. Author manuscript; available in PMC: 2016 Jul 1.

Published in final edited form as: Parallel Comput. 2016 Jul;55:17–27. doi: 10.1016/j.parco.2015.10.015

Table 2.

Comparison of VMD parallel movie rendering performance for a 1,079 frame HIV-1 test movie on BlueWaters, after implementing several new optimizations within the TachyonL-OptiX GPU-accelerated ray tracing engine. For reference, the best previously published performance results are included for both GPU and CPU rendering [17]. Repeated timings reveal overall movie rendering performance gains ranging from a factor of 1.2× faster for presentation-sized movies (not shown in table) up to as much as 1.46× faster for high-definition 1920 × 1080 movies.

Movie	Ray Tracing Engine	Nodes	Full-Movie Wall Clock Execution Time
Movie	Ray Tracing Engine	Nodes	Script Load	State Load	Geometry and Rendering	Total

HIV-1 Capsid HD 1920 × 1080 144 AO samples/pixel	New TachyonL-OptiX (GPU) VMD 1.9.3	64 XK7	2 s	39 s	435 s	476 s
	New TachyonL-OptiX (GPU) VMD 1.9.3	128 XK7	3 s	62 s	230 s	295 s

	TachyonL-OptiX (GPU) VMD 1.9.2-UltraVis’13 [17]	64 XK7	2 s	38 s	655 s	695 s
		128 XK7	4 s	74 s	331 s	410 s
		256 XK7	7 s	110 s	171 s	288 s

	Tachyon (CPU) VMD 1.9.2-UltraVis’13 [17]	256 XE6	7 s	160 s	1,374 s	1,541 s
	Tachyon (CPU) VMD 1.9.2-UltraVis’13 [17]	512 XE6	13 s	211 s	808 s	1,032 s