Afrigraph Tutorial B: Interactive RayTracing

1
Afrigraph Tutorial BInteractive Ray-Tracing

• Ingo Wald
• Philipp Slusallek
• Saarland University
• Computer Graphics Group
• http//graphics.cs.uni-sb.de

2

• For almost 20 years, researchers have argued that
  eventually, Ray-Tracing will become faster than
  rasterization

3

• For almost 20 years, researchers have argued that
  eventually, Ray-Tracing will become faster than
  rasterization
• And nothing happened...
• Well, almost ...

4
UNC Powerplant (12.5 Mtris, gt10 fps)
5
Four Power Plants (50 Mtris)
6
Tutorial Overview

• Introduction
• Introduction to Ray-Tracing
• Discussion Ray-Tracing versus Rasterization
• Previous Work
• Approximating Ray-Tracing
• Accelerated Ray-Tracing
• Interactive Ray-Tracing on PCs
• Coherent Ray-Tracing Implementation
• Comparisons (SW / HW)
• Distributed RT of Massive Models
• Outlook Hardware-Architectures for Ray-Tracing
• Future Research and Conclusions

7
Tutorial Overview

• Introduction
• Introduction to Ray-Tracing
• Discussion Ray-Tracing versus Rasterization
• Previous Work
• Approximating Ray-Tracing
• Accelerated Ray-Tracing
• Interactive Ray-Tracing on PCs
• Coherent Ray-Tracing Implementation
• Comparisons (SW / HW)
• Distributed RT of Massive Models
• Outlook Hardware-Architectures for Ray-Tracing
• Future Research and Conclusions

8
Introduction to Ray-Tracing

• In principle Very simple algorithm
• For each pixel
• Create ray through that pixel
• Cast ray into scene and find closest intersection
• Shade ray at intersection point
• Can also shoot new rays during shading
• Determine visibility of point lights by shadow
  rays
• Compute reflected/refracted light by recursively
  tracing reflection-/refraction-rays
• Basically, thats all

9
Ray-Tracing Algorithm
10
Introduction to Ray-Tracing

• Only three main components
• Generating rays
• Finding the closest intersection of a ray
• Ray traversal
• Ray-object intersection
• Shading

11
Ray-Generation

• Generate initial ray for each pixel
• Other camera models are trivial
• Fisheye lens
• Non-linear distortions/Lens effects
• Motion blur, depth of field
• Options
• More samples for anti-aliasing
• Adaptive Sampling
• Combine with IBR
• E.g. RenderCache Reuse samples by reprojection

12
Ray-Traversal
Grid (2D)

• Need to find objects quickly
• Exhaustive search infeasible
• Build spatial index structure
• Grid, octree, BSP-tree, BVH, ...
• Advantages
• Logarithmic complexity
• Occlusion culling
• Early ray termination
• Problems
• Multiple intersection computations
• (objects often in multiple voxels)
• Dynamic scenes ?

Octree (2D)
13
Ray-Object-Intersection

• Need to compute intersections fast
• Requires many floating point operations
• But typically dominated by traversal (21)
• Plenty of algorithms
• Plenty of primitives
• Even for triangles
• Optimizations
• Use SIMD CPU-extensions (SSE, AltiVec, 3D-Now)
• Data parallel execution
• Proper caching of data

14
Shading

• Lots of reflection models possible
• Phong, Cook-Torrance, Ward,
• Direct use of Shading Languages (Renderman)
• Shading after visibility has been computed
• No overhead due to overdraw
• Every ray is shaded exactly once
• Can generate new rays
• Shadow, reflection, transmission, ...
• Need to deal with recursion
• Rendering cost linear in rays traced

15
Introduction to Ray-Tracing

• Only three main components
• Generating rays
• Finding the closest intersection of a ray
• Ray traversal
• Ray-object intersection
• Shading
• Problem
• Find closest intersection is very expensive
• And Lots of rays per image

16
Rasterization Pipeline
Application

• In Contrast Rasterization
• Efficient HW implementation
• Use of object coherence
• Many new features
• Rendering is driven by App.
• Application submits geometry
• Visibility determined at end
• Z-buffer fragment test

TL, Vertex Ops
Rasterization
Texturing
Fragment Ops
Fragment Tests
Framebuffer
17
RasterizationDrawbacks

