Hello.
A friend has just tried the 185.13 drivers on Debian 64b
And all his WU are in computation error.
The error given is :
<core_client_version>6.6.15</core_client_version>
<![CDATA[
<message>
process exited with code 193 (0xc1, -63)
</message>
<stderr_txt>
# Using CUDA device 0
SIGSEGV: segmentation violation
Stack trace (17 frames):
acemd_6.59_x86_64-pc-linux-gnu__cuda[0x4baac9]
/lib/libc.so.6[0x7f48eab1af60]
/usr/lib/libcuda.so.1[0x7f48eba0bce0]
/usr/lib/libcuda.so.1[0x7f48eba11a44]
/usr/lib/libcuda.so.1[0x7f48eb9d79df]
/usr/lib/libcuda.so.1[0x7f48eb65f9cb]
/usr/lib/libcuda.so.1[0x7f48eb6702cb]
/usr/lib/libcuda.so.1[0x7f48eb6580c1]
/usr/lib/libcuda.so.1(cuCtxCreate+0xaa)[0x7f48eb65224a]
../../projects/www.gpugrid.net/libcudart.so.2[0x7f48ebc8cd58]
../../projects/www.gpugrid.net/libcudart.so.2[0x7f48ebc8d2a9]
../../projects/www.gpugrid.net/libcudart.so.2(cudaThreadSynchronize+0x1d)[0x7f48ebc7374d]
acemd_6.59_x86_64-pc-linux-gnu__cuda[0x414253]
acemd_6.59_x86_64-pc-linux-gnu__cuda(sin+0x16ac)[0x408a3c]
acemd_6.59_x86_64-pc-linux-gnu__cuda(sin+0x31b)[0x4076ab]
/lib/libc.so.6(__libc_start_main+0xe6)[0x7f48eab071a6]
acemd_6.59_x86_64-pc-linux-gnu__cuda(sinh+0x49)[0x407489]

Exiting...

</stderr_txt>
]]>

Any ideas ?

There are issues when running s@h and GPU-Grid together. I don't know everythig, but it seems that if seti errors out the PC will need a reboot to use CUDA again (GPU-Grid also erros). But there may be more.

Stock shaders for 9600GT are 1.625 GHz. I'd put the clocks at NV stock for one WU or 2 and switch to the old values afterwards [if successful ;) ].

MrS

Thanx i put it on these to see if it helps also
I slowly am raising the clock speed of the core but i guess it won't matter much untill it reaches the normal clock of 650 Mhz
Ill keep on setting a step higher till stock speeds are reached on the shaders as well, as i understood memory speed doesn't do much i keep it close to stock.
I should have started boinc under x64 vista instead of under normal xp :D
Then i could have used the full 8 Gb mem xD
Or i have to run the machine totally dry but thats gonna take a while since CPDN has downloaded a large one.

I think I may have fixed this issue:
If you have Windows XP Home and you want to switch profiles whilst using Boinc, follow the steps below found here:
http://setiathome.berkeley.edu/forum_thread.php?id=50929

I didn't even know you could get to the NTFS permissions on XP Home.

Regards,

Rob

Turns out that this didn't work for me. The GPU task runs for a few seconds and then fails out followed by any queued tasks for that day.

I have reinstalled Boinc as a service so changing profile should not interfere with the tasks. I'll update this thread with my findings.

Regards,

Rob

I think I may have fixed this issue:
If you have Windows XP Home and you want to switch profiles whilst using Boinc, follow the steps below found here:
http://setiathome.berkeley.edu/forum_thread.php?id=50929

I didn't even know you could get to the NTFS permissions on XP Home.

Regards,

Rob

Turns out that this didn't work for me. The GPU task runs for a few seconds and then fails out followed by any queued tasks for that day.

I have reinstalled Boinc as a service so changing profile should not interfere with the tasks. I'll update this thread with my findings.

Regards,

Rob

It's the same for me... Perhaps we have to wait for the next CUDA version???

