Results for: Matrices

The QR Algorithm Computes Eigenvalues and Singular Values

Posted by Cleve Moler,

The QR algorithm is one of the world's most successful algorithms. We can use animated gifs to illustrate three variants of the algorithm, one for computing the eigenvalues of a nonsymmetric matrix, one for a symmetric matrix, and one for the singular values of a rectangular matrix. In all three cases, the QR iteration itself is preceded by a reduction to a compact form. All transformations are orthogonal similarities using Givens and Householder transformations. These are numerically stable, preserve eigenvalues, and preserve any symmetry.... read more >>

Hadamard Matrices

Posted by Cleve Moler,

I have just returned from the ICIAM2019 conference in Valencia, Spain. It was a huge conference -- 4,000+ attendees, dozens of prize and invited talks, hundreds of parallel minisympsia. I gave a talk in a two-part minisymposium organized by Nick Higham and Rob Corless. I outlined the first part of the talk in this blog a month ago. This post outlines the second part, which was about Hadamard matrices. Some of it is taken from a post in this blog five years ago.... read more >>

Benchmarking a GPU 5

Posted by Cleve Moler,

I recently acquired a GPU, a graphics processing unit. It's called a GPU because such processors were originally intended to speed up graphics. But MATLAB uses it to speed up computation. Let's see how the gpuArray object benchmarks on my machine. I have been doing computer benchmarks for years. I like to do profiles where I vary the size of a task and see how the amount of memory required affects performance. I always learn something unexpected when I do these profiles. Ben Todoroff is on the MathWorks Parallel Processing team. Last year he contributed #34080, gpuBench to the MATLAB Central File Exchange. He has been able to compare several different GPUs. I am going to consider the performance of only one GPU, but in more detail. Important note. This is only about double precision. Single precision is another story. ... read more >>

Experiments with Variable Format Half Precision

Posted by Cleve Moler,

For the past month I have been working with the variable format 16-bit floating point arithmetic that I described in this post. It has been frustrating work. I have found that the limited precision and limited range of half precision make it barely usable for the kind of experiments with matrix computation that I like to do. In this post I will describe a few of these experiments.... read more >>