• Drawbacks of this approach
• Use of object coherence
• Only if triangle is large
• Rendering is driven by App.
• Application has to know what is visible
• Efficient occlusion culling is hard
• Visibility determined at end
• Overdraw Discard all but one fragments
• High depth complexity very inefficient

18
Ray-Tracing versus Rasterization

• Flexibility
• Handling unstructured groups of rays
• Image-based rendering, reflections, shadows
• Generality
• Ray-Tracing is the basis for many algorithms
• Global illumination, visibility,
• Used in many disciplines
• Physics, Biology, Chemistry, Telecom,

19
Ray-Tracing versus Rasterization

• Simple and Efficient Shading
• Shading happens after visibility computation
• Direct use of Shading Languages
• Correctness Image Quality
• Rasterization inherently relies on approximations
• Environment maps, shadow maps, ...
• Ray-traced images are correct by default
• True reflections and shadows
• Use of approximations is optional

20
Ray-Tracing versus Rasterization

• Parallel Scalability
• Ray-Tracing is embarrassingly parallel
• (e.g. each pixel independent of all others)
• Scales well with the available hardware
• Needs fast access to scene data base

21
Ray-Tracing versus Rasterization

• Scalability with Scene Size
• Occlusion Culling Logarithmic Complexity
• RT never even looks at invisible geometry
• RT traversal allows for efficient searching
  O(log N)
• Rasterization shows linear behavior O(N)
• ? RT wins for complex scenes
• But rasterization is improving

22
Ray-Tracing versus Rasterization

• Coherence
• Key to efficient rendering
• Rasterization Object coherence
• Allows for efficient HW implementation
• But only really efficient for large triangles
• Ray-Tracing Ray coherence
• Improved caching reduced bandwidth
• Allows for data parallel computation
• RT has much more coherence than assumed
• But harder to exploit

23
Ray-Tracing versus Rasterization

• Conclusion of that Comparison
• Ray Tracing has many advantages
• These advantages become ever more pronounced
• Not only qualty, also efficiency
• But Ray-Tracing is (still) costly
• Have to make it faster !

24
Tutorial Overview

• Introduction
• Introduction to Ray-Tracing
• Discussion Ray-Tracing versus Rasterization
• Previous Work
• Approximating Ray-Tracing
• Accelerated Ray-Tracing
• Interactive Ray-Tracing on PCs
• Coherent Ray-Tracing Implementation
• Comparisons (SW / HW)
• Distributed RT of Massive Models
• Outlook Hardware-Architectures for Ray-Tracing
• Future Research and Conclusions

25
Previous and Related Work

• Two ways to achieve ray-tracing like quality
  interactively
• Trace less rays per frame Approximative
  ray-tracing
• Rasterization hardware
• Image-based techniques
• Interpolation of ray-traced results
• Trace more rays/sec Accelerated ray-tracing
• Better data structures
• Better algorithms
• Better implementations
• Parallel processing

26
Previous and Related Work

• Two ways to achieve ray-tracing like quality
  interactively
• Trace less rays per frame Approximative
  ray-tracing
• Rasterization hardware
• Image-based techniques
• Interpolation of ray-traced results
• Trace more rays/sec Accelerated ray-tracing
• Better data structures
• Better algorithms
• Better implementations
• Parallel processing

27
Approximated Ray-TracingRasterization Hardware

• HW-Accelerated vista/shadow buffers
• Compute visible geometry in HW
• Lookup of geometry in frame buffer
• Only works for primary rays and point lights
• Creates artifacts (e.g. shadow buffer resolution)
• Augmenting hardware with RT effects
• Selective ray-tracing
• Integrate ray-tracing with OpenGL rendering
• Rasterization for diffuse objects
• Textures or splatting Stamminger/Haber 00/01
  for ray-traced samples

28
Approximated Ray-TracingCorrective Textures
29
Approximated Ray-TracingImage-Based Techniques

• RenderCache Walter et al. 99
• Store ray samples per pixel (color, depth, ...)
• Reproject samples for next frame
• Detect and fill holes by sending few new rays
• Heuristic algorithms based on neighborhood
• Locate and correct errors (shadow, etc)
• Pseudo-randomly sample a few other pixel
• Adaptively sample near error regions
• But Reprojection and Heuristics are expensive
• Pays off (only) when pixels are very expensive to
  compute directly (e.g. global illumination)
• Scales badly with CPUs

