I never tried the test because I would need to put 1.2 back on my computer.
I would also say it depend on your set up.
On my computer as an exemple the depth maps calculation (for dense cloud) is made at 5% by CPU and 95% by the GPU.
For the alignment, detecting and matching points are calculated on both CPU and GPU, whereas the camera location estimation is calculated on CPU only. The time-saving is huge !
For other better result, I would say the ground extraction tool, which had several problem in 1.2, is working perfectly on 1.3 (I have extracted ground from flowers by using 5 cm max distance...)
Many new tools are pretty useful, such has shapes on pictures, or ruler which can measure point coordinates directly on pictures, model or orthomosaic...
You really miss something
Regards