[Mageia-dev] NVIDIA CUDA 5 has landed

[Dimitri]

Hi,

Finally, NVIDIA CUDA Toolkit 5.0.35 has landed in Cauldron. I apologize
for the delay, and ask everyone interested to test the packages
(especially on x86_64).

New in packaging:
=================

* From now on, we are using SLES official packages, because others
(Ubuntu, Fedora) contain cuda-gdb binary that is linked against
libtinfo.so.5 that we do not ship;
* those lame /usr/{nvvm,open64,libnvvp,libnsight} are moved to the
appropriate place, namely %_libdir;
* samples (ex-CUDA SDK) have been included.

New in CUDA 5.0:
================

* NVIDIA Nsight, a full-featured CUDA IDE, based on Eclipse CDT;
* dynamic parallelism - simply speaking, ability to spawn new kernels
from kernels;
* more here:
http://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html

Known issues:
=============

* GCC 4.7 is not supported. However, you can fool nvcc and compile a
program with -D__GNUC_MINOR__=6;
* In Eclipse based products, you may notice that toolbar buttons don't
respond to mouse hover (look flat all the time). This is a well-known
issue and will be fixed in the upcoming release of
gnome-themes-standard;
* Eclipse-based products are unable to open web browser due to crash in
libxul.so (should we update xulrunner?)

The work on the package is still not finished. The package contains an
init script (nvidia) that creates device nodes and loads kernel module.
This "service" is intended to be started on GUI-less compute nodes. I
want to ask for assistance in migrating this script to systemd unit. As
far as I know, with latest kernels and NVIDIA drivers the device nodes
always get created automatically (never noticed them not being
created). So probably the only thing left is to load kernel module.
Seems like putting it into modules-load.d/* is a bad idea, because if
the module is absent, systemd-modules-load.service will fail. Should we
simply make a unit with ExecStart=/usr/sbin/modprobe nvidia-current instead?

Also I'd like to thank Pascal Terjan who helped me to upload a part of
source. Uploading another 700 MBytes took me 2 hours - that was painful.
I remember there was a proposal to provide packagers with their
dedicated VMs with shell accounts on a build cluster - what's the status
of that?

Regards,
Dimitri