All tasks which involve upscaling (eg, 2X/4X in benchmark) are severely limited by CPU/RAM speed. This has been shown multiple times by people in the benchmark thread and can be seen by anyone in the GPU underutilization during any task involving scaling.
At the same time, TVAI doesn’t make full use of CPUs with more than 8 cores. My 16 core (+ 16 hyper threading virtual cores) performs best when only using 8 cores without hyperthreading.
These 2 issues could possibly be solved if TVAI would upscale more images in parallel on any CPU with more than 8 cores, so that the GPU utilization can be improved.
