Recently I have switched to the Blender 2.66 version and I faced the problem with a CUDA kernel compilation.
When I render the scene 1st time, the CUDA kernel compilation process starts and finishes with an error:
ptxas info : Compiling entry function ‘kernel_cuda_tonemap’ for ‘sm_11’
ptxas info : Used 7 registers, 40+16 bytes smem, 1168 bytes cmem, 20 bytes cmem
ptxas info : Compiling entry function ‘kernel_cuda_shader’ for ‘sm_11’
ptxas info : Used 24 registers, 4968+0 bytes lmem, 16+16 bytes smem, 1168 bytes cmem, 644 bytes cmem
ptxas info : Compiling entry function ‘kernel_cuda_path_trace’ for ‘sm_11’
ptxas fatal : Memory allocation failure
CUDA kernel compilation failed, see console for details.
Do you have any clue how could I avoid this problem?
My system specification:
OS: Linux, 32bit, kernel version: 2.6.32-5-686-bigmem (Debian)
Graphic Card: GeForce 9600GT
nVidia driver: 310.32
I am aware that my graphic card is quite old, but until Blender 2.65a it works with Cycles on Experimental mode, and still gives me quite big benefits in rendering time (more than 50%-60% (GPU/CPU rendering time ratio) depending on scene), so if you have any suggestions how to force blender work with my card I will be very grateful