It seems that there's a rising demand for HPC software programmers in the marked who can write cutting edge software for the processing powerhouses that GPUs and computer clusters have become.
Of course, the unanswered question that remains is whether companies want programmers with this background but only want to pay them regular salaries or there's truly a shortage in this type of expertise.
Anyhow, this HPCwire article outlines some aspects of this and it's quite an interesting text.
Could it be that OpenCL and the new HPC tools available for GPUs will allow someone to create the new Google or Apple? Time will tell.
Hi. We're back this year and hope everyone had great holidays. We'd like to share a few thoughts about how OpenCL and GPU computing will go about this year:
- Our first impression, considering that OpenCL 1.2 didn't change dramatically it's core, is that OpenCL is reaching the maturity we all have been hoping so far (Khrono's OpenCL 1.2 article).
- We hope to see the first OpenCL programmable mobile devices this year, in the form of FPGAs, tablets with programmable GPUs and powerful microprocessors.
- On the software side, we expect to see high performance algorithms coming from the manufacturers which are easier and less clumsy to use. In this aspect, we at CMSoft consider that we have been contributing our share by making Convex Optimization code available, as well as tutorials and image processing tools.
- And, of course, we at CMSoft are going to wrap up our OpenCL accelerated k-means algorithm and regularized least-squares and logistic regression and make them available as soon as we feel they're stable. We also intend to include image integrals and Haar feature extraction using OpenCL as a case study in the OpenCL tutorial.
A quadratic programming problem is a special type of mathematical optimization that involves minimization of a quadratic function (hence the name) subject to linear inequality and equality constraints.
We are proud to bring to the developer community an open source QP solver, still undergoing development but fully functional, which uses GPU acceleration to solve large dense problems.
Make sure to visit our new Optimization section to read more about the QP solver and download the software.
- Fixed eye inversion and swap buffer bug in stereoscopic GLRender framework; - Included Optimization tools to allow robust Curve Fitting and quadratic programming. Check out Optimization section.
For the second year, we bring an OpenCL / GL interoperation demonstration on Christmas. This time, we simulate particles that are attracted by three centers and repelled by 1 center.
The pattern is quite interesting, as shown in the video below:
I'd also like to say that we're closing for the year. Next year, we'll continue with the optimization resources and bring OpenCL accelerated least squares and logistic regression.
If you're interested in data categorization, you're probably familiar with the Support Vector Machine algorithm. SVMs are considered to be state-of-the-art classifiers and are used in a broad variety of fields, from handwritten digit recognition to spam classification.
This time we bring a SVM demonstration that shows the intuition of a maximum margin classifier using a RBF kernel, adequate to fit very nonlinear models.
