The test matrices are all from the UF sparse matrix collection, which need long time in numerical factorization.
Table 1 lists the time of numerical factorization between GSS and PARDISO. PARDISO’ s version is from Intel Composer XE 2013 SP1. GSS 2.4 use CPU-GPU hybrid computing. The testing CPU is INTEL Core i7-4770(3.4GHz) with 24G memory. The graphics card is ASUS GTX780 (with compute capability 3.5). NVIDIA CUDA Toolