[solved] nVidia 295.51 beta driver problems - Failed tasks - Cannot create a symbolic link in a registry key that already has subkeys or values. (0x3fc) - exit code 1020 (0x3fc)
Talking about WHQL drivers: the latest WHQL release for the GeForce 600 series is version 301.10. Has anyone ever tested this version? Does it work flawlessly? We would really appreciate if those you can reproduce that issue could test this version and provide feedback.
Background: as Bernd announced, CUDA tasks are shipped again for drivers >= 301.xx which includes 301.10. However, we currently do have positive feedback only for version 301.24. I just want to make sure we set our driver version filter correctly...
Talking about WHQL drivers: the latest WHQL release for the GeForce 600 series is version 301.10. Has anyone ever tested this version? Does it work flawlessly? We would really appreciate if those you can reproduce that issue could test this version and provide feedback.
Background: as Bernd announced, CUDA tasks are shipped again for drivers >= 301.xx which includes 301.10. However, we currently do have positive feedback only for version 301.24. I just want to make sure we set our driver version filter correctly...
You would have to ask one of the limited number of 'early adopters' about the bug - I'll see if I can find anything reported in any of the many similar threads on other project message boards.
@ Oliver That was the one that I actually tested on. 301.10, I only used 301.24 breiefly, since it appeared to be 301.10 except for 5xx and below.
Currently I'm actually using 301.25
Again, I tested on 301.10 and that WHQL is for 680 ONLY. I just checked NVIDIAs website, and when I typed in 5xx series filter only 296.10 and 301.24 popped up.
EDIT: lol someone beat me to it.... I will be switching back to launch disk 300.83? Since Intel got into NVIDIAs business and they "kicked" PCIE 3.0 off of these other driver releases. Ordered a 2011 and x79 setup in order to have more lanes, and until NVIDIA adds a new driver with support for PCIE 3.0, I will be using the launch (disk) driver.
Guys, I know that 301.10 is GTX 680 only, but still, I'd like to know whether this contains the bug fix included in 301.24 or not. Otherwise the GTX 680 users will get CUDA tasks that might fail on their system, and we don't want that.
@5pot: this means that you tested 301.10 and it worked for you, although you didn't use it on a Kepler device? So, you can confirm the actual bug discussed here is fixed in 301.10 as well?
Yes Oliver. When i ran my tests it was on 301.10 on a 680, and the bug was not discovered. I do see that your app is for 301+ so i have still not decided on whether or not i will revert back to 300.83 in order to use pcie 3.0. Depends on how long NVIDIA takes to fix this pcie issue on x79 devices, which i am installing currently.
Well, I am using the 301.24 drivers with a GTX 560Ti, and unfortunately I am still seeing the CUDA failures. I'll try the workaround of disabling the monitor power mgmt, but at least in my case, the bug is not fixed.
Well, I am using the 301.24 drivers with a GTX 560Ti, and unfortunately I am still seeing the CUDA failures. I'll try the workaround of disabling the monitor power mgmt, but at least in my case, the bug is not fixed.
I see that today your host succeeded to crunch several CUDA units. Any idea what fixed it?
Is not finding a gpu device different from failing to initialize cuda? Thought bug only prevented cuda from starting and not finding gpu. Meaning there would have been a couple seconds from starting than it would fail. Not, not starting at all.?
Talking about WHQL drivers:
)
Talking about WHQL drivers: the latest WHQL release for the GeForce 600 series is version 301.10. Has anyone ever tested this version? Does it work flawlessly? We would really appreciate if those you can reproduce that issue could test this version and provide feedback.
Background: as Bernd announced, CUDA tasks are shipped again for drivers >= 301.xx which includes 301.10. However, we currently do have positive feedback only for version 301.24. I just want to make sure we set our driver version filter correctly...
Thanks,
Oliver
Einstein@Home Project
RE: Talking about WHQL
)
According to http://www.nvidia.com/object/win7-winvista-32bit-301.10-whql-driver.html (see 'supported products' tab), the WHQL v301.10 driver only supports the GTX 680 ('Kepler') GPU - which was only released a month ago, as discussed in cruncher's corner.
You would have to ask one of the limited number of 'early adopters' about the bug - I'll see if I can find anything reported in any of the many similar threads on other project message boards.
@ Oliver That was the one
)
@ Oliver That was the one that I actually tested on. 301.10, I only used 301.24 breiefly, since it appeared to be 301.10 except for 5xx and below.
Currently I'm actually using 301.25
Again, I tested on 301.10 and that WHQL is for 680 ONLY. I just checked NVIDIAs website, and when I typed in 5xx series filter only 296.10 and 301.24 popped up.
EDIT: lol someone beat me to it.... I will be switching back to launch disk 300.83? Since Intel got into NVIDIAs business and they "kicked" PCIE 3.0 off of these other driver releases. Ordered a 2011 and x79 setup in order to have more lanes, and until NVIDIA adds a new driver with support for PCIE 3.0, I will be using the launch (disk) driver.
Guys, I know that 301.10 is
)
Guys, I know that 301.10 is GTX 680 only, but still, I'd like to know whether this contains the bug fix included in 301.24 or not. Otherwise the GTX 680 users will get CUDA tasks that might fail on their system, and we don't want that.
@5pot: this means that you tested 301.10 and it worked for you, although you didn't use it on a Kepler device? So, you can confirm the actual bug discussed here is fixed in 301.10 as well?
Oliver
Einstein@Home Project
Yes Oliver. When i ran my
)
Yes Oliver. When i ran my tests it was on 301.10 on a 680, and the bug was not discovered. I do see that your app is for 301+ so i have still not decided on whether or not i will revert back to 300.83 in order to use pcie 3.0. Depends on how long NVIDIA takes to fix this pcie issue on x79 devices, which i am installing currently.
Well, I am using the 301.24
)
Well, I am using the 301.24 drivers with a GTX 560Ti, and unfortunately I am still seeing the CUDA failures. I'll try the workaround of disabling the monitor power mgmt, but at least in my case, the bug is not fixed.
Interesting: Yours aren't
)
Interesting:
Yours aren't even starting, and the stderr output is different too, at top of this thread it looks like this:
Activated exception handling...
[01:53:12][2316][INFO ] Starting data processing...
[01:53:12][2316][ERROR] Couldn't initialize CUDA driver API (error: 100)!
[01:53:12][2316][ERROR] Demodulation failed (error: 1020)!
01:53:12 (2316): called boinc_finish
Not saying it isn't the drivers, but yours looks different. Hmmm.
FYI, this error means that
)
FYI, this error means that the application couldn't find any GPU device.
Cheers,
Oliver
Einstein@Home Project
RE: Well, I am using the
)
I see that today your host succeeded to crunch several CUDA units. Any idea what fixed it?
Cheers
HB
Is not finding a gpu device
)
Is not finding a gpu device different from failing to initialize cuda? Thought bug only prevented cuda from starting and not finding gpu. Meaning there would have been a couple seconds from starting than it would fail. Not, not starting at all.?