| Author |
Message |
|---|
 Michael Goetz Volunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13877
ID: 53948
Credit: 383,325,983
RAC: 118,067
                              
|
|
Numerous participants have had tasks fail due the Nvidia's buggy 364.xx driver. If you're using this driver, I recommend reverting to an earlier version of the driver with a major version number lower than 364.
If you're currently running an older version of the driver, I strongly recommend that you do not "upgrade" to 364.xx. I've also heard reports of this driver "bricking" GPUs, and that goes way, way beyond merely having some BOINC tasks fail.
____________
My lucky number is 75898524288+1 |
|
|
RafaelVolunteer tester
Send message
Joined: 22 Oct 14
Posts: 909
ID: 370496
Credit: 515,461,478
RAC: 218,071
             
|
Numerous participants have had tasks fail due the Nvidia's buggy 364.xx driver. If you're using this driver, I recommend reverting to an earlier version of the driver with a major version number lower than 364.
If you're currently running an older version of the driver, I strongly recommend that you do not "upgrade" to 364.xx. I've also heard reports of this driver "bricking" GPUs, and that goes way, way beyond merely having some BOINC tasks fail.
Just as a heads up, I've also heard a report of POEM Home (another Boinc project) failing with 364.xx, but doing perfectly fine with 362.00 and below.
If you upgraded and didn't have issues by now, you are most likely going to fine. But if you haven't upgraded yet, it's advised to NOT upgrade, or at the very least proceed with caution. |
|
|
mackerelVolunteer tester
Send message
Joined: 2 Oct 08
Posts: 2609
ID: 29980
Credit: 566,938,418
RAC: 18,312
                   
|
|
364.47 was the one that some reported significant problems with during install. That got pulled quickly, and was replaced later with 364.51. Since then they have also released 364.72. I'm not aware of general or gaming problems with the updated drivers. '72 is currently installed on both my laptop and main desktop without problems, and '51 previous to that, but I don't GPU crunch so can't comment there. |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13877
ID: 53948
Credit: 383,325,983
RAC: 118,067
|
364.47 was the one that some reported significant problems with during install. That got pulled quickly, and was replaced later with 364.51. Since then they have also released 364.72. I'm not aware of general or gaming problems with the updated drivers. '72 is currently installed on both my laptop and main desktop without problems, and '51 previous to that, but I don't GPU crunch so can't comment there.
364.16 is also problematic. I'll look through the forums and see if there's any others that are known to be bad.
____________
My lucky number is 75898524288+1 |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13877
ID: 53948
Credit: 383,325,983
RAC: 118,067
|
|
364.xx driver versions that are known to be problematic:
364.16
364.47
364.51
364.72
364.74 ??? (might be a typo of 364.47)
Each driver version above has had at least one person report having problems running with that driver that went away when they changed to a driver below 364.xx.
____________
My lucky number is 75898524288+1 |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1943
ID: 352
Credit: 5,924,750,150
RAC: 1,520,730
                              
|
|
How about 364.96 - do we know any status on those?
http://international.download.nvidia.com/Windows/364.96/364.96-geforce-win10-64bit-international.hf.exe
http://international.download.nvidia.com/Windows/364.96/364.96-desktop-notebook-win8-win7-winvista-64bit-international.hf.exe
____________
My stats |
|
|
|
|
|
364.72 without problems for me too (actually manual sieving on GFN N=65536). The past week, I also run PPS (Sieve) v1.39 (cudaPPSsieve) work units without problems.
____________
Founder of CRUNCHERS SANS FRONTIERES
www.crunchersansfrontieres.org
CSF lucky number 22872882^65536+1 |
|
|
|
|
|
Nothing new to add, just a thank you.
I had a series of failed GFN17Mega WUs on 364.51 e.g. 700569742, I just thought I'd reached some limit for my card so I switched back to PPS Sieving and didn't get any problems with those WUs.
Now that I've seen the comments about the driver issues I've reverted to 359.00
Thanks for letting us know,
Pete.
It's a pity we can't let all contributors who are working on GFN units and are running those drivers know about the problem, let's hope they read your message boards or pick up the warning from the BOINC client.
____________
35 x 2^3587843+1 is prime! |
|
|
axnVolunteer developer
Send message
Joined: 29 Dec 07
Posts: 285
ID: 16874
Credit: 28,027,106
RAC: 0
           
|
|
*Never mind* |
|
|
|
|
|
Is this problem only with OpenCL? I processed the last couple of Genefer World Record work units, the 5 1/2 day jobs, using cuda and they seem to have processed ok.
Although a few unknown errors plague the stderr output, this seems normal for a wu that has completed and was validated
(http://www.primegrid.com/result.php?resultid=700320661)
I upgraded to driver 364.72 for VR support and don't particularly want to downgrade |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13877
ID: 53948
Credit: 383,325,983
RAC: 118,067
|
Is this problem only with OpenCL? I processed the last couple of Genefer World Record work units, the 5 1/2 day jobs, using cuda and they seem to have processed ok.
Although a few unknown errors plague the stderr output, this seems normal for a wu that has completed and was validated
(http://www.primegrid.com/result.php?resultid=700320661)
I upgraded to driver 364.72 for VR support and don't particularly want to downgrade
It's unclear under exactly what circumstances there's a problem. Anecdotally, it does seem to me that this driver's problems are related only to the OpenCL programs. However, given that there's the possibility of destroying your GPU with this driver, we're not about to go and do any testing on it to figure out exactly what is and is not working. We're just avoiding this driver like the plague.
Given the ""Oh, and by the way, it can also brick (destroy) your GPU" nature of this problem, I don't expect many people will opt to stay with this driver no matter what benefits it provides. Unless, of course, one is looking for an excuse to upgrade. :)
It's a pity we can't let all contributors who are working on GFN units and are running those drivers know about the problem, let's hope they read your message boards or pick up the warning from the BOINC client.
That's exactly what happens when I post anything in the news forum. The post goes out and appears in the notifications tab of every BOINC client that connects to PrimeGrid. This only applies to the first post in the thread. This is the mechanism by which BOINC projects send notifications to the BOINC clients. It's unreasonable to expect everyone to read the forums, although many do. But the news posts get pushed to everyone's client, which should reach a wider audience, especially those who can read and understand English.
____________
My lucky number is 75898524288+1 |
|
|
|
|
Numerous participants have had tasks fail due the Nvidia's buggy 364.xx driver. If you're using this driver, I recommend reverting to an earlier version of the driver with a major version number lower than 364.
If you're currently running an older version of the driver, I strongly recommend that you do not "upgrade" to 364.xx. I've also heard reports of this driver "bricking" GPUs, and that goes way, way beyond merely having some BOINC tasks fail.
What is "bricking"? |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13877
ID: 53948
Credit: 383,325,983
RAC: 118,067
|
What is "bricking"?
TL;DR: When referring to electronics, it means "to make completely unusable, permanently".
brick
/brik/
verb
gerund or present participle: bricking
1. block or enclose with a wall of bricks.
"the doors have been bricked up"
2. informal
cause (a smartphone or other electronic device) to become completely unable to function, typically on a permanent basis.
"installing an unofficial OS voids the warranty and may brick the phone"
The term "bricking" is derived from the concept of turning a piece of useful electronic gear into something no more useful than a brick. Afterwards, you can use it as a paperweight, but beyond that it's pretty useless.
In most cases, bricking is the result of a failed attempt to change the BIOS or other firmware in a device, with the end result that you can no longer start up the device or correct the firmware. For all intents and purposes, it's completely dead. It's now a "brick".
____________
My lucky number is 75898524288+1 |
|
|
GDBSend message
Joined: 15 Nov 11
Posts: 284
ID: 119185
Credit: 3,806,299,758
RAC: 2,307,459
   
|
|
I had problems with GFN sieving for n=22 until I downgraded my Nvidia driver. |
|
|
|
|
|
Before downgrade to 362.00, problems centered on openCL. Cuda on PPS seive worked fine. Bricks are easy to find, don't need to turn my GPU into one :) |
|
|
|
|
|
Greetings!!! I'm one of the Linux users & was surprised to see this in BOINC Manager's feed... I belong to nVidia dev forums & have posted this thread there...Waiting to hear back from the Devs about this.........Must say that I use the 364.15 beta Linux driver & have not seen any problems with it...will keep my eyes open & report anything I hear back from nVidia. |
|
|
|
|
|
Here is a cut and paste from the geforce forums:
OpenCL is still broken with 364.96, causing constant driver crashes...
End quote.
This is a beta driver for the new Doom with little bug fixes. |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13877
ID: 53948
Credit: 383,325,983
RAC: 118,067
|
|
There's now evidence that the 364.xx driver is causing faulty results. As such, sometime in the near future I will be blocking this driver completely and the server will no longer send tasks to computers running that driver. Reverting to an earlier driver will be mandatory. "Near future" is probably sometime in the next 5 to 10 hours.
More details will be made public as soon as possible.
____________
My lucky number is 75898524288+1 |
|
|
|
|
|
I am using 364.72 WHQL - I don't use beta drivers except in rare cases for gaming, only. I had a few GFN-18 tasks validate by other crunchers using either an older GeForce driver (361.xx) or AMD drivers.
I can switch to an older GeForce driver like 361.xx or 362.xx if PrimeGrid servers are configured to reject 364.xx, but I'm not sure that 364.xx causes consistent problems for every OCL app on every Windows machine. |
|
|
RafaelVolunteer tester
Send message
Joined: 22 Oct 14
Posts: 909
ID: 370496
Credit: 515,461,478
RAC: 218,071
|
I am using 364.72 WHQL - I don't use beta drivers except in rare cases for gaming, only. I had a few GFN-18 tasks validate by other crunchers using either an older GeForce driver (361.xx) or AMD drivers.
I can switch to an older GeForce driver like 361.xx or 362.xx if PrimeGrid servers are configured to reject 364.xx, but I'm not sure that 364.xx causes consistent problems for every OCL app on every Windows machine.
It seems to be a Maxwell specific issue, that's why your tasks are validating. But better be safe than sorry and downgrade drivers... |
|
|
|
|
It seems to be a Maxwell specific issue, that's why your tasks are validating. But better be safe than sorry and downgrade drivers...
Agreed. |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13877
ID: 53948
Credit: 383,325,983
RAC: 118,067
|
|
We've made changes to the validator rather than blocking the tasks. So while we're still going to send tasks out, the validator will require at least one matching result that does not come from a 364.xx GPU task in order to validate.
We do not fully understand the exact circumstances where this driver is failing, so we're treating all results produced by the driver as suspect.
____________
My lucky number is 75898524288+1 |
|
|
|
|
We've made changes to the validator rather than blocking the tasks. So while we're still going to send tasks out, the validator will require at least one matching result that does not come from a 364.xx GPU task in order to validate.
We do not fully understand the exact circumstances where this driver is failing, so we're treating all results produced by the driver as suspect.
That sounds like a pretty good compromise. I'll just watch for new posts on the subject should the policy on 364.xx change in the future. |
|
|
|
|
|
It's because of issues like this that I rarely, if ever, upgrade my Nvidia drivers.
____________
 |
|
|
|
|
We've made changes to the validator rather than blocking the tasks. So while we're still going to send tasks out, the validator will require at least one matching result that does not come from a 364.xx GPU task in order to validate.
We do not fully understand the exact circumstances where this driver is failing, so we're treating all results produced by the driver as suspect.
That sounds like a pretty good compromise.
