ABINIT Discussion Forums

Dear friends,

I compiled abinit 6.6.3 with mpi (mpich2) and cuda support, everything seems fine - no errors, warning etc.
The log of every task contains following information
=== Build Information ===
Version : 6.6.3
Build target : x86_64_linux_gnu4.6
Build date : 20110523

=== Compiler Suite ===
C compiler : gnu4.6
CFLAGS : -g -O2 -mtune=native -march=native -mfpmath=sse
C++ compiler : gnu4.6
CXXFLAGS : -g -O2 -mtune=native -march=native -mfpmath=sse
Fortran compiler : gnu4.6
FCFLAGS : -g -ffree-line-length-none
FC_LDFLAGS :

=== Optimizations ===
Debug level : basic
Optimization level : standard
Architecture : intel_core2

=== MPI ===
Parallel build : yes
Parallel I/O : yes
Time tracing : no
GPU support : yes

=== Connectors / Fallbacks ===
Connectors on : yes
Fallbacks on : yes
DFT flavor : libxc-fallback+atompaw-fallback+bigdft-fallback+wannier90-fallback
FFT flavor : none
LINALG flavor : netlib
MATH flavor : none
TIMER flavor : abinit
TRIO flavor : netcdf-fallback+etsf_io-fallback

so I guess abinit recognized my gpu nvidia gtx275
I use mpirun -np 2 abinit < root.files >& log to run my "root.files"
Anyway i dont get any speedup and my gpu doesnt get any extra load while abinit running on really hard tasks (lots of k-points and bands).
my config:

debian 6
cuda 3.2
abinit 6.6.3
nvidia driver is 270.41.06 (not developer one, may be this is source of trouble)

core2duoE6600 3.2 Ghz
6 gb ram
nvidia gtx 275

Some gay said that i should use ngfft not less then 3*16 to enable cuda, but it doesnt worked.
Any ideas?)

Dear Magniff,
For the moment GPU are used only in a subset of ABINIT capabilties.
In particular GPU are used only in the recursion_method, which is a DFT for high temperature systems (direct space calculation) and
BigDFT (wavelet code- indipendent and distribuited also with ABINIT).

For the rest I konw that it is in project to use GPU to do FFT and Matricial linear algebra, but I am pretty sure that it is not implemented in the
current version.

Marco

thanx for reply, mancini
so, as i understood there is not any way to accelerate gs calculation in current version

HI ,
I am trying to install the parallel version of abinit with open-mpi still I couldn't get it to compile correctly , I was wondering whether you can guide me in the correct direction
here are the posts by me on compilation ;
http://forum.abinit.org/viewtopic.php?f=3&t=1099

thanks