Kind regards

Joe

I have reinstalled Boinc as a service so changing profile should not interfere with the tasks. I'll update this thread with my findings.

I don't think you can run the GPU tasks as a service (a limitation of CUDA itself I believe).

I don't think you can run the GPU tasks as a service (a limitation of CUDA itself I believe).

That's correct. The Windows installation process for both 6.4.5 and 6.4.7 explicitly states that you can not install BOINC as a service if you want to use CUDA.

I have BOINC running as a service on all my Windows machines *except* for the one where I'm running CUDA. Can't install it as a service there.

Mike

I'm testing 6.6.16 Beta http://boinc.berkeley.edu/dl/boinc_6.6.16_windows_intelx86.exe and it seem to work now with XP Prof 32 Bit. Working now since 4 hours without an error... Let me see more tomorrow.

Kind regard

Joe

I don't think you can run the GPU tasks as a service (a limitation of CUDA itself I believe).

That's correct. The Windows installation process for both 6.4.5 and 6.4.7 explicitly states that you can not install BOINC as a service if you want to use CUDA.

I have BOINC running as a service on all my Windows machines *except* for the one where I'm running CUDA. Can't install it as a service there.

Mike

Mike,

I am running 6.6.15 and BOINC is running CUDA whilst installed as a service.

They must have resolved whatever issue there was with the stable editions.

Rob

The issue is only with Vista.
On XP it worked all the time with BOINC installed as a service...

---

pixelicious.at - my little photoblog

I am running 6.6.15 and BOINC is running CUDA whilst installed as a service.

Sweet!

In that case, I'm certainly looking forward to the release of a stable 6.6.x client. I much prefer running BOINC as a service.

I prefer not to run the beta versions of the client, so I'm sticking to the stable release versions. I like running CPDN, and with the length of those work units, too much work gets lost if they error out. Yeah, I know I can backup/restore those WUs, but then I have to remember to back them up, and restoration is less than a pleasant process when you're running lots of projects.

Mike

Edit:

The issue is only with Vista.
On XP it worked all the time with BOINC installed as a service...

Well, maybe not so sweet then...

A few minutes before finishing the WU I got "incorrect function (0x1) exit code 1" with 6.6.16... Now I'm testing 6.6.17...

Any ideas for the error 193 with the 185.13 drivers on Debian 64b ?

Just had 4 errors on my Vista machine......looks like my fellow crunchers have errored out too. Is there a bad batch of Wu's out there atm? Just thought its worth checking as my GTX 295 is a factory overclock card and it could possibly cause some errors.

http://www.gpugrid.net/workunit.php?wuid=318412
http://www.gpugrid.net/workunit.php?wuid=317639
http://www.gpugrid.net/workunit.php?wuid=317457
This one was completed by someone else but had error at same time as one of the other WU's....possibly killed when he other one went...i dont know
http://www.gpugrid.net/workunit.php?wuid=317194

[quote]I am running 6.6.15 and BOINC is running CUDA whilst installed as a service.

Sweet!

In that case, I'm certainly looking forward to the release of a stable 6.6.x client. I much prefer running BOINC as a service.
/quote]

I thought installing it as a service would fix the problem, but my tasks again failed out when logging into the machine as two users simultaneously on XP Home.

