On Wed, Jul 4, 2012 at 8:17 AM, Alberto Cirillo <cirilloalberto at gmail.com>wrote:
> Hi, > > I have a question about memory management in CUDA environment, using TAO > toolkit. > When i create a vector with VECSEQCUSP, it's placed on GPU memory, > transfered to CPU memory and returned back when operations are done > (function evaluation is made by CPU). > If I have a VECSEQCUSP vector with dimension that causes the exhaustion of > GPU's memory, there is a possibility to swap this vector on CPU's memory in > order to avoid "bad alloc" error? > For example: if GPU memory is 4GB and CPU memory is 8GB and my working > vector is 6GB, PETSc doesn't have a method to manage automatically the > vector avoiding the crash? > No. We would recommend that you run in parallel on multiple machines. Matt > Thanks > Alberto > -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120704/3f222401/attachment.html>
