Thanks Roeland,roeland wrote:What is the first error shown?grimm wrote:I think I spoke a bit too soon. The only ocs file I have gotten to work has been the benchmark. I tried all of the other demo files and they all would start to work for just a couple of seconds before stopping. The errors are the same as before, here is a sample:
Maybe a graphical desktop indeed suffers more from running a CUDA application. Is it running much slower than with version 3.01? Were you using the PMC kernel back then? This kernel has less effect on the responsiveness of the desktop than the path tracing and direct lighting kernels.I haven't been able to use Octane much this summer so I think it must have been 2.58 or thereabouts. "less responsive" is, when this version is working, things like button clicks take a second or two to register. Panning or rotating the camera is very slow and the rest of the interface is just not as snappy as it was before.
The only differences that I can think of from then to now, besides the OS upgrade and the new version of Octane, is that I have changed my desktop environment to Cinnamon. Cinnamon is a fork of Gnome 3 that now uses a lot of graphic effects that might affect the GPU more than before? The only other thing is that I had to do is a link to the libGLEW library to get Octane to run. Even though I'm running Fedora 17 the libGLEW in the repository is only up to version 1.6. So I linked /lib64/libGLEW.so.1.7 to the 1.6 version. Sometimes this works, other times not. In any case it might be an issue?
We will try to remove the libGLEW dependency. I don't know if linking like this is a problem.
--
Roeland
Here is the start of one of the sessions, don't remember which one I was trying here:
Code: Select all
CUDA error 700 on device 0: An exception occurred on the device while executing a kernel. The context cannot be used anymore and must be destroyed. All existing device memory allocations from this context are invalid and must be reconstructed.
-> Kernel execution failed (mprev)
CUDA device 0: Preview failed
CUDA error 700 on device 0: An exception occurred on the device while executing a kernel. The context cannot be used anymore and must be destroyed. All existing device memory allocations from this context are invalid and must be reconstructed.
-> Failed to load symbol data to the device (kernel data)
The slow-down is very noticeable, I tried daylight, path tracing, and pmc, they were all very slow and unresponsive. The samples per second appeared to be the same from what I was getting before. Jason
