after awhile. From what I can tell, only people who know what they're doing didn't use those drivers anyway, and since BOINC blacklisted them it won't matter when NVIDIA releases WHQL.

@Michael Goetz:

It's possible that I misread the issues with the newer BOINC Manager and Client.
What they wrote, is that when we install BOINC as a Service, OR in Protected Execution, GPU detection won't work anymore.

I'm not trying to convince anyone of anything. My point is that if you are using the WHQL driver for game development and the like (i play games) than you would upgrade to newest WHQL in order to have their ( NVIDIA) latest software. In this case you would be upgrading to 300 series whenever the WHQL is released if I'm not mistaken. This was my point, if your using the latest currently, than why not upgrade when newest is released. I personally use NVIDIA website so i can do a clean install. When I made my comment I was merely saying if boinc is currently being run on the side, than I would ASSUME you would want the latest. This being 300, which is a good thing for everyone all around.

My apology. I thought that I was being urged to upgrade to the beta driver, etc.. I can upgrade to the 300.xx driver as soon as it becomes WHQL, just because my current setup will continue to work for now.

And I did just upgrade my client software to 7.0.25, as requested.

Dirk

Quite allright. I've been trying to spread the word to various sites, b/c as Michael had stated, it has been a HUGE problem from the projects standpoint. MANY MANY errors have been caused by this monitor sleep bug, and when I said, "only the people that know what they're doing don't use it anyways" I should have been more clear. I meant to mean the BOINC ONLY crowd, but since this can be a rather small percentage on some sites, many users who aren't BOINC ONLY (they attach project and leave it) w/o ever checking results, they just keep producing errors w/o knowing it.

No need to go to beta if you already prevent monitor from sleeping, but not everyone does this, and this was the problem. People who play games etc. in spare time want/need the latest drivers in order for their system to function properly (whether it's WHQL or not).

All in all, it's great news for the BOINC community as a whole, b/c now everyone's happy (will be soon anyways when WHQL is released). From both the projects side (valid WU), and latest and greatest PhyX etc.

As always Happy Crunching

@Michael Goetz:

It's possible that I misread the issues with the newer BOINC Manager and Client.
What they wrote, is that when we install BOINC as a Service, OR in Protected Execution, GPU detection won't work anymore.

I'm pretty sure that's a feature that's been in BOINC for many years now, and certainly isn't driver specific. It definitely was in the 6.12.34 client, and possibly in all of the 6.x.x clients.

Toshiba has now released a new display driver for notebooks that hava a NVIDIA card installed.
The driver seems to be the 296.31 version.
Has anyone tested this toshiba driver?
I guess I better stick with my older version which has worked without a problem and wait for a new toshiba version based on the 300.xx series.
____________

Toshiba has now released a new display driver for notebooks with a NVIDIA card installed.
The driver seems to be the 296.31 version.
Has anyone tested this toshiba driver?

for anyone trying to reproduce the problem, I have found that the problem occurs when Windows powers off the monitor first, and then BOINC tries to start or resume a CUDA task while the monitor is off. This means that, if you try to reproduce it using tasks that are already running before Windows powers down the monitor, those tasks will not fail. But any tasks that try to start or resume, while the monitor is off, will fail... according to my testing.