Home My Page Projects OpenNL
Summary Activity Forums Tracker Lists Tasks Docs News SCM Files

Project Filelist for OpenNL

File Release Notes and Changelog

Release Name: PSM_1.4.0

Release Notes
New CUDA implementation for OpenNL solvers. 
Expect 3X to 10X performance gain for nlSolve() (measured up to 30Gflops with sparse matrices in double precision on GTX1080, 
up to 6Gflops on Quadro M1000M, 
if somebody gots a P100 that has good FP64 capabilities I'd be curious, probably >100GFlops).

Tested under Linux. To be tested on Windows.

To use: nl:CUDA=true on command line 
Change Log
* Removed old Concurrent Number Cruncher from OpenNL
* New BLAS abstraction layer 
* Implementation of iterative solvers using BLAS abstraction layer
* CUDA implementation of abstract BLAS and abstract matrices
* New experimental support of Intel MKL multicore sparse matrix vector product