Like a lot of people, I’ve been taking a look at Fusion again now that Blackmagic Design has purchased Eyeon and will be giving the app the development and support it’s been needing for quite a while. As a compositing application, it has a very reasonable learning curve, with major concepts being easy to pick up in a few hours, while still having the depth and power required for high-end visual effects and motion graphics work.
Something that’s helpful to pin down with any creative application is a set of optimal settings that allow you to iterate your ideas quickly and view your work in context. Fusion isn’t exactly “real-time” per say, but it can get pretty close if you’re willing to experiment a bit. I spent the better part of a day working through the Fusion user guide looking for ways to improve Fusion’s performance, and found a number of different tools you can take advantage of depending on what you need to do.
Here is a detailed breakdown on the preferences I’ve found to work well. I’ve also included some notes on the system resources Fusion heavily relies on, and hardware changes you can make that will improve performance the most.
Enable Local Caching of Loaders: On
Cache Multi-Frame Files: On
Cache Location: Set to a Local SSD or SSD RAID Volume.
Limit Caching to 70% if Physical RAM
Simultaneous Branching: On
Render Frames at Once: 10 (Adjust Depending on Number of CPU Cores)
Simultaneous Branching: On
OpenCL Tools: Auto
Device: Select, Then Choose GPU.
Render Previews Using Proxy Scaling: On
When checked, Fusion will scale the images down to the preview size at the loader and creator tools. This allows for much faster rendering. If this control is de-selected, frames will be rendered at full size, then scaled down.
Skip Frames To Maintain Apparent Framerate: On
When checked, Fusion will skip frames during playback of flipbooks and file sequences to maintain the framerate setting.
Enable Direct Reads: On
This checkbox enables Direct Reads of data from the drives into memory, using a method that is more efficient when loading a large chunk of contiguous data. The advantage is that the CPU load is reduced somewhat for I/O operations. The disadvantage is that this feature employs an undocumented ability of the operating system and may have other effects on performance that are unknown.
Read Ahead Buffers: 4-8 (Test on your system)
This slider determines the number of 64K buffers Fusion will use to read ahead in a file I/O operation. The more buffers, the more efficient Fusion is likely to be at loading frames from disk, but the less responsive it will be to changes that require disk access interactively.
Use float16 Textures: On (If your graphics card supports it)
Fusion Performance Optimizations
RAM Caching of Loaders and Effects is Fusion’s strongest performance feature. Other than increasing your CPU clock, you will benefit more from additional RAM than any other single hardware upgrade. Fusion can cache to formats as large as 32-Floating Point Files, which can use up memory quickly. 32GB of RAM minimum, with 64-128GB or more being optimal. Large compositions several seconds long with multiple loaders can easily use up all of it.
Highest clock speed possible, with 3.0GHz or faster being optimal.
6 Cores is optimal. More can be beneficial, just don’t add them at the expense of clock speed.
Fast memory access, at least 1333MHz, with 1600MHz+ and beyond being optimal.
Nvidia GPU with as much VRAM as Possible, with 4GB or more being optimal.
Good options according to various guides are the GTX Titan Black, or the Quadro K5000/5200 for Stereo 3D Work.
Fusion benefits from fast storage, but not as much as additional RAM, so prioritize RAM over storage in your hardware budget. It’s good to have at least one SSD as a disk cache location for Image Loader nodes in Preferences > Loader > Cache > Cache Location.
Possible Purchasing Strategy: Buy 64GB of RAM, then one 256GB SSD for cache. If additional funds are available buy up to 128GB of RAM, then use any additional funds for more disk storage.
Do not spend a lot of money on a multi-SSD RAID array. A two-SSD RAID0 stripe is plenty for disk caching. Spend hardware budget elsewhere like more RAM, faster CPUs, or an Nvidia GPU with more Vram.
Image Loader Nodes (“Loaders”)
The two best “free” steps you can take to help maintain as much real-time performance as possible during playback.
1) Cache loaders to disk right away.
2) Set good In/ Out Points in Loader settings to eliminate un-needed rendering and disk cache usage.
Caching a Loader Node to Disk
Right-Click Loader node.
Select Cache to Disk…
Verify Cache Location is set to your correct cache volume.
Verify Enable Cache is checked On.
Setting Loader Node In/Out Points
Single Click Loader Node.
In Loader settings panel to the right, set the Global and Trim IN/Out points so they include only the the part of the image file you need.
Specifies the In-Out position of the clip within the length of your composition. (In an NLE this would be the RECORD In and Out points).
Specifies the IN-Out points used from the clip itself. (In an NLE this would be the SOURCE In and Out points).
If you change the IN-Out points for Trim, you will probably need to Right-Click the Loader Node and Cache to Disk again.