30
Approximated Ray-TracingImage-Based Techniques

• Holodeck Ward 98
• Similar to RenderCache, but
• Long term storage of ray samples on disk
• Fast access to samples based on grid structure
• Builds light-field-like data representation

31
Approximated Ray-TracingImage-Based Techniques

• Interpolation in the image plane
• Pixel-selected ray-tracing Akimoto, 89
• Coarse sampling grid
• Adaptive refinement based on error criteria
• Linear interpolation between samples
• General ray interpolation Bala, 99
• Object-/Ray-/Image-Space
• Time
• Error bounded

32
Previous and Related Work

• Two ways to achieve ray-tracing like quality
  interactively
• Trace less rays per frame Approximative
  ray-tracing
• Rasterization hardware
• Image-based techniques
• Interpolation of ray-traced results
• Trace more rays/sec Accelerated ray-tracing
• Better data structures
• Better algorithms
• Better implementations
• Parallel processing

33
Accelerated Ray TracingBetter Data
Structures/Algorithms

• Best data structure (Grid vs BSP vs) ?
• Always scene and implementation dependent
• In practice, most do about equally well
• Well-reserached topic ? New data structures are
  unlikely to be found
• But Potential for better algorithms
• Can we better exploit coherence ?
• Can we build data structures faster ?
• Can we build data structures fully automatically
  ?
• Also Need for dynamic data structures

34
Accelerated Ray-TracingParallelization on
SuperComputers

• RT of large CSG models Muuss 95
• Motivation Interactively render complex data
  sets
• Idea Use raytracing
• Flexibility Avoid tessellation of CSG-models
• Take advantage of logarithmic complexity of RT
• Exploit parallelism
• Implementation
• Optimized, general RT algorithm
• 96 CPU, SGI PowerChallenge, shared memory
• Results
• 1-2 frames per second _at_ video resolution (in
  95!!!)

35
Accelerated Ray-TracingParallelization on
SuperComputers

• Utah Parallel RT System Parker 99
• Similar approach to Muuss
• Parallelization on shared memory machine
• Supports general primitives and volume data sets
• Results
• Has shown scalability up to 128 CPUs
• Importance of caching analysis
• New goal interactive visual cues for
  visualization(Same information at less cost)

36
Tutorial Overview

• Introduction
• Introduction to Ray-Tracing
• Discussion Ray-Tracing versus Rasterization
• Previous Work
• Approximating Ray-Tracing
• Accelerated Ray-Tracing
• Interactive Ray-Tracing on PCs
• Coherent Ray-Tracing Implementation
• Comparisons (SW / HW)
• Distributed RT of Massive Models
• Outlook Hardware-Architectures for Ray-Tracing
• Future Research and Conclusions

37
IRT on PCsWhat to keep in mind

• PC hardware has changed dramatically
• Processors become much faster
• But increase in ray-tracing speed is gradual
• Increasing gap between speed of CPU and memory
• But ray-tracing algorithm did not change
• SIMD extensions
• Flops become increasingly cheap
• But difficult to take advantage of in ray-tracing
• Fast (and cheap) networking network of PCs
• But good performance on non-shared-memory is hard
• Small clusters are around everywhere

38
IRT on PCsWhat to keep in mind

• PC hardware has changed dramatically
• Have to adapt our algorithms !
• Special emphasis on
• Keeping the CPU busy
• Memory Caching(1 cache miss can cost several
  triangle intersections)
• SIMD
• Not so important any more
• Instruction count, avoiding float ops

39
General Optimizations Cache

• Main memory is too slow for CPU (110)
• (bandwidth and latency)
• Keep relevant data in caches
• Design algorithms for cache reuse ? coherence
• Align data to cache lines (32 bytes)
• Separate data according to usage
• Separate volatile from non-volatile data
• Store intersection data separate from shading
  data(e.g. shading normals not needed for
  intersection)
• Prefetch data
• Design algorithms to enable data access prediction

40
General Optimizations Cache

• Cache Reuse Example Triangle Data Structure
• Variant 1 Struct Triangle Vec3f a,b,c
• Intersect() routine works on this structure
• Prefetching hard (2 levels of indirection)
• Data stored in 4 different memory regions
• (1 struct 3 vectors)
• Worst case 8 cache misses
• (if each of the 4 data overlaps cacheline border)

