Update for CULA Sparse Beta

by John

Hello all, we are nearing the official launch of the CULA Sparse Beta program. First off, I'd like to thank everyone who has submitted applications and to encourage others to so do as well here. Notifications for program acceptance will start going out at the beginning of next week, with the Beta 1 installers available very shortly thereafter. I would like to note that receiving an acceptance will require fully completing the form, which is quite short.

As for the package itself, the contents have been disclosed at other venues, but the final Beta 1 is list shaping up to include:

  • Cg, BiCg, and GMRES solvers
  • CSC, CSR, COO storage formats
  • Jacobi, Block Jacobi, ILU0¬†preconditioners
  • Double precision only
  • GPU solvers only
  • A single, C-style simple interface that is suitable for initial testing
  • Support for all the standard CUDA platforms; Linux 32/64, Win 32/64, OSX

Maybe the most surprising one on that list is that this first release will be double precision only. So please evaluate speed only on a suitable double precision platform, such as the Tesla 2000-series and Quadro 6000. Functionality can still be evaluated with a gaming card such as the GTX580 or lower end computing part such as the C1060 or Quadro 5800 as long as it is Compute 1.3 or greater. This will be covered in detail in the release notes as well.

We intend to grow the capabilities list dramatically in Beta revisions (see the release notes for CULA's Betas to see the scope of changes we intend). The immediate feedback we are looking for is on what will best serve our users in future versions, be it CPU solvers so as to have a fallback, more advanced interfaces, different preconditioners, etc. We're excited to hear what our users think!

Comments (0) Trackbacks (1)

Leave a comment

Trackbacks are disabled.