Scientific Computing Scientific Computing on Graphics Hardware on Graphics Hardware GPGPU Applications GPGPU Applications Examples Showcase Examples Showcase
Jan 10, 2016
Scientific ComputingScientific Computingon Graphics Hardwareon Graphics Hardware
GPGPU ApplicationsGPGPU Applications
Examples ShowcaseExamples Showcase
2
OverviewOverview
• Interactive Demos, Videos– Computational Fluid Dynamics
– Scientific Visualization
– Protein Matching
– Path Planning
– FFT Applications
– Adaptive Data Structures
• Feel free to ask questions!
Note: This is a small selection to show some of the possibilities and arouse further curiosity, alternative implementations to the above and many more applications exist.
3
Computational Fluid DynamicsComputational Fluid Dynamics
Real-time simulation of fluid flow governed by the Navier-Stokes equations
Live Demo:
NVIDIA “GPGPU FLUID”
Youquan Liu et al., Pacific Graphics 2004
3D fluid with complex obstacles
4
Scientific VisualizationScientific Visualization
Jens Krüger et al., TU München
A Particle System for
Interactive Visualization
of 3D Flows
Winner of the IEEE
Visualization Contest 05
Live Demo...
5
Fuzzy Protein Matching with HMMerFuzzy Protein Matching with HMMer
FR
NT
TP
F
G
FR
N
T
TP
F
G
F
FRNT
TP
F
GR
TT
FP
F
G
FR
TT
NP
F
GFNNN
NP
P
RR
F
T
TP
F
T
N
FNNN
NP
P
PR
TN
TR
F
G
Probabilistic model
Database
Query
Junk
Junk Junk Junk Junk Junk
Del Del Del Del
Start EndF R N T PT
[Daniel Horn et al., Supercomputing 2005]
6
ClawHMMer PerformanceClawHMMer Performance
0
10
20
30
40
Colipase Connexin50
Adenovirus Arfaptin PGI DUF#499
Rela
tive P
erf
orm
an
ce
P4
G5
X1800XT
7800 GTX
7
FlexiPlan - Path Planning for Deformable Robots in FlexiPlan - Path Planning for Deformable Robots in Complex EnvironmentsComplex Environments
[Russell Gayle et al.,Systems and Science, 2005]
Application: Liver Chemoembolization A long thin catheter is navigated from the femoral to the hepatic arteries near a tumor in the liver for direct drug injection.
Movie ...
8
FFT ApplicationsFFT Applications
θ θ
Spatial domain Frequency domain
f(x,y) F(u,v)
pθ(r) Pθ(w)
projection slice
1D-DFT
2D-DFT
Fourier Volume Rendering
• Inverse Filtered Back-Projection– Slice in frequency = Projection
– Needs high-quality interpolation
• 512³ rendered in >25 frames/sec– But needs slightly more space!
Filtered Back-Projection
• Create volume from projections– Blending of “FFT-filtered” projections
– Visualization of intermediate results!
• Faster by an order of magnitude!– 4.5 min (CPU) vs. 24 sec (GPU)
[Thomas Jansen et al., VMV 2004]
Movie ...
9
Glift : Generic, Efficient, Random-Access GPU Glift : Generic, Efficient, Random-Access GPU Data StructuresData Structures
[Aaron Lefohn et al.,ACM TOG, 2006]
Movie ...
STL-like abstraction of data containers from algorithms for GPUs
10
More DemosMore Demos
See GPGPU case studies in …
• Siggraph 2005 course noteshttp://www.gpgpu.org/s2005/
• Visualization 2005 course noteshttp://www.gpgpu.org/vis2005/