Second workshop in NIRICT GPGPU Reconnaissance workshop series


Workshop Programme


9:45 welcome with coffee

10:00 GPU Computing at the Netherlands eScience Center - Ben van Werkhoven (Netherlands eScience Center)

Abstract
This talk gives an overview of the GPU Computing projects at the Netherlands eScience Center. The eScience center has many GPU projects in different scientific disciplines, including digital forensics, particle physics, radio astronomy, super-resolution microscopy, and climate modeling. Next to an overview of the GPU work, we will zoom in on a few GPU applications and discuss them in more detail.

10:30 Inference of gene regulatory networks on GPUs - Dragan Bosnacki (TU/e)

Abstract
Knock out experiments are main stream methods for inference (reverse engineering) of gene regulatory networks. The goal of these experiments is to discover connections between genes in the living organisms. To this end, in each experiment one of the genes of a given set is deactivated (knocked out) and the effect on the other genes from the set is measured. Based on the obtained data, the regulatory network is reconstructed. We present efficient and scalable parallel algorithms for network inference from knock out experiments. To this end the notion of transitive reduction for weighted graphs is introduced which allows us to distinguish between direct and indirect interactions. The GPU implementations of these algorithms can achieve sped ups of two orders of magnitude compared to their sequential counterparts. Besides the acceleration, the concept of transitive reduction also results in improved quality of the reconstruction. (The presentation is based on a joint work with Maximilian R Odenbrett, Anton Wijs, Willem Ligtenberg and Peter Hilbers.)

11:00 coffee

11:15 GPU Usage in Gravitational Dynamics - Jeroen Bedorf (Sterrenwacht, U. Leiden)

Abstract
Ever since the introduction of CUDA (even before that actually) astronomers have been looking at accelerating their simulations using GPUs. With a long history in using accelerator techniques to speed-up gravity computations the acceptance of GPUs was a given. With production simulations using GPUs since 2011 today it is no longer a question if a GPU should be used but rather which model and how many. In this talk we describe how GPUs have changed the field, how accelerated software is integrated in the astronomers workflow and how we are slowly transitioning to including GPUs in multi-physics simulations such as combined gravity and hydrodynamics.

11:45 OpenCL: Industry Examples and 3.0 Preview - Vincent Hindriksen (StreamComputing)

Abstract
TBA

12:15 lunch

14:00 CLBlast: A Tuned BLAS Library for Faster Deep Learning - Cedric Nugteren (TomTom)

Abstract
We'll demonstrate how to accelerate dense linear algebra computations using CLBlast, an open-source OpenCL BLAS library providing optimized routines for a wide variety of devices. It is targeted at deep learning training and inference and thus provides a fast matrix-multiplication routine (GEMM) to accelerate the convolutional layers: the computational heart of all deep-learning frameworks (TensorFlow, Caffe, etc.). CLBlast has four main advantages over other BLAS libraries: 1) it can be explicitly tuned for specific matrix-sizes and hardware platforms, 2) it runs on less common devices (and it is fast), such as embedded and low-power GPUs, 3) it can perform operations in half-precision FP16 format, saving precious bandwidth, time, and power, and 4) is supports batched routines to accelerate multiple small operations.

14:30 Using GPUs in Tomography - Willem Jan Palenstijn (CWI)

Abstract
Tomography is the field of reconstructing three dimensional objects from two dimensional objects. The most common use of this is in medical CT (Computed Tomography) scans, but it is also extensively used in materials science, biomedical research, industrial inspection, and many other applications. GPUs are now commonly used for implementations tomography reconstruction algorithms. In this talk, we will look at a brief overview of the general field of tomography, the open source ASTRA Toolbox (started in 2008) with GPU implementations of tomography primitives, and our current plans and challenges, especially for more real-time and distributed computation.

15:00 GPU computing for scientific visualization and computer graphics - Andrei Jalba (TU/e)

Abstract
Early graphics processing units (GPUs) were designed as fixed-function accelerators tailored to fulfill specific needs of computer graphicstasks. Today, GPUs have become powerful computing platforms on par with modern CPUs. Therefore, hundreds of applications have been optimized to leverage their compute power. In this talk I will give an overview of several GPU-accelerated, compute-intensive applications from visualization and graphics. Finally, I will discuss a few of them in some more detail.

15:30 coffee

16:00 Speeding up sequence alignment enables new DNA sequence approaches - Sven Warris (U. Wageningen)

Abstract
Finding similarities in DNA sequences has been a key approach in bioinformatics for many years. One the underlying algorithms, the Smith-Waterman sequence alignment algorithm, has been implemented in CUDA and OpenCL. We combined these implementations with pyCUDA and pyOpenCL to create an easy to use application in Python: pyPaSWAS. This implementation was extended to find palindromic sequences in erroneous, long reads from platforms such as PacBio and Oxford Nanopore. Now we are able to correclty determine the DNA with long read technology of single (cancer) cells and even of single chromosomes.

16:30 Efficient Imaging in Radio Astronomy Using GPUs - Bram Veenboer (ASTRON )

Abstract
Realizing the next generation of radio telescopes such as the Square Kilometre Array requires both more efficient hardware and algorithms than today's technology provides. We'll present our work on the recently introduced Image-Domain Gridding (IDG) algorithm that tries to avoid the performance bottlenecks of traditional AW-projection gridding. We'll demonstrate how we implemented this algorithm on various architectures. By applying a modified roofline analysis, we show that our parallelization approaches and optimization leads to nearly optimal performance on all architectures. The analysis also indicates that, by leveraging dedicated hardware to evaluate trigonometric functions, NVIDIA GPUs are much faster and more energy-efficient than regular CPUs. This makes IDG on GPUs a candidate for meeting the computational and energy-efficiency constraints for future telescopes.

17:00 closing and drinks