Differences

This shows you the differences between two versions of the page.

--- faq:cuda_support [2014/02/25 10:02] – created oschuett
+++ faq:cuda_support [2014/04/06 14:32] – vondele
@@ Line 1: / Line 1: @@
 ====== Which parts of CP2K are CUDA-accelerated? ======
-  * FFTs, when compiled with ''%%-D__PW_CUDA%%''
+  * Anything that uses ''dbcsr_multiply'', i.e. sparse matrix multiplication, when compiled with ''%%-D__DBCSR_CUDA%%''. This benefits in particular the [[doi>10.1021/ct200897x| linear scaling DFT]] code. See also [[http://dbcsr.cp2k.org | the DBCSR project.]]
-  * Anything that uses ''dbcsr_multiply'', when compiled with ''%%-D__DBCSR_CUDA%%''. This includes in particular the [[linear scaling DFT]] code.
+  * If linked against an accelerated scalapack/blas library (in particular pdgemm/pdsyrk/dgemm) that executes these calls on the GPU. The impact of this is most visible for MP2 and RPA calculations. On the hybrid Cray XC30 linking against libsci_acc makes this happen.
-  * Anything that uses ''dgemm'' (**not** ''pdgem''), when linked against Cray's accelerated LibSci. Currently, this is most useful for the ''MP2'' method.
+  * FFTs, when compiled with ''%%-D__PW_CUDA%%''.