Technology for a better society. SINTEF ICT, Applied Mathematics, Heterogeneous Computing Group

Transcription

1 Technology for a better society SINTEF, Applied Mathematics, Heterogeneous Computing Group Trond Hagen GPU Computing Seminar, SINTEF Oslo, October 23,

2 Agenda 12:30 Introduction and welcoming Trond Hagen, SINTEF 12:45 The State of the Art in GPU Computing Jon Hjelmervik, SINTEF 13:45 Commercial Overview Chris Butler, NVIDIA 13:55 The NVIDIA Tesla GPU Compute Platform: Hardware, Software Developer Tools, and Support Timothy Lanfear, NVIDIA 14:30 Coffee, discussions, and mingling 2

3 Department of Applied Mathematics The Simulation group in Trondheim focuses on developing robust and efficient numerical methods for computational mechanics and geophysical flows. The Simulation group in Oslo develops robust and efficient computational methods for subsurface flow (petroleum, CO2, groundwater). Group for Heterogeneous Computing (Oslo) performs research on heterogeneous computing, many-core and data-stream processing. Group for Geometry (Oslo) focuses on computational geometry and development of 3D technology for the IT industry. Group for Optimization (Oslo) develops advanced optimization methods for applications within, among others; transportation, (maritime) logistics, and health care planning. 3

4 Heterogeneous Computing Group Seven Research Scientists and PhD students. 4 Geometry Simulation, Oslo Simulation, Trd. Optimization Heterogeneous Computing

5 Heterogeneous Computing 5

6 Heterogeneous Computer A heterogeneous computer (HC) is a tightly coupled system of processing units with distinct characteristics. A modern desktop or laptop computer is an example of such a system, as most systems include both a task-parallel, multi-core CPU and one or more data-parallel processors in the form of programmable graphics processing units(gpus). CPU (Intel Nehalem) GPU (NVIDIA Fermi) 6

7 Heterogeneous Computing Heterogeneous computing is the strategy of using multiple types of processing elements within a single workflow, and allowing each to perform the tasks to which it is best suited. aka Hybrid Computing aka Accelerated Computing aka General-Purpose Computing using GPUs (GPGPU) aka GPU Computing 7

8 Architecture and Algorithm Interaction ARCHITECTURE ALGORITHMS The key to performance is to understand the architecture and algorithm interaction. Need to know how to utilize the computational power of parallel accelerators in modern computer hardware. Designing compute intensive algorithms for obsolete hardware will most likely give bad performance. (sequential code / single-core CPUs) 8

9 Research Activities in the HC Group 9

10 Research Activities View-dependent tessellation Preparation of finite element models Soliving partial differential equations 10

11 Research Activities contd. Silhouette refinement Self-intersection detection of NURBS surfaces Registration of medical data Visualization of algebraic surfaces 11

12 Research Activities contd. Inpainting Navier-Stokes: Fluid dynamics 12

13 Research Activities contd. Volume visualization Electric activity in a human heart. Water injection in a fluvial reservoir 13

14 Research Activities contd. Matlab Interface to the GPU Cluster of GPU s Linear algebra / load balancing CPU - GPU 14

15 Research Activities contd. Ray-tracing, reservoir Adaptiv tessellation 15

16 Research Activities contd. Reservoir Simulation Operator split: Evaluate the pressure on the CPU in parallel with the GPU transport evaluation. Proposed a model for simulating in corner-point grids on the GPU: 16

17 High-speed Marching Cubes using Histogram Pyramids Marching cubes on GPUs: CPU: not interactive GPU: 200 fps Extraction of iso-surfaces Based on joint research with Gernot Ziegler, NVIDIA HPMC Open source library released soon 17

18 Smoothed Particle Hydrodynamics 18

19 Shallow-Water Equations 19

20 Heterogeneous Computing in the Cloud Heterogeneous computers (CPUs &GPUs) Web based clients Workstations 20

21 Thank you 21