Fusion Edits Crashing on Render, 32GB not enough?

Hello all, I'm new to exploring Fusion, and I'm attempting what seems to be a pretty demanding effect from a tutorial I was trying to follow along with:

Youtube: Morbius Smoke Trails by Nomad R Productions (links not allowed)

I can assume since the default value of the particle generator is 100, it's perhaps pushing things a bit when the tutorial author sets it to 1000, lol. This definitely brings my machine grinding to a halt:

Dell XPS 15, 2023
Windows 11 Pro
32GB RAM
Nvidia 4070

I went and added a boatload of extra VRAM to Windows, but I don't think that helps very much. I reduced the particle count to as low as 200, but it still chokes my laptop to death. I can preview individual frames and make adjustments even though it takes a long time, but when I go to export the 600-frame clip, Resolve simply counts up the expected render time until it inevitably crashes.

I've uploaded a screenshot of my RAM usage as the laptop is trying to encode the file en route to crashing - it's spooling up gigantic amounts of memory. Is this just a really inefficient way to go about achieving this effect? Is it using Fusion improperly? Or is my laptop simply not up to the task? I'm super new to this kind of graphics-intensive work, so please forgive my ignorance, and any guidance would be greatly appreciated!

Kindly,
Tristan

Update: having tested this further, reducing the pEmitter particle count to 100 definitely helps the laptop stay in the game a little long, but not by much. Currently, it's frozen showing RAM utilization at 70% - 22765MB on the Fusion page; I couldn't even get to the export tab.

Is this normal? I've seen another similar tutorial use a pEmitter value of 3000; do these folks just have god-like computers, or what? If there's something I have configured or settings that aren't correct on my end, I would love to learn!

Or can my laptop simply not handle a pEmitter node?

Many thanks!

Right click on the pRender node and select "cache to disk". Select some place with enough room for 600, 13Mb files and select the cache button at the bottom of the window. I forget what it exactly says. This will pre-render just the particle set up and keep it from over loading when you go back to the edit page. You can also put down a Saver node and render the pRender to disk. Then add a Loader and load in the flatbed sequence directly after in the flow. This is the preferred way to cache something complex.

Right click on the pRender node and select "cache to disk". Select some place with enough room for 600, 13Mb files and select the cache button at the bottom of the window. I forget what it exactly says. This will pre-render just the particle set up and keep it from over loading when you go back to the edit page. You can also put down a Saver node and render the pRender to disk. Then add a Loader and load in the flatbed sequence directly after in the flow. This is the preferred way to cache something complex.

Oh mah gods Ben, thank you. I literally spent almost 10 hours yesterday relentlessly crashing Resolve trying to figure what to do. Caching the pRender node worked perfectly; after I did that, I was able to alter and preview lots of parameters at a very acceptable rate.

Would you be willing to explain what's going on there? If sending the pRender node to cache means it doesn't have to re-load it, why am I able to change some of the parameters that it's rendering so easily now, like the particle count on the pEmitter node? Or even some of the pRender settings, i.e. Blur, Glow, etc? I would expect I would need to cache that new output all over again, but that doesn't seem to be the case. I don't think I fully understand how the caching system works. Or if you know of a resource that would explain it, I'm all ears!

Also, I'm not quite following the Saver-Loader workflow - is rendering the pRender node to disk different than caching it? I couldn't find an option for that, and loading the first of the chached sequence into the Loader node didn't seem to do much. Apologies for the hand-holding, as I mentioned, this is the first time I'm encountering all this info. You've already helped a ton, thank you!

Kindly,
Tristan

I'm glad it got the results you were hoping for! Fusion is mostly a CPU and RAM page. There are some GPU enhancements but not a lot. When you turn on a node's cache, it starts saving processed frames to disk. To get to that point the pEmmiter runs through the frame count in the pRender and preprocessed all the particles it's going to need. This can take a long time. Once that's done, the pRender will process the frames. I didn't know why it's easier to tweak settings on the pRender after caching. Maybe there is some glitch in Resolve that loads up the memory of something. I didn't think I've NOT cached a pRender in five years.

The Saver Loader thing is a technique to correct problems with Fusion's caching system. Often Fusion will just ignore a cache on a node. It can happen for a lot of reasons. At the end of the day, it's annoying and I don't want to trouble shoot something like that. So to counter that, you add a node called a Saver. This will allow you to save out an image sequence as exrs or other formats to disk. To make the Saver render, you select"render all Savers" from the Fusion menu at the top. It's important to disable a Saver after you've used it by hitting Ctrl p. Otherwise when you use that drop down command, it will re-render any saver that's enabled over again. It's a lot like how the cache works except that it doesn't auto load the image sequence generated into the same node. When you change things on a flow with a cache, the changes will make the cache overwrite itself or blow it out completely if you cancel it for some reason. If you use a Saver, it's in a folder and untouched until you re-render it. Then directly after the Saver you add a Loader node and select the image sequence that you made with the Saver. So your original flow will be broken off but you'll have the Loader now connected where the cached pRender would have connected. The main difference being that you have complete control over when and how the cache is written. Now you can also use the same sequence in multiple locations without any lines going everywhere. There's more you can do to make this stuff easier to handle but that's a big one.

Fusion is great but you either have to have a monster system or a healthy about of disk space so you can cache everything you need. My last project had about 300Gb of caches. When I move the project, I'll have to move them with it or recache everything.

Saver-Loader Test.png (969.47 KiB) Viewed 109 times

I'm glad it got the results you were hoping for! Fusion is mostly a CPU and RAM page. There are some GPU enhancements but not a lot. When you turn on a node's cache, it starts saving processed frames to disk. To get to that point the pEmmiter runs through the frame count in the pRender and preprocessed all the particles it's going to need. This can take a long time. Once that's done, the pRender will process the frames. I didn't know why it's easier to tweak settings on the pRender after caching. Maybe there is some glitch in Resolve that loads up the memory of something. I didn't think I've NOT cached a pRender in five years.

Yeah, that's a huge help, thank you so much! I had zero chance of figuring that out on my own, LOL. So, just to make sure I'm understanding correctly, with the Saver-Loader workflow, you get your settings in all nodes up to and including pRender to where you want them, and then cache the pRender to keep your machine from dying. To improve performance even more, we then:

1. add a Save after pRender and export that to disk
2. disable that Save node so that it doesn't re-export should we repeat this process later with another scene
3. add a Loader after the Save which automatically populates with its own Merge node, and select the output of the Save node to be loaded

If I'm understanding correctly, this basically caches the output of everything up to the Loader, so that none of that needs to be processed again, is that right? Then everything downstream of the Loader can be adjusted with less of a compute hit?

In the little test case I'm working on, I changed the Blur settings post-Loader and that worked well. Then just to see what would happen I added a Glow after the Blur node and Resolve immediately crashed. I've uploaded a screen cap of the node structure for the project, if you have the interest, could you tell me if I'm understanding/implementing this correctly?

Also, please only invest the attention you care to - you've already helped a ton and I really appreciate your time!

Kindly,
Tristan

PS - I tried to reply sooner, but kept encountering errors because I'd included contraband emojis in my message.