41
General Optimizations Cache

• Cache Reuse Example Triangle Data Structure
• Variant 2 With preprocessed intersection data
• All necessary data packed into 48 aligned
  bytes(see paper)
• Con Additional data to store (48b/triangle)
• But several advantages
• At most 2 cache misses
• 1 continuous memory region ? Trivial to prefetch

42
General Optimizations Cache

• This was only one example Similarly for
• BSP Nodes (even more important)
• Triangle lists
• Materials
• Shading Data

43
General Optimizations Simplification

• Today's CPUs have very long pipelines
• Simplify the code to avoid pipeline stalls
• Choose simple algorithms
• KISS wins(KISS keep it simple and stupid)
• E.g. BSP-tree traversal simpler than grids
• Easier to maintain and optimize (e.g.
  prefetching)
• Write tight inner loops
• E.g. better caching and handling of branches
• Avoid conditionals/relative jumps in inner loops
• E.g. support only triangles
• Avoid memory-access stalls
• ? Caching, caching, caching !!!

44
OptimizationSIMD Extensions

• Most CPUs provide SIMD extensions
• Intel SSE (Others 3D-Now!, AltiVec, ...)
• Use SIMD higher speed lower bandwidth
• Up to four parallel floating point operations
• ? For the cost of 1 !
• Fetch data once to reduce bandwidth to cache
• Amortize loading cost over 4 operations
• ?Factor 4 in bandwidth reduction
• Overhead due to restricted instruction set
• E.g. no SSE dot product
• Con Programming in assembly language

45
OptimizationSIMD Extensions

• How to use SIMD Extensions ?
• Either Instruction-parallel
• Combine 4 computations in normal algorithm
• E.g. the 4 mults in a dot product
• Or Data-parallel
• Run algorithm on 4 different data in parallel
• E.g. 4 independent dot products

46
SIMD Intersection

• SIMD best used in data parallel fashion
• Little instruction-level parallelism (in RT)
• ? Just doesnt work
• Data parallel 1 ray ? 4 triangles
• Hard to always have four triangles ready
• Data parallel traversal for 1 ray ?
• Data parallel 4 rays ? 1 triangle
• Must traverse rays in parallel ? ray packets
• Standard intersection code
• Overhead for terminated rays(E.g. 1 ray hits, 3
  rays miss)

47
SIMD Intersection

• Performance Results
• Comparison against already optimized C code
• Amortized cost for SSE code
• ? 20-36 million intersections/sec! (P-III, 800
  MHz)

48
SIMD BSP-Traversal

• Recursive Traversal Algorithm

49
SIMD BSP-Traversal

• SIMD-Traversal
• Traverse four rays in parallel
• Intersection with split plane traversal
  decision
• Combine decisions flags
• All rays must perform the same traversal
• Make sure order is consistent
• Easy to guarantee Same ray origin or same signs
  of direction vector
• Avoid recursion function calls
• Maintain stack manually
• Worst case as bad as before

50
SIMD BSP-Traversal

• Overhead of SIMD-Traversal (in )
• Fixed resolution at 10242 (l), fixed 2x2 packet
  (r)
• Traversal still dominates rendering cost
• Overall speedup factor 2 to 2.3

51
Coherent Algorithm Tracing Ray Packets

• Many rays are very similar
• e.g. primary and shadow rays, but others too
• Handle rays together in packets of 4 rays
• Process them in lock-step (? SIMD)
• Reorder computations to be partly breadth-first
• Load data once and use it for all rays
• Reduces memory bandwidth (e.g. SSE Factor 4 !)
• Increases Cache Utilization
• Coherence increases with image resolution
• more rays in same view frustum

52
Coherent Algorithms Shading

• SIMD Phong-Shading
• Fixed cost per image
• Rearrange data from ray packets
• Different depth non-coherent shadow rays
• Different materials different shaders
• Algorithm
• Parallel shadow rays to light sources
• SIMD shading using shadow flags
• Constant shading texturing cost (lt10)
• Procedural shading is easy (noise)

53
Coherent Ray-Tracing Summary

• Speedup
• Prerequisite Expose coherence in ray-tracing
  algorithm
• Factor gt5 General optimizations
• Factor gt2 SIMD computations
• Further optimizations are possible
• Better prefetching, more efficient shading
• Performance
• 200K to 1.5M primary rays/s (800 MHz, P-III)
• Almost linear in of reflection shadow rays

54
Comparison Test Scenes
55
Comparison Software Ray-Tracers

• Time per primary ray (1 CPU, 5122, in ?s)
• Main memory RTRT 256MB, others up to 1GB
• Rayshade Best grid resolution

56
Comparison OpenGL Hardware

• Frame rate with SGI-Performer (5122, fps)
• HW Octane V8, Onyx3/IR3, Geforce II GTS
• CPUs Onyx 8, nVidia 2, RTRT 1

57
Comparison Scaling with Scene Size

• Render time of subsampled terrain (spf)
• Typical linear scaling of rasterization HW
• Worst case for RT No occlusion
• Only 1 CPU !

