Am intrigued by GPU based renderers like Octane or fryrender and would love to test them out. Unfortunately I have a history of trashing my linux systems during even routine upgrades.

I currently have only one workstation… so if I were to go through the trouble of installing the CUDA 3 drivers, new linux libs, and who knows what else (it looks like a lot) just to test these renderers out… will Houdini still function properly on such a system?

Thanks,

Floyd

I,m on ubuntu 10.10, with cuda 3.Octane and Houdini runs with no problem

Thanks. Good to hear.

Will now try to work up the courage and take the plunge.

Cheers.

BTW, octane keeps crushing with default nvidia drivers, so i also installed latest drivers from nvidia.

Hi Floyd,

Installing Cuda is quite non-invasive, it just goes in /usr/local and only gets used when you add it to LD_LIBRARY_PATH and PATH. So, pretty easy to isolate.

3D Coat uses CUDA on Windows and Linux, but I have not yet successfully gotten Linux to work, likely the combo of CUDA that 3D Coat expects vs the OGL drivers I have on the machine.

Cheers,

Peter B

Hi Peter,

I've been trying to install this for a few days now.

Up-to-date Debian Lenny (2.6.26-2-amd64)
quadro FX2700M
nvidia driver 260.19.44

CUDA 3.2 toolkit and SDK fail during extraction.

CUDA 3.0 toolkit and SDK run (extract) fine but fail to make the binaries…. with a lot of “warning: type qualifiers ignored on function return type” errors

Paths are fine.

That's as far as I've got so far… and my eyeballs hurt from all the readme's and related sites I've visited.

A bit frustrating cuz this GPU rendering stuff looks very cool. Should probably post on some of the appropriate sites to get some help.

One bright spot… I haven't trashed my OS yet, and everything else seems to be working fine.


edit: Never was able able to install cuda3.2. Went with cuda3.0. Cuda SDK stuff never compiled, but was able to run OctaneRender… after installing libglew-dev stuff. Have only played with OctaneRender for a few minutes… my card doesn't have much horsepower. Looking forward to spending more time with it.

Finally got the cuda3.2 toolkit and SDK to install.

Regarding the extraction fails… I changed (vi) the TMPROOT variable at the beginning of the cuda run files to another (larger) location on my system and then they ran fine. Strange… had no problem with the cuda3.0 extractions. Perhaps my /tmp directory was too small?

Even with cuda3.0 I was never able to compile the binaries that came with the SDK. Did a few things differently when installing cuda3.2… not sure if one or all of these solved the problem.

- apt-get install libxmu-dev - had only installed freeglut3-dev, libxi-dev & libglew-dev previously.

- Did NOT set the LD_LIBRARY_PATH as suggested in several locations… instead created “/etc/ld.so.conf.d/cuda.conf” and added the lines /usr/local/cuda/lib64 & /usr/local/cuda/lib to it. Some instructions say to add these paths to your /etc/ld.so.conf file… but on my system that file already had the line… “include /etc/ld.so.conf.d/*.conf”… so it reads any .conf files in the ld.so.conf.d directory. Has the LD_LIBRARY_PATH env been deprecated?

- Also did the blacklist bit recommended on many sites… not sure if this was necessary in my case…
Created /etc/modprobe.d/blacklist-nouveau.conf and added the following lines…
blacklist nvidiafb
blacklist nouveau
blacklist rivafb
blacklist rivatv
blacklist vga16fb
options nouveau modeset=0

Cuda3.2 binaries compiled. deviceQuery runs and passes. SDK sample files run. Oh… and Octane Beta2.46 runs!

Here are some of my other system details…

Debian Lenny 64 bit | Quadro FX2700M | NVIDIA-Linux-x86_64-260.19.44

Cheers.