Download presentation
Presentation is loading. Please wait.
Published byClaude Shields Modified over 6 years ago
1
GPU Computing Jan Just Keijser Nikhef Jamboree, Utrecht
10 December 2012 1
2
What is a GPU? 2
4
What can you do with a GPU?
From: Massively Parallel Computing with CUDA GPUs are excellent for performing the Same operation (Instruction) on Multiple Data elements (SIMD) Vector processing, anyone? 4
5
How do you program a GPU? From: An Introduction to GPU Computing and CUDA Architecture 5
6
How do you program a GPU? From: An Introduction to GPU Computing and CUDA Architecture 6
7
How do you program a GPU? From: An Introduction to GPU Computing and CUDA Architecture 7
8
Programming languages
NVIDIA CUDA AMD/ATI Stream Computing Open standard: OpenCL Supported by NVIDIA, AMD, Intel, ARM Language bindings C/C++ FORTRAN Python, Perl R, MATLAB, Mathematica 8
9
Possible applications
Fluid dynamics Tomography Monte Carlo simulations Chroma Library for Lattice Field Theory ... ? At Nikhef: ANTARES/KM3NET VIRGO ... ? 9
10
No silver bullet Requires code rewrite and optimization
GPU architectures rapidly change, so do the optimization techniques Limited by amount of memory on the GPU card (≤ 8 GB) Number of cores on a CPU is also rapidly increasing 10
11
Available GPUs @ Nikhef
Some desktops already have a GPU! NVIDIA Quadro NVS290 (low end) AMD Firepro V5900 1 server with 2 x NVIDIA Tesla M2070 Q2 2013: Intel Xeon Phi testboard Elsewhere in the Netherlands: SARA: GPU cluster (Tesla C1070) RuG: GPU cluster Successor to Huygens will use GPUs Leiden: Little Green Machine 11
12
12
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.