58

• Demo / Video

59
Distributed RT of Massive Models
60
Reference Model (12.5 Mtris)
61
Previous Work

• Rendering of Massive Models Aliaga 99
• Framerate 5 to 15 fps for single power plant
• Needs shared-memory supercomputer (SGI)
• Framework of algorithms
• Textured-depth-meshes (96 reduction in tris)
• View-Frustum Culling LOD (50 each)
• Hierarchical occlusion maps (10)
• Extensive preprocessing required
• Entire model 3 weeks (estimated)
• Only semi-automatic

62
Distributed RT of Massive Models

• Ray-Tracing and massive models just match
• Logarithmic scaling in primitives
• Ideal for big models
• Preprocessing
• Simple and fast spatial sorting, fully automatic
• Distributed computing
• Parallel scalability to many networked computers
• No scene replication
• ? Our Approach Use coherent ray-tracing
• Caching of scene data in network
• Deal with network issues by reordering

63
Ray-Tracing Issues

• Distributed Scene Management
• Several GB of scene data
• File size and virtual address space (32 bit)
• Cannot use OS caching (demand paging)
• Cache miss will stall the entire process
• 1ms network latency time to trace several
  hundred rays
• Reordering would need non-blocking memory read
• Need to handle cache manually
• No longer limited by address space
• Allows reordering of computations
• Do not wait for missing data
• Continue with other rays while data is being
  fetched

64
Massive Models Caching

• 2-Level BSP-Trees
• Caching based on voxels
• Voxels are completely self-contained

65
Structure of the BSP-Tree
66
Distribution Issues

• Preprocessing
• Simple spatial sorting
• Need out-of-core algorithm due to model size
• Simplistic implementation 2.5 hours
• Estimated with optimizations lt 30 min
• Model Server
• Single server provides all model data
• Potenial bottleneck
• Should be distributed as well
• At least for more than 10 clients
• Trivial to implement

67
Distribution Issues

• Load Balancing
• Tile based (32x32 pixels)
• Demand driven
• Avoid idle-times
• prefetching tiles
• Asynchronous communication
• Frame-to-Frame Coherence
• Keep rays on the same client
• Simple Keep tiles on the same client
• Better Assign tiles based on reprojected pixels
• Larger effective cache size
• Increases with number of clients

68
Results

• Setup
• Seven dual Pentium-III 800-866 MHz
• FastEthernet (100Mbit) for normal clients
• GigabitEthernet only for display model server
• Performance for one Power Plant
• 4-5 fps without SSE optimization
• Factor 2 speedup with SSE
• Almost perfect scaling from 1 to 14 CPUs
• Never tried any more than that

69
Animation Framerate vs. Bandwidth
70
Speedup
71

• Demo / Video

72
Tutorial Overview

• Introduction
• Introduction to Ray-Tracing
• Discussion Ray-Tracing versus Rasterization
• Previous Work
• Approximating Ray-Tracing
• Accelerated Ray-Tracing
• Interactive Ray-Tracing on PCs
• Coherent Ray-Tracing Implementation
• Comparisons (SW / HW)
• Distributed RT of Massive Models
• Outlook Hardware-Architectures for Ray-Tracing
• Future Research and Conclusions

73
Ray-Tracing Hardware

• Summary so far
• RT has many technicaladvantages
• Better performance forlarge scenes, (logN vs N)
• Better image quality, more features
• But High initial cost onmain CPU
• ? Hardware support would help

74
Ray-Tracing HardwareWhy today ?

• The setting has changed
• Real scenes arent suited for rasterization any
  more
• High depth complexity
• Large scenes, small triangles
• Shading becomes more expensive
• Demand for more features (shading,
  programmability)
• Advantages of raytracing finally come to play
• Also Flops arent that expensive any more
• Number of Gigaflops per Gforce ?
• Neither is memory

