2025-01-23, 07:23 AM
(This post was last modified: 2025-01-23, 07:27 AM by nyanmisaka. Edited 2 times in total.)
> This varies a lot depending on the input and in reality it barely does one 4k30 for most of the time (two will stall the stream from time to time). I might test some lite implementation that trade off flexibility for such slow GPUs to have more useable tone mapping performance.
Probably because my test stream does not contain DOVI P8 metadata, reshape kernel is expensive on this ~0.5 TFLOPS GPU.
> I don't think we really need VPP hardware for this SoC as its GPU looks fast enough for most of the processing tasks to me, at least they are fast enough on paper.
Let's wait and see. This new GPU should be 3 or 4 times faster than the previous GPU used on the RK3588.
Probably because my test stream does not contain DOVI P8 metadata, reshape kernel is expensive on this ~0.5 TFLOPS GPU.
> I don't think we really need VPP hardware for this SoC as its GPU looks fast enough for most of the processing tasks to me, at least they are fast enough on paper.
Let's wait and see. This new GPU should be 3 or 4 times faster than the previous GPU used on the RK3588.