I’ve been trying out Topaz on a couple computers:
2019 Intel MacPro
2021 M1 Max MacBook Pro
Custom PC with a Threadripper 1920X and 24GB RTX Titan
The PC has the best benchmarks by far. But I have not found those numbers to be relevant in real world performance. The M1 Mac consistently has the best times in everything I’ve run with real footage. I’m mostly upscaling old 480i footage to 1080p using Iris.
The PC’s benchmark with Iris 2x was 5.51 fps and I was getting about 3.6 on an actual render.
The M1’s benchmark with Iris 2x was 1.39 and I was getting about 4.5 on an actual render.
Has anyone else had similar experiences with benchmark accuracy? Are there any obvious things to change that might explain the problem?
For what it’s worth, the Intel MacPro was the worst in both benchmarks and real world performance.
The benchmark measures the speed of the AI engine running on your hardware, but does not simulate the additional workload of decoding an input video and encoding an output video. This is the most likely reason for the difference between benchmark and real performance on your Windows machine.
For macOS our team is currently looking into some improvements that can be made for Iris performance, but it’s also possible that the size of the input video is causing the model to run a different scaling mode when exporting vs. benchmarking.
I can check the app logs to see if Iris 2x is being utilized during exports:
To gather logs, please select Help > Logging and make sure that there is a check mark next to File Logging. Next, recreate the issue and then return to the Help menu and select > Get Logs for Support and attach the “logsForSupport.tar.gz” file to your reply.
Is ProRes a factor? I’m using ProRes for both the source and output file. Both because it’s a format I tend to work in and because it’s the lightest compression available in the output.
Everything I use Topaz for goes to editing anyway so big files are completely fine. I wasn’t sure if ProRes on Windows was somehow slowing it down more than Mac.
