I am running VAPOR 2.5.0 on a Windows 8.1 PC with AMD A5700 APU (with Radeon HD 7660D). Perform DVR of QCLOUD field with high resolution Jangmi data set. Change time step to 34 as suggested in tutorial. I get approximately 20 FPS without preintegration and 10 FPS with preintegration turned on.
As I recently read a paper by K. Engel claiming preintegration can greatly accelerate rendering performance, why does performance results of VAPOR show otherwise? Could the authors provide some insight, thanks.
Pre-integrated volume rendering will only improve performance if, as the Engel et al paper suggest, advantage is taken of the greater accuracy afforded by pre-intergrated volume rendering and fewer samples are used. The vaporgui DVR uses the same number of samples for pre-integrated and post-classification algorithms. Nevertheless I'm suprised you are seeing a noticable difference between the two. Typically the biggest cost penalty of pre-integrated volume rendering only occurs when you change the transfer function. On my Intel HD Graphics 4000 system I get about the same performance with either method.