Jump to: Board index » General » Fusion

Cache to disk vs Rendering

Learn about 3D compositing, animation, broadcast design and VFX workflows.
  • Author
  • Message
Offline

ResolveNoob

  • Posts: 85
  • Joined: Wed Jul 31, 2019 1:42 pm
  • Real Name: Tumelo Mogale

Cache to disk vs Rendering

PostWed Jun 22, 2022 8:24 pm

I have been experimenting with the depth map and while I think it's a very powerful tool, it runs quite slowly even on a good system.

In my recent project I tried to cache the node to disk but not only did this take forever to finish caching, it didn't make a significant difference to the playback when it was done. I ended up just rendering out the depth map (in the deliver page) then bringing it back into fusion.

I just want to find out if this is everyone's experience with the cache to disk function? Generally speaking, is it better and faster to just render things out if it's something like a depth map?
Offline

bentheanimator

  • Posts: 431
  • Joined: Mon May 13, 2019 10:38 pm
  • Location: Minneapolis, MN
  • Real Name: Ben Hall

Re: Cache to disk vs Rendering

PostThu Jun 23, 2022 3:40 am

The short answer is yes.

The longer answer involves a lot of complaining about how the caching system in Fusion works/doesn't work. There are a lot of things that can and do go wrong in Fusion because of how much back and forth with the GPUs occur. Sometimes nodes won't let go of cached frames from RAM, sometimes is never caches it at all because the GPUs are kicking in just right. Then you'll do something like add a particle system and it eats your GPU for lunch. You go back to your flow farther down and now what was playing back fine is a stuttering mess. One way to solve this is to turn off all GPU calculations. I don't think I've ever done this but it would at least eliminate the GPU caching issue by way of cutting off your proverbial nose to scratch an itch on it.

So, you can cache any node. But should you just cache any node or should you set up a known system that you can identify three months from now when you crack open the project again? Like everything in Fusion, it's up to you.

The first recommendation I could suggest would be to set a Boolean at the end of any flow you want to cache and name it "thisFlowCache". Then enable Cache to Disk and stuff the cache in a Cache folder in your project directory and name it the node's name. Like "\\project\fu\fuCahes\thisFlowCache\DiskCache0000.raw"
This will allow you to enable and disable cache as you like and use the faster ".raw" format. It's a way of not completely abandoning the idea of "Caching" in Fusion.

The second recommendation is to abandon the idea of Caching in Fusion. You can set up a practice of absolute rock solid principle... The Saver/Loader flow.

At the end of any flow branch you want to keep you set up a Saver. Then immediately after it you set up a Loader that loads in what was just rendered. All you have to do is disable the Saver and the branch is cleanly tucked away. Need to revise something? Re-enable the Saver and overwrite what you need to and then Disable again. Most will set up a folder in their project much like the caching system where you tuck away the Gigs and Gigs of exrs you make with this set up. It's heavy on the IO but it is rock solid and does work. You can also do tricks like use a Resize node right before the saver to make half size proxy versions in the Saver and then another resize to scale it back up after the Loader. Saves a lot of data and then you just go back and re-render at full size once you're ready.

There's even a script in Reactor to auto set up the two automatically.

Those are the core principals of what you asked. A depth matte is just a black and white map. The more bit depth it has the more data per frame you have to calculate. If you can get away with a 16bit depth pass nobody is stopping you. Might even be able to use an 8bit in some circumstances. The best and worst thing about Fusion is that it lets you hang yourself with your own rope if you chose to. Once you get the hang of it though, it gives you the ability to really make what you want.
Resolve & Fusion Studio 18.6.5
Windows 10
Intel Xeon CPU 2699A @ 2.40GHz | 128GB RAM | 2xRTX3090 | 512NVME System | 8TB NMVE Scratch | 80TB 8Gbps Fiber

MacOS 12.7.2
MacBook Pro 13,3 | 16GB | Radeon 460 4GB | 256GB System | 256GB Scratch
Offline

ResolveNoob

  • Posts: 85
  • Joined: Wed Jul 31, 2019 1:42 pm
  • Real Name: Tumelo Mogale

Re: Cache to disk vs Rendering

PostFri Jun 24, 2022 1:22 pm

Thanks so much for the comprehensive reply.

I tried the loader/ saver method in my new project and it works. I think think this is going to be my preferred workflow from henceforth whenever I'm dealing with heavy scenes.

Return to Fusion

Who is online

Users browsing this forum: No registered users and 39 guests