laptop for matlab
in [DSP]
|
From: pravi on 9 Mar 2010 06:43 On Mar 9, 10:17 am, Rune Allnor <all...(a)tele.ntnu.no> wrote: > On 9 Mar, 10:05, pravi <ravindra.perav...(a)gmail.com> wrote: > > > > > On Mar 9, 9:59 am, Rune Allnor <all...(a)tele.ntnu.no> wrote: > > > > On 9 Mar, 09:37, pravi <ravindra.perav...(a)gmail.com> wrote: > > > > > Hello all, > > > > I need to buy a new laptop mainly for doing compute intensive 3D image > > > > processing using Matlab. > > > > Does anyone have any suggestions for a good configuration or product > > > > suggestions? > > > > Skip matlab. That step alone will boost performance by a > > > factor 10-50-100. > > > > No kiddin'. > > > > Apart from that, the obvious comment is that if you intend > > > to do graphics, then you will need to address the graphics > > > performance of the candidate computers. > > > > Rune > > > Thanks, Rune! > > > Eventually, the application will be ported to a stand-alone processor. > > However, the initial proof-of-concept has to demonstrated with matlab. > > So, currently, the interest is not in performance as meant by time, > > etc. > > but more to see if our algorithmic design functions. > > Again, skip matlab. Matlab's matrix-centric semantics is a > straight-jacket you would like to stay away from whenever > possible. Such a simple operation as computing the histogram > of an image is a major understaking in matlab, both what coding > time [*] and run-time is concerned, but can be coded in mere > minutes in C++, using the STL containers. And the C++ version > runs a *lot* faster, 10-50-100 times faster, than the matlab > version. > > [*] I know; matlab have built-in functions to compute the > histogram. For the sake of this argument, though, take a minute > to contemplate how you would do it to implement your own version. > > > Any suggestions for laptops with very good graphics performance? > > No. Ask your local dealer. > > Rune Hello Rune! Thanks again for your feedback. Your point is taken!! :) I am a computational biologist with a reasonably good signal processing theory background. However, I have very little C/C++ skills. I work mostly with algorithmic issues to solve problems (detection, estimation, tracking,etc.) on the large data sets that we collect hourly for days together. For people like us, Matlab/LabView have certain features that help in our science. If we can prove that our thinking is right, then, we can start to look for optimized implementations. For this, later, we can get to collaborate with people having advanced programming skills. But, at the beginning, we prefer to use existing scientific computing tools (although we write our "own" algorithms and use the built-in functions scarcely!). Thanks again for replying! Further suggestions will be nice to discuss!! Ravi.
From: Richard Dobson on 9 Mar 2010 07:46
On 09/03/2010 11:43, pravi wrote: ... > > I am a computational biologist with a reasonably good signal > processing > theory background. However, I have very little C/C++ skills. I work > mostly > with algorithmic issues to solve problems (detection, estimation, > tracking,etc.) > on the large data sets that we collect hourly for days together. > For people like us, Matlab/LabView have certain features that help > in our science. > Both Clearspeed and Nvidia have plugin libraries to accelerate Matlab using their custom cards. I am far from certain, but I ~think~ the Nvidia one can support an on-board GPU such as may be found on a highpend laptop. Only Windows and Linux supported. Clearspeed of course needs a PCI-e bus slot, which excludes laptops, I suspect. Richard Dobson |