CPU/GPU recommendations: many static parts, few moving parts

I'm new to Fusion and have to upgrade my computer, and I'm asking for advice on what level of CPU/GPU I'll need.

The 4K videos I'm making, I use Fusion to 'draw' many dozens of static parts. Below is a sample of 4 static parts, but in my videos there will be up to 50 parts. (In the videos, of course the parts will look different).

examplefusion.jpg (195.99 KiB) Viewed 1186 times

In the 4K videos, at any given instant, perhaps one or two parts might move slightly. But for most of the video, the parts are static and not animated.

Throughout the video, which could be 10-60 minutes long, these static Fusion elements are present throughout. But the actual movement/animation of parts happens sporadically (often to illustrate an idea in the dialogue). A new 'scene' of Fusion parts would occur about every 1 minute through the hour video.

I'm new to Fusion, and my present late 2015 iMac's i5 3.2GHz CPU, 2GB GPU is totally inadequate.

My question is, is what I am doing considered a low or high load on the CPU/GPU for Fusion? Note that the Fusion illustrations, even though static for most of the time, are visible throughout the potentially a 1 hour video. Each Fusion 'scene' changes to a new one say every minute throughout the video. So there could be several new Fusion compositions coming into play every minute, with a handful of parts animated.

For the sake of calculation, there could be 180 Fusion compositions spread across a 1 hour video (since I could split the diagrams into many Fusion compositions).

For this manner of videos I'll be making, I'm not experienced enough with Fusion to know if this is regarded as low or high workload for the CPU/GPU. Do I need something like a Ryzen 9 3950X or 3900X, and RTX 2080 TI? Memory 64GB?

Does the fact that most of the objects are static mean it is a low-load for Fusion, or the fact that several Fusion compositions are being displayed, changing each minute, mean it is a high-load Fusion task?

Fusion can relatively easily handle large numbers of shapes of that kind, and the length doesn't really matter to it—it doesn't care if a timeline is 10 frames or 10,000 in terms of the resources it requires. On the other hand, it also doesn't give you much benefit from re-using pixels from frame to frame. It will always calculate the entire image on every frame.

So if you have long periods where the shapes don't move at all, it would be most efficient to replace the Fusion comp over those times with a single still frame, since I imagine Resolve will cache that much more efficiently than having a live comp present at all times.

Prior to version 16, Fusion got only minimal benefit from GPU acceleration, but now it gets a significant boost. I'm not sure if anyone's done any benchmarking of v16 yet or not, but just my brief experiments make me feel as though it's taking much better advantage of my GPU than v9 does.

If you're trying to prioritize where to spend your money, Fusion and Resolve will both benefit the most from fast access to media, followed by as much RAM as you can get, then a GPU—prioritize VRAM size over clock speed, then CPU—priority should go to number of cores over clock speed.

CPU—priority should go to number of cores over clock speed.

https://www.pugetsystems.com/labs/artic ... 950X-1616/

In this article from Puget systems, they found that, for Fusion, the higher base clock speed was better for Fusion than number of cores. What are your thoughts?

fusioncores.jpg (121.5 KiB) Viewed 1049 times

I'd trust someone who's actually done a benchmark over somebody who's shooting from the hip (me).

Bryan is right, RAM is more important then almost anything else. A 3960X and 3900X have the same clock speed, but the 3960X has twice the RAM capacity (and twice the memory bandwidth and twice the core count).

I think it's bizarre that Puget would test a system with a 24GB GPU, but only 64GB of RAM.

Also, Puget only tested renders, not interactivity. A workstation should be interactive, a render farm renders.

Just something to keep in mind.

For this particular use case, Fusion will set the cache validity range for static elements, so it's not like you'll need to render every frame. You only need to render the frames that are different (as much as Fusion knows they are different, it doesn't look at the pixels or the values of inputs, but whether or not the inputs are changing).

Bryan is right, RAM is more important then almost anything else. A 3960X and 3900X have the same clock speed, but the 3960X has twice the RAM capacity (and twice the memory bandwidth and twice the core count).

I think it's bizarre that Puget would test a system with a 24GB GPU, but only 64GB of RAM.

Also, Puget only tested renders, not interactivity. A workstation should be interactive, a render farm renders.

Just something to keep in mind.

For this particular use case, Fusion will set the cache validity range for static elements, so it's not like you'll need to render every frame. You only need to render the frames that are different (as much as Fusion knows they are different, it doesn't look at the pixels or the values of inputs, but whether or not the inputs are changing).

The AMD Threadripper 3960X is probably out of my price-range.

From what you've said, for the Fusion and video editing I'm doing, how would you choose between the Ryzen 3900X and 3950X? Or would you recommend me something else, even in the Intel range?