Pros: OpenCL is truly fast and convenient. The node system is traditionally convenient. Great curves and math. The first half of the build is pure pleasure. Here we give upvotes and hearts.
Cons: No timeshift, which is very bad; it's essential for clipping and shifting animations. A switch node is an analogue for layer clipping, but shifting the animation start or disabling animations is a dead end. Changing everything globally immediately puts us in the red and misses timeshift. Blocks and invoke are very useful, but how do we pass parameter values? We can easily pass millions of pixels, but passing a couple of numbers is a very non-trivial task. KKnowing OpenCL, Python, VEX, and HScript to be able to pass the number "5" as a parameter is a fundamental flaw. There's no motion blur, and creating a streak blur isn't possible either.
The "lamp" needs to be redesigned; I made several variations to simulate metal and good highlights. The glow is very bad; I also had to test several variations using blur and resizing, vacillating between speed and quality.
The equalizer is very slow for animation; I had to figure out OpenCL and do a two-step pass to calculate the min and max values, which resulted in an almost 10x increase over the native node and made it suitable for animation. It's a long process, but these are things we can fix.
The fundamental problem was where I least expected it to be.
Ratserize is very slow, but that's half the problem. There's already a solution, and we're just waiting for the next version. For now, let's save it to a file and read it. The cache in PIC format is only 80 MB, 1 frame is 600 KB. It's negligible. BUT the file node immediately dropped the FPS to 12-14 for PIC and 7-8 for EXR. SSDs deliver from 500 MB/sec to 3 GB/sec, dozens of processor cores, tens of gigabytes of memory, and tens of thousands of GPU cores. How can you read a file at 5-7 MB/sec? What's going on here? I honestly don't understand. We cache all output from "files," but the cache works very strangely. Sometimes it works, sometimes it doesn't. The performance monitor shows that the file node is eating up more than half the time, even though it's covered in caches like a Christmas tree with toys. But sometimes the caches don't work. The FPS fluctuates between 20 and 2. Although a couple of times I got the expected 60 FPS. No explanation. But with 5-6 caches for tiny footage from 8 MB to 80 MB with all frames, it was possible to get 1 frame every 3 seconds and consume all 16 GB of video memory. And clearing the entire cache, like in AE, doesn't seem to be possible. Basically, the File node is currently a solid tombstone for the idea of comfortably composing footage. What's the point of optimizing anything if the very first "file" sends the FPS to the bottom? And this second half has ruined all the fun, speed, and convenience of the first half. Alas, we can't say goodbye to AE just yet, but we eagerly await the next version of Copernicus. If they optimize loading and caching, it will be a blast.
Image Not Found
