cuda problem with latest fedora updates (march 08) strange behavior and system hung

Accelerated Computing CUDA CUDA Programming and Performance
1

hi guys,
i hope this is the right place for bug reporting. i didn’t find any bugzilla links on nvidia site.

i have a laptop with fedora 8 (x86_64), nvidia driver x86_64-169.12 and cuda toolkit 1.1 installed

the hardware is a dell vostro 1700 with a geforce 8600m gt with 256mb ram.

all the toolkit examples worked until a week ago (i didn’t run them every day so i’m not sure
which software updates introduced problems).
today i just discovered that even the simple alignedTypes does not work every time i launch it.
some time the system hang or the program end with the message TEST FAILED and the
X session become dirty (i see many random black pixel like a spray effect all over the desktop session).
if i reboot the system (cold reboot) i can rerun the toolkit examples but they work only for the
first 3/4 run. then i hit problems.

i would be pleased to post much more detail if someone can help me to run the toolkit examples
in a more verbose mode or help me to collect debug infos.

many thanks in advance
Luigi

ps: in attach dmesg and rpm -qa outputs
nvidia_bug_report.log.txt (130 KB)
rpm_qa.txt (14.6 KB)
dmesg.txt (32.1 KB)

2

Please generate and attach an nvidia-bug-report.log.

3

done. thank you.

i opened this post not just because i want to play with toolkit examples but because while writing a simple application i found strange behavior of cuda.

my program compute the md5 on the gpu but while hunting for a bug i found that cuda continued to execute an “old” kernel even if i did significative changes and recompiled a new binary.

i even find a case where a soft reboot didn’t clean properly the video card ram/settings and cuda continued to execute and old kernel (one of before the reboot!).

to solve that situation i had to do a cold shutdown and restart. i was starting to think i was crazy but then i realized that even the toolkit starts to behave strangely.

thanks again

Luigi

4

hi,
i have good news! i just reinstalled the Nvidia driver and all problem with the sdk went away. it seems that i did an xorg server update and i broke some link created during the nvidia driver installation.
In fact i usualli run sh NVIDIA-Linux-x86_64-169.12-pkg2.run -K to recompile the kernel module after a linux kernel update but i didn’t thought about links and library dependencies.
so the solution is to run a normal driver installation (even the same version) to overwrite the xorg default links.

now i can go back to my bug hunting :P

bye bye
Luigi