Apparently, in Matlab one now uses maxNumCompThreads(1) at the prompt. After doing this (and verifying that it returns 1 when subsequently called), the timing for calling eigs with a tol value of 1e-14 was 1.01 seconds. For Julia -t1, and setting BLAS.set_num_threads(1) the time needed for the same tolerance setting was 2.76 seconds. So it doesn’t look like threading affects either function very much. Both calls used the sigma input variable to request eigenvalues close to a given value, so I assume shift/invert is being done. I haven’t implemented the instrumentation for counting iterations yet. May not get to that till later tonight or tomorrow.