Their FAQ page lists answers to most questions one may have about the service. I've uploaded an example of a project which includes the baked cache for particle systems, on Github, here. As mentioned in their FAQ, the steps to do this are:
1. Make all paths relative - in Blender, choose
File --> External Data --> Make All Paths Relative
2. If you need to render a baked fluid/particle system, zip the blender file and the cache directory, ensuring that the cache directory is also set as a relative path. In my case, I had set the cache directory to be inside the project directory to simplify matters. In the project above, this is located in the Physics tab of the Smoke Domain object as seen in the screenshot below. As you can see, the current directory is indicated by a leading double-slash - //.
3. One point to note is that choosing File --> Save As and saving the project to another directory, and then copying the cache directory there, does not work (as in Blender 3.2 on Windows.) This is probably because the file paths get mangled? Anyway, if you want to break up the project into multiple renders due to the bake files being too large, copy-paste the project outside Blender instead, don't use File --> Save As.
4. There is a script available on the Forums for splitting the project, in case you want to automate the process. Or, like I did, you can just choose to bake all, and then copy only the .vdb files in the data subfolder of the cache for a limited number of frames. For the zoomed out view, for frames after 275, this was something like 30-50 frames per project. And choose to render only those frames. Then create a fresh zip file for the next set of frames. And so on.
Here are some project statistics, for the zoomed in project render, which was taking around 3 minutes per frame on my machine:
Summary
- Storage used: 971.8 MB
- Cumulated time of render: 9h01m
- Points spent: 5,767 points
- Real duration of render: 1h23m
- On reference per frame rendertime: 0m51s
On my machine, the 600 frames would have taken 30 hours to render, and even on the Studio Mac it would have taken 5-6 hours, but SheepIt finished it in less than 90 minutes due to the power of distributed processing.
So, nowadays I leave the sheepit client running whenever I switch on my machine, to accumulate points. A borrowed CPU from VV fitted with the GTX1050 Ti card which I purchased for my research also helps at the office. The 1050 Ti card is only 4% the speed of the standard GPU, but since most of the projects specify GPU, it usually gets some project to render, as against the i5-1235U laptop CPU, which often goes many minutes with "No jobs to render".
Statistics of the three client machines:
| CPU | Intel(R) Core(TM) i7 CPU M 620 @ 2.67GHz x4 |
|---|---|
| RAM allowed | 1.3 GB |
| RAM available | 3.9 GB |
| Max render time per frame | |
| Power CPU | 6 % |
| Scheduler | very_slow_computer |
|---|
------------------------------
| CPU | Intel(R) Core(TM) i5-4430 CPU @ 3.00GHz x4 |
|---|---|
| RAM allowed | 4.8 GB |
| RAM available | 8.3 GB |
| Max render time per frame | |
| GPU | GeForce GTX 1050 Ti |
| VRAM | 4.3 GB |
| Compute device | CPU, GPU |
| Power CPU | 34 % |
| Power GPU | 4 % |
| Scheduler | Default |
------------------------------
| CPU | 12th Gen Intel(R) Core(TM) i5-1235U x12 |
|---|---|
| RAM allowed | 2.9 GB |
| RAM available | 8.1 GB |
| Max render time per frame | 1h |
| Power CPU | 72 % |
| Scheduler | Default |
And finally, here's a youtube video of the rendered output. This is a "Fulldome" video, meant for projection in a planetarium. (Since this is my first effort, the launch needs more work to look more realistic - currently the motion of the rocket looks a bit cartoonish.)
No comments:
Post a Comment