Skip to content

jacobi - a benchmark by solving 2D laplace equation with jacobi iterative method. GPU or Xeon Phi can be used.

Notifications You must be signed in to change notification settings

dc-fukuoka/jacobi

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

jacobi - a benchmark by solving 2D laplace equation with jacobi iterative method.
GPU or Xeon Phi can be used.

how to run:

# for intel (intel compiler and intelmpi are required)  
$ make mic  

# for GPU (PGI compiler and its MPI are required)  
$ make gpu  

$ cat fort.18  
10000       ! iter_max  
4096 4096   ! m, n  
1 2         ! np_m, np_n  
$ vi fort.18 <-- adjust the paramters  

$ mpirun -np $NP ./jacobi_mpi.intel   # for Xeon phi  
or  
$ mpirun -np $NP ./jacobi_acc_mpi.pgi # for GPU
# where $NP = np_m * np_n

# if you do not have GPU nor Xeon Phi,
$ mpirun -np $NP ./jacobi_mpi.intel.nooffload
or
$ mpirun -np $NP ./jacobi_mpi.pgi

# to view the result
$ gnuplot # splot "jacobi.dat" w l

performance comparison:

iter_max=10000
m=4096, n=4096

with Xeon E5-2680(8 cores/socket, 2 sockets) + Xeon Phi 5110P x2, 2 nodes (np_m=1, np_n=4):

only CPU[s]: 66.7373
CPU + MIC[s]: 29.1181

with Xeon E5-2680 v4(14 cores/socket, 2 sockets) + NVIDIA Tesla P100 x4, 2 nodes (np_m=1, np_n=8):

only CPU[s]: 94.7290
CPU + GPU[s]: 5.29180

the following is an example of calculation result.

m=1024 n=1024
with GPU:

$ mpirun -x LD_LIBRARY_PATH -np 8 -npernode 4 -bind-to socket ./jacobi_acc_mpi.pgi
 iter_max:        10000
 m:         4096 n:         4096
 np:            8
 np_m:            1 np_n:            8
    0  0.499878
 1000  0.000357
 2000  0.000178
 3000  0.000118
 4000  0.000088
 5000  0.000071
 6000  0.000059
 7000  0.000050
 8000  0.000044
 9000  0.000039
10000  0.000035
Total CPU time:    5.29180
Device init   :    1.91826
Setup problem :   0.841110E-01
Data copyin   :   0.291271E-01
Computation   :    3.13129
Data copyout  :   0.277512E-01
Output        :   0.101262

Alt text

with Xeon Phi:

$ srun --mpi=pmi2 -N2 -n4 -c8 --cpu_bind=cores -t0:05:00 ./jacobi_mpi.intel
 iter_max:       10000
 m:        1024 n:        1024
 np:           4
 np_m:           1 np_n:           4
 
Number of threads = 236
    0  0.499512
 1000  0.000347
 2000  0.000171
 3000  0.000113
 4000  0.000084
 5000  0.000066
 6000  0.000055
 7000  0.000047
 8000  0.000041
 9000  0.000036
10000  0.000032
Total CPU time:    6.98289
Device init   :   0.400000E-05
Setup problem :   0.640400E-02
Data copyin   :   0.208880E-01
Computation   :    6.82102
Data copyout  :   0.292500E-02
Output        :   0.131650

Alt text

About

jacobi - a benchmark by solving 2D laplace equation with jacobi iterative method. GPU or Xeon Phi can be used.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published