Paper 3 Interactive TextureBased Volume Rendering for Large

Paper 3 Interactive Texture-Based Volume Rendering for Large Data Sets Joe Kniss University of Utah Patrick Mc. Cormick, Allen Mc. Pherson, James Ahrens, Jamie Painter, and Alan Keahey Los Alamos National Laboratory Charles Hansen University of Utah Electronic Visualization Laboratory (EVL) University of Illinois at Chicago

Introduction • a scalable, pipelined approach for rendering data sets too large for a single graphics card. • 5 fps is considered near interactive rates. • 5 fps on a 128 -CPU, 16 -pipe SGI Origin 2000 with IR 2 graphics hardware. • Rendering isn’t bottleneck , I/O is. • Data is preprocessed and quantized from native datatype to 8 /12 bit unsigned integer data • Data then split into sub-volumes to fit available texture memory on graphics pipes. Electronic Visualization Laboratory (EVL) University of Illinois at Chicago

TRex Pipeline Electronic Visualization Laboratory (EVL) University of Illinois at Chicago

Pipeline , I/O issues • Each stage is multi-threaded and has 2 main parts – event, functional managers • IO rates must match 300 MBps of texture download rates on IR pipes. • 16 parallel pipes -> 5 Gbps • Practically 4 Gbps reached – 4 channel fiber controller Electronic Visualization Laboratory (EVL) University of Illinois at Chicago

Renderer, Compositor • A renderer initializes multiple image buffers for simultaneous rendering with the other stages because the compositing and user interface stages both rely on the image buffer • Gets messages from UI and uses 3 D textures VAP or VCSS. • A Compositor begins once N renderers complete and composites back to front and done in software. • Pointer to renderer’s shared memory image buffer and sub-volume’s distance from eye point • A compositor is overwritten A=A+B. Electronic Visualization Laboratory (EVL) University of Illinois at Chicago

Compositor Electronic Visualization Laboratory (EVL) University of Illinois at Chicago

Overivew Electronic Visualization Laboratory (EVL) University of Illinois at Chicago

Immersive TRex • Separate daemon • Stereo pairs. So half the resolution • Adaptive tessellation of spherical shells. Electronic Visualization Laboratory (EVL) University of Illinois at Chicago