+1 :)
____________
Founder of CRUNCHERS SANS FRONTIERES
www.crunchersansfrontieres.org
CSF lucky number 22872882^65536+1 |
|
|
|
|
We've made changes to the validator rather than blocking the tasks. So while we're still going to send tasks out, the validator will require at least one matching result that does not come from a 364.xx GPU task in order to validate.
We do not fully understand the exact circumstances where this driver is failing, so we're treating all results produced by the driver as suspect.
Michael,
Does your validator protect against 365.xx? Basically, all R364 drivers are bugged for me, including 365.10.
|
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13877
ID: 53948
Credit: 383,325,983
RAC: 118,067
|
We've made changes to the validator rather than blocking the tasks. So while we're still going to send tasks out, the validator will require at least one matching result that does not come from a 364.xx GPU task in order to validate.
We do not fully understand the exact circumstances where this driver is failing, so we're treating all results produced by the driver as suspect.
Michael,
Does your validator protect against 365.xx? Basically, all R364 drivers are bugged for me, including 365.10.
Yes it does. Jim modified the validator such that anything greater than 363.xx would be treated as suspect. It will stay that way until we can verify that a new driver in fact fixes the problem.
We did anticipate that the problem might not be fixed immediately.
____________
My lucky number is 75898524288+1 |
|
|
|
|
...We did anticipate that the problem might not be fixed immediately. I'm using 359.00 and getting inconclusive tasks appear for partners using 365.10 so it looks as if you're right on the money.
I won't be upgrading again until I get the all clear from this site.
Pete.
|
|
|
|
|
|
What I've seen so far indicates that 368.22 fixes the problems if it will install properly - but the installation fails on some hardware/Windows combinations in a way that makes the computer unbootable.
It installed properly on my Windows 10 computer and is running without errors.
However, on my Windows Vista computer (Intel Q9650 CPU, GTX 560, 8 GB memory, 64-bit Windows Vista SP1), the screen went black perhaps halfway through the installation, and stayed that way until I turned off the computer about an hour later. Attempting to reinstall 362.00 ended with messages that newer versions of some parts were already present, and were not changed. The computer has gone downhill over another 6 hours and now will not boot, either normally or in safe mode. While I was still able to run the hardware diagnostics, they reported no errors.
I've already reported this to Nvidia, but can any of you suggest what to do to make that computer run properly again? |
|
|
|
|
What I've seen so far indicates that 368.22 fixes the problems if it will install properly - but the installation fails on some hardware/Windows combinations in a way that makes the computer unbootable.
It installed properly on my Windows 10 computer and is running without errors.
However, on my Windows Vista computer (Intel Q9650 CPU, GTX 560, 8 GB memory, 64-bit Windows Vista SP1), the screen went black perhaps halfway through the installation, and stayed that way until I turned off the computer about an hour later. Attempting to reinstall 362.00 ended with messages that newer versions of some parts were already present, and were not changed. The computer has gone downhill over another 6 hours and now will not boot, either normally or in safe mode. While I was still able to run the hardware diagnostics, they reported no errors.
I've already reported this to Nvidia, but can any of you suggest what to do to make that computer run properly again?
See my reply here:
http://boinc.fzk.de/poem/forum_thread.php?id=1205&postid=11079
368.22 does not support Vista. 365.19 is the latest Vista driver. |
|
|
|
|
368.22 does not support Vista. 365.19 is the latest Vista driver.
Then it's missing something important - a test for attempts to install it under Vista, that will give the user a chance to stop trying to install it. |
|
|
|
|
What I've seen so far indicates that 368.22 fixes the problems if it will install properly - but the installation fails on some hardware/Windows combinations in a way that makes the computer unbootable.
It installed properly on my Windows 10 computer and is running without errors.
However, on my Windows Vista computer (Intel Q9650 CPU, GTX 560, 8 GB memory, 64-bit Windows Vista SP1), the screen went black perhaps halfway through the installation, and stayed that way until I turned off the computer about an hour later. Attempting to reinstall 362.00 ended with messages that newer versions of some parts were already present, and were not changed. The computer has gone downhill over another 6 hours and now will not boot, either normally or in safe mode. While I was still able to run the hardware diagnostics, they reported no errors.
I've already reported this to Nvidia, but can any of you suggest what to do to make that computer run properly again?
Download, install, and run DDU by Wagnard Tools under Windows Safe Mode to do a thorough cleaning of all NVIDIA driver components and the Windows registry. Then re-install driver 362.00. Download links and basic instructions on DDU are available here.
EDIT: Caution - Use at your own risk. DDU utility is not produced by NVIDIA but they allow the author to maintain and support DDU on the official GeForce Driver forum. I've used it successfully to remove all traces of old drivers before doing a fresh install and have not had problems with its use, but as always, your mileage may vary. |
|
|
|
|
Download, install, and run DDU by Wagnard Tools under Windows Safe Mode to do a thorough cleaning of all NVIDIA driver components and the Windows registry. Then re-install driver 362.00. Download links and basic instructions on DDU are available here.
EDIT: Caution - Use at your own risk. DDU utility is not produced by NVIDIA but they allow the author to maintain and support DDU on the official GeForce Driver forum. I've used it successfully to remove all traces of old drivers before doing a fresh install and have not had problems with its use, but as always, your mileage may vary.
I looked at the web page for DDU. Nothing was said about whether it can run if neither normal mode nor safe mode will boot (my situation). There is no way to ask the author without logging in, and no indication of what to sign up for so I will have an account that can log in. Nothing was said about whether it will run under Vista.
I have a newer computer almost ready to install, but the room cannot supply enough power to run both the new computer and my Windows Vista computer at once. It currently looks best to hurry up installation of the new computer, and start leaving the new computer turned on and the Vista computer turned off.
|
|
|
|
|
I looked at the web page for DDU. Nothing was said about whether it can run if neither normal mode nor safe mode will boot (my situation). There is no way to ask the author without logging in, and no indication of what to sign up for so I will have an account that can log in. Nothing was said about whether it will run under Vista.
I have a newer computer almost ready to install, but the room cannot supply enough power to run both the new computer and my Windows Vista computer at once. It currently looks best to hurry up installation of the new computer, and start leaving the new computer turned on and the Vista computer turned off.
From the first post in the linked thread:
-The tool can be used in Normal mode but for absolute stability and effectiveness, when using DDU, Safemode is always the best and using the button#1 (Clean+restart) are the recommended options.
-You do NOT need to uninstall the driver before using DDU.
<snip...>
Keep note that NVIDIA/AMD/INTEL did not have anything to do with this, I do not work at or for NVIDIA/AMD/INTEL and they should not be held responsible for anything that may go wrong with this application.
NOTE: .NET framework 2.0 or higher.
Windows XP up to 10 is supported X86/X64
Again, if you are not accustomed to running a utility such as DDU, don't do so. It's only a suggestion. You already have a new computer almost ready to install.
You might also check your system BIOS to see if onboard graphics is enabled. A system should be able to boot even without a graphics card unless some settings have been changed in the BIOS. |
|
|
|
|
Windows XP up to 10 is supported X86/X64
Again, if you are not accustomed to running a utility such as DDU, don't do so. It's only a suggestion. You already have a new computer almost ready to install.
You might also check your system BIOS to see if onboard graphics is enabled. A system should be able to boot even without a graphics card unless some settings have been changed in the BIOS.
Good - Vista is supported.
I'm accustomed to running many utilities, but not that one.
The system BIOS settings do not offer anything related to on-board graphics, and there is no connector for them to use.
I finally found a way to tell the computer to boot from a DVD, and plan to run the extra diagnostics from the System Recovery DVDs before I try anything that will wipe out the data on the computer with problems. |
|
|
|
|
|
I now managed a normal boot, and BOINC is running with the 362.00 driver. |
|
|