75
Ray-Tracing HardwarePrevious Work

• Over the last decade Several research systems
• Often suffered from lack of resources
• Memory and Flops too expensive 10 years ago
• Offline-Ray-Tracing AR250 (ART)
• Accelerated offline rendering, bandwidth limited
• Volume-Ray-Casting systems
• Full volume ray casting on a chip
• Many, some already commercially successful

76
Ray-Tracing HardwareThe SHARP Architecture

• SHARP architecture Tim Purcell, Stanford
• Mixed SW/HW approach
• Based on SmartMemories Mai 00
• Multiprocessor on a Chip
• Roughly 64 R10k, with 8GB/s (!) memory bandwith

77
Ray-Tracing HardwareThe SHARP Architecture

• Conclusions from SHARP(Also see Siggraph 2001,
  Course 13)
• Simple caching works very well
• Good ray coherence
• Off-chip bandwidth is minimal
• Simple memory access design
• Performance (512x512)
• Conference scene 50 fps
• Reconfigurability allows to adapt to demands
• Adapt number of shading/traversal units to scene

78
Ray-Tracing HardwareOther Architectures

• RAYA (MERL, Siggraph 2001, Course 13)
• Based on Memory Coherent Ray-Tracing Pharr
• CORA (Saarbrücken)
• Hardware version of Coherent RT Algorithm
• Custom-design chip
• Est. performance 30/25 fps at 1024x768
• Cruiser 3.5 Mtris, 2 lights
• BunnyQuake 110 Ktris, 2 lights, 3 reflection
  levels

79
Tutorial Overview

• Introduction
• Introduction to Ray-Tracing
• Discussion Ray-Tracing versus Rasterization
• Previous Work
• Approximating Ray-Tracing
• Accelerated Ray-Tracing
• Interactive Ray-Tracing on PCs
• Coherent Ray-Tracing Implementation
• Comparisons (SW / HW)
• Distributed RT of Massive Models
• Outlook Hardware-Architectures for Ray-Tracing
• Future Research and Conclusions

80
What you should take home with you

• Interactive Ray Tracing IS feasible
• If importance is paid to underlying hardware
• Its not only feasible, its already there
• Not only a theoretical phantasy any more
• And even on cheap PCs
• Not only better, it can even be faster
• At least for certain applications

81
The Future

• IRT enables completely new applications
• Just think what has been done OpenGL
• Large scale visualization engineering,
• Handling of huge models
• Interactive global illumination (?)
• Need to adapt algorithms to new situation
• Flexible rendering
• Gaze tracking and non-uniform sampling density
• Image-Based or Frameless rendering
• Question What can IRT do for you?

82
Open Research Problems

• Can we make it even faster ?
• Hardware
• What is the best HW architecture?
• Dynamic Scenes
• Optimized rebuild or transformation of index?
• API
• Better alternative to OpenGLs push model?
• OpenGL not suited for Ray-Tracing
• Global Illumination
• Efficient new algorithms

83
Acknowledgements

• AMD
• Generous support, sponsoring and collaboration
  soon 24-node dual-Althlon IV, 1.5GHz cluster
• Presenters of the Siggraph 2001 Course 13
• Images, material, and information
• Tim Purcell Pat Hanrahan (Stanford)
• Many discussions and ideas
• The Max-Planck-Institute at Saarbruecken
• Collaboration and use of their Graphics Hardware
• C. Benthin M. Wagner others
• Work on the RT implementation and discussions

84
Links

• mailto//wald_at_graphics.cs.uni-sb.de
• For any questions or comments
• http//graphics.cs.uni-sb.de/rtrt
• The Saarland Universities RealTime RayTracing
  Project
• http//graphics.cs.uni-sb.de/pub/afrigraph01
• Tutorial Notes (Slides, Papers)
• http//www.openrt.de
• The OpenRT Interactive Raytracing API (not yet
  online)

85
The Future

• Applications on compute clusters
• Visualization of large models
• Previewing of animations with full shading
• Hardware support for IRT
• At least for specialized applications
• Convergence between RT and TR
• Occlusion culling
• Improved shading capabilities
• Eventually based on the same API?

86
Open Research ProblemsGlobal Illumination

• New situation
• Ray-tracing bottleneck is gone (Well, almost)
• New challenges
• Need for coherence
• Efficient computations
• Usage of view-importance
• High-degree of parallelism
• Small communication overhead
• Interactivity !!!
• Can we trade quality for speed ?