uname -a output: Linux iceraven 3.14-2-amd64 #1 SMP Debian 3.14.15-2 (2014-08-09) x86_64 GNU/Linux (jessie)
CPU: Intel i7-4700MQ
GPU: NVIDIA 780M
Desktop Environment: GNOME 3 (latest)
GPU Driver: nvidia-driver package on jessie repo (340.29)
CUDA Driver: nvidia-cuda-toolkit package on jessie repo (5.5; just stopped working as support for pre-6.0 just dropped)
Broken: Latest Dev Version (2.71.6, Hash: 24ea3ab)
Worked: None yet. Testing 2.71 stable as I'm typing.
I have a complex scene where a bullet impacts a coffee cup, with coffee inside, that goes everywhere. It involves a large domain, high resolution fluid simulation, and hundreds of collision objects that are keyframed from a Rigid Body simulation.
At one point this worked flawlessly. However, now it merely stops once reaching enough complexity. Here's what I've been able to gather:
- Lowering the resolution makes the simulation continue baking a bit longer, but only the lowest resolutions (120-) actually complete.
- top tells me that the CPU is being used plenty. The RAM usage is well within what should work.
- Blender's memory usage is never above 600M, according to itself. top agrees; however I'm using GPU for rendering, so this could have been an issue.
It probably means nothing, but I uncommented the Experimental GPU features in kernel_types.h before compiling Blender.
--Reproducing the error:
1. Load file.
2. Set offset in the fluid simulation to 25.
3. With the resolution at 425, attempt a bake.
4. The bake should, without any kind of error, simply stop, uncompleted, after 50 frames
NOTE: In this file all the keyframing has been moved back 27 frames, so you don't have to wait more than neccessary for a complex high-res bake.
--Filesize: 25MB. https://docs.google.com/uc?id=0Bx0YaKfeuWSrUnFScDAtLVdtU2s&export=download
Thank you in advance!