As we've previously mentioned, the performance of Fast CinemaDNG software in most cases is limited by input/output bandwidth, not by GPU model. From that point of view, the SSD speed for read/write could be the bottleneck for RAW processing on GPU.
We've recently done some benchmarks for DNG to JPEG conversion on GPU. We've utilized the same 8-bit 12-MPix DNG image and very simple pipeline: raw image reading from SSD and loading to GPU, WB, demosaicing with MG algorithm, DCP profile, gamma, conversion to JPEG 4:4:4 and quality 100%, storage to SSD. On the NVIDIA GeForce RTX 2080ti we've managed to process 3000 images per minute, which we consider to be pretty fast.
That performance was limited by our SSD with 600 MB/s read/write capabilities. This is 50 fps to process 12-MPix input images, and this is exactly the result we've got (3000 images per minute). Actually, processing on GPU was around 3 times faster, so the bottleneck was with SSD performance, not with GPU. GPU load was around 30% for one-thread application, and that's why we see great potential for further acceleration of RAW-to-JPEG transform on GPU due to faster SSD and higher GPU load.
