Unfortunately, I did not bookmark the page I saw that comment on so I cannot provide a reference for it. However, the question was not intended to be an accusation. Personally, I would like to use Octave but I own a number of SPARC workstations/servers (Sun V880, Sun V440, Sun Blade 2000/1000, Sun Ultra 30) and some Sun Opteron systems (Sun Ultra 40, 4 Sun V40z's, and an HP DL585 server running Solaris 10). I would like to create a compute cluster and use them for computations. So I just wanted to know if Octave has been optimized for these processors or are the optimizations really just making sure you link the optimized linear algebra libraries available with sun studio 12 for example. My understanding of the lapack or blas libraries that come with Sun Studio is that they are hand optimized (in assembler) for the specific architectures that the compiler supports. If I just download ATLAS or Goto BLAS and I compile them I am not sure I would get the same performance. On the other hand maybe all or part of my assumptions are incorrect. This is what my post on this forum is intended to find out.
Thanks, Juan
On Dec 17, 2010 4:02pm, "John W. Eaton" <address@hidden> wrote: > On 17-Dec-2010, address@hidden wrote: > > > > | I read in a forum online > > > > Which forum? > > > > | that gnu octave was optimized primarily for the intel architecture chips. > > | Is this true today with version of octave at 3.0 and above? > > > > I don't think that has ever been true. > > > > | I would like to begin using octave on my hard to do analysis but > > | I am wondering if the performance will be enough. > > > > How can we answer this for you when we don't know any details about > > what you are trying to do, or what you might consider to be acceptable > > performance? > > > > | If in fact this true would > > | compiling octave and using the blas and lapack optimized libraries available > > | with the sun studio compilers address performance issues? > > > > Using optimized BLAS and LAPACK functions will improve the performance > > of linear algebra operations compared to what you would see using the > > reference implementations of BLAS and LAPACK. We recommend ATLAS or > > Goto BLAS because they are free software. > > > > jwe > > >