Are you looking for ways to speed up compressed sensing? If you work in the areas of medical image reconstruction, image acquisition or sensor networks, you probably are. This paper, Parallel Implementation of Compressed Sensing Algorithm on CUDA-GPU, compares CPUs running Matlab and GPUs running Jacket using a Basis Pursuit Algorithm for compressed sensing. They compared an Intel Core 2 Duo T8100 (2.1GHz and 3.0 GB memory) running Matlab with a NVIDIA GeForce series 8400m GS (256 MB video memory, DDR2 and bus width of 64bit) using an older version of Jacket, Version 1.3. The CPU and GPU setups were used to run their Basis Pursuit Algorithm on six MRI images. These are some samples: The implementation using Jacket …
Our Point of View & Twitter Comedy
“Great businesses have a point of view, not just a product or service.” ~37 Signals At AccelerEyes, our point of view is that GPU software can and should deliver great results on real applications. With this point of view, we’ve kept our heads down solely focused on delivering a great runtime system for GPUs. All our energy has been devoted to the task of emitting optimized low-level code from high-level matrix notation. These efforts are now paying off in a big way! Jacket is consistently delivering awesome results in real applications, read examples here and here. Alternative choices apparently have a different point of view. Yesterday’s twitter stream contained a comical, but all-to-common indication of frustration with the recent GPU …
Digital Holograms Faster than Ever
REAL3D is a digital holography project funded by the EU and brings together nine participants from academia and industry under the FP7. As part of the project Nitesh Pandey, Damien Kelly, Bryan Hennelly and Thomas Naughton from the National University of Ireland, Maynooth demonstrate utilizing pre-computation and quantization of chirp matrices with GPUs running Jacket from AccelerEyes speeds up the reconstructions of digital holograms. Digital holography is a powerful imaging technique with many new applications like true 3D display. It allows the capture of both amplitude and phase information of the light reflected off the surface of 3D objects. Researchers at the National University of Ireland, Maynooth are developing techniques based on digital holography for 3D display applications. Reconstruction of …
Feature Learning Architectures with GPU-acceleration
Stanford researchers in Andrew Ng’s group used GPUs and Jacket to speed up their work on Feature Learning Architectures. They wanted to know why certain feature learning architectures with random, untrained weights perform so well on object recognition tasks. The complete write up can be found in On random weights and unsupervised feature learning in ICML 2011. They decide to use GPUs and Jacket for this study because of “the need to quickly evaluate many architectures on thousands of images.” Jacket taps into the immense computing power of GPUs and speeds up research utilizing many images. This is the architecture used in the study: They started by studying the basis of good performance for systems and found convolutional pooling …
A better way to time Jacket code
Whether you are a new Jacket programmer or a GPU maestro, you are bound to speed-test Jacket at some point. There are many factors to keep in mind while benchmarking Jacket code – a simple tic-func()-toc won’t do. For example, this is some typical benchmarking code: % warm up x = rand(n,’single’); x = grand(n, ‘single’); geval(x); % CPU timing tic for r = 1:reps x = rand(n,’single’); end cpu_time = toc; % GPU timing gsync, tic for r = 1:reps x = grand(n,’single’); geval(x); end gsync, gpu_time = toc With Jacket 1.7, this entire code chunk is now replaced by two lines: cpu_time = timeit(@() rand(n,’single’)); gpu_time = timeit(@() grand(n,’single’));
Improved Fat/Water Reconstruction Algorithm with Jacket
Case Western Reserve University researchers turned to GPUs running Jacket to develop a fast and robust Iterative Decomposition of water and fat with an Echo Asymmetry and Least-squares (IDEAL) reconstruction algorithm. The complete article can be found here. The authors report that “GPU usage is critical for the future of high resolution, small animal and human imaging” and Jacket “enables GPU computations in MATLAB.” Their research was performed on a desktop system with 32GB RAM, dual Intel Xeon X5450 3.0 GHz processors, an NVIDIA Quadro FX5800 (4GB RAM, 240 cores, 400 MHz clock), and MATLAB R2009a 64bit. Jacket v1.1, an older version, was used to produce these results. Reconstruction tests with different sized images were performed to evaluate computation times …
Hybrid GPU & Multicore Processing for LU Decomposition
One of the hot areas in supercomputing is hybrid compute: balancing the computational load between one or more CPUs and GPUs. Along these lines Nolan Davis and Daniel Redig at SAIC recently presented work on Hybrid GPU/Multicore Solutions for Large Linear Algebra Problems where they developed a novel algorithm for LU decomposition, one of the most important routines in linear algebra. Here’s a snapshot view of their setup: System Specs: GPU Nvidia® Tesla™ 2050 448 processing cores3 GB dedicated memory Multicore Host 24 cores64 GB system memory Red Hat® Enterprise Linux 5 Two AMD Opteron™ 6172 12-core processors Host-to-GPU Communications PCIE 2.0 16 channels at 500 MB/sec/laneTheoretical peak bandwidth of 8 GB/sec Their initial results are very promising. For …
CUDA over Remote Desktop now available for Tesla GPUs
Update: Jacket over Remote Desktop is now available for Quadro devices too! Read this post. Jacket over Remote Connections is also documented extensively on the AccelerEyes Wiki. Over the past several years, many Jacket programmers have requested support for Remote Desktop in Windows. We are pleased to report that recent NVIDIA drivers now enable Jacket to run over Remote Desktop, for some system configurations. Specifically, the requirements to make this work include: Windows Vista, Windows 7, Windows HPC Server 2008, or Windows HPC Server 2008 R2 The latest NVIDIA driver (as required by Jacket) Tesla GPU TCC-mode enabled on at least one (Tesla) GPU To enable TCC, the Tesla cannot be connected to a display. This means you need to …
Unraveling Speedups: Two Important Questions
One Jacket programmer recently emailed the following to us: Our chief scientists asked me a question that I’d like to pass on to you. I think I know the answer, but you guys can be much more definitive than I can. He recently read about people achieving ~10x speedups by converting parts of their code to MEX files. He was wondering how much of the observed speedup is due to that MEX and how much is due to CUDA and the GPU. Two Questions You Should Ask Yourself When contemplating an effort to optimize a piece of code, it is important to unravel the effort into two separate questions. Both need to be addressed to improve performance: How well-written is …
Stanford GPU Benchmarks: Jacket vs PCT/GPU
Researchers in the Pervasive Parallelism Laboratory at Stanford University recently published work describing a novel framework for parallel computing with a paper entitled, “A Domain-Specific Approach to Heterogeneous Parallelism.” As part of their research, they compared Jacket to the GPU support in the Parallel Computing Toolbox™. The results clearly show that Jacket’s optimizations make a big difference in performance. In this blog post, we highlight 4 algorithms included in their research: NAME DESCRIPTION INPUT Gaussian Discriminant Analysis (GDA) Generative learning algorithm for modeling the probability distribution of a set of data as a multivariate Gaussian 1,200×1,024 Matrix Restricted Boltzmann Machine (RBM) Stochastic recurrent neural network, without connections between hidden units 2,000 Hidden Units 2,000 Dimensions Support Vector Machine (SVM) Optimal …