Matthew Riggieri
Well-known member
So while I think this is probably badass and much needed on lower core count systems, it kinda has the opposite effect for High core count systems.
Example
I have a Mac Pro 7,1, 16 core with Dual Radeon VII (aftermarket, not MPX).
For 6K and under, it makes more sense for me to do CPU decode, and leave more GPU overhead for color transforms, blurs and NR. In 7K and 8K projects, it is pretty sweet to hit 24-40 FPS full debayer decode to 4K Prores. But it uses such a large amount of GPU resources, that with just a tetrahedral LUT transform and some basic corrections, I am already maxed out on GPU resources. Not enough overhead for full grades.
Hopefully some further improvements and BIG Navi will make this a moot point. I Think Puget system already came to this conclusion with CUDA decode.
I wish the switch from CPU to GPU decode did not require a restart of Resolve.
Example
I have a Mac Pro 7,1, 16 core with Dual Radeon VII (aftermarket, not MPX).
For 6K and under, it makes more sense for me to do CPU decode, and leave more GPU overhead for color transforms, blurs and NR. In 7K and 8K projects, it is pretty sweet to hit 24-40 FPS full debayer decode to 4K Prores. But it uses such a large amount of GPU resources, that with just a tetrahedral LUT transform and some basic corrections, I am already maxed out on GPU resources. Not enough overhead for full grades.
Hopefully some further improvements and BIG Navi will make this a moot point. I Think Puget system already came to this conclusion with CUDA decode.
I wish the switch from CPU to GPU decode did not require a restart of Resolve.