20/03/2009 16:34:22 GPUGRID Computation for task up108704-pYEpYI_US530000-0-10-ignasi_1 finished
20/03/2009 16:34:22 GPUGRID Starting WS10117-SH2_US_8-0-10-SH2_US_8270000_0
20/03/2009 16:34:23 GPUGRID Starting task WS10117-SH2_US_8-0-10-SH2_US_8270000_0 using acemd version 662
20/03/2009 16:34:24 GPUGRID Computation for task WS10117-SH2_US_8-0-10-SH2_US_8270000_0 finished
20/03/2009 16:34:24 GPUGRID Output file WS10117-SH2_US_8-0-10-SH2_US_8270000_0_1 for task WS10117-SH2_US_8-0-10-SH2_US_8270000_0 absent
20/03/2009 16:34:24 GPUGRID Output file WS10117-SH2_US_8-0-10-SH2_US_8270000_0_2 for task WS10117-SH2_US_8-0-10-SH2_US_8270000_0 absent
20/03/2009 16:34:24 GPUGRID Output file WS10117-SH2_US_8-0-10-SH2_US_8270000_0_3 for task WS10117-SH2_US_8-0-10-SH2_US_8270000_0 absent
20/03/2009 16:34:25 GPUGRID Started upload of up108704-pYEpYI_US530000-0-10-ignasi_1_0
20/03/2009 16:34:25 GPUGRID Started upload of up108704-pYEpYI_US530000-0-10-ignasi_1_1

Any suggestions how I can fix this? Should I disable the option on install to allow all users to run Boinc? Anything else?

Rob

Any ideas for the error 193 with the 185.13 drivers on Debian 64b ?

Just a check on your environment. I just brought up GPUGRID on my Ubuntu 8.10 64-bit machine. Brand new install. And every WU would error out in seconds. No error 193, but it gave me the "output file absent" message.

Check to see if the 32-bit runtime libraries are installed. If not, try that on the next set of WU's. After I installed the IA32 libraries, and the "microcode.ctl" package, the next set of WU's are running fine.

I'm not saying this will fix your problem, but it's something to check.

I'm running nvidia 180.11 drivers, since that's what Ubuntu "likes". And trying to upgrade to the released versions from Nvidia's website just wasn't going too well, so I reverted to "stock".

Also, I'm running an 8600GTS card, and BOINC 6.6.15.

Mark

A long string of compute errors to report (task ids):

426788 4 errors
426924 3 errors
427001 4 errors, 1 canceled
427056 2 errors
427583 3 errors
425600 1 error

The error counts are as of the posting of this note ... obviously these may rise by tomorrow or when ever you all at the project look at these ...

These all look like "new" and "improved" tasks so ... maybe back to the drawing board ...

Certainly one of the batches sent yesterday was corrupted.
*pYIpYV1*
I am canceling them out.

sorry about that,
ignasi

The aborts came through ok - it picked up an additional two bad ones in the queue lurking from the same batch as the three that bombed out earlier this morning UTC time.

Many Thanks - Crunch On :)

Regards
Zy

Help! I have 3 identical winXP AMD 4200 x2 systems with evga 9600gso 384mb cards that have been running GPUGRID happily for at least a month, and suddenly about a week ago ALL of them started giving errors simultaneously. All WUs now crash within seconds of starting. I have tried lots of fixes: going back to stock clocking, underclocking, rebooting, updating drivers, etc. Nothing makes any difference. There have been no changes to my system other than WinXP auto updates. My conclusion is that there was a change in GPUGRID WUs and the problem is not on my side. Here is my typical crash:

http://www.gpugrid.net/result.php?resultid=410378

<core_client_version>6.4.5</core_client_version>
<![CDATA[
<message>
The system cannot find the path specified. (0x3) - exit code 3 (0x3)
</message>
<stderr_txt>
# Using CUDA device 0
# Device 0: "GeForce 9600 GSO"
# Clock rate: 1350000 kilohertz
# Total amount of global memory: 402325504 bytes
# Number of multiprocessors: 12
# Number of cores: 96
Cuda error in file '..\cuda/cutil.h' in line 305 : out of memory.
Memory usage: host: bytes device: bytes
Assertion failed: 0, file ..\cuda/cutil.h, line 305

This application has requested the Runtime to terminate it in an unusual way.
Please contact the application's support team for more information.
----

TIA for any help.

My conclusion is that there was a change in GPUGRID WUs and the problem is not on my side. Here is my typical crash:

http://www.gpugrid.net/result.php?resultid=410378

Nope. These WUs haven't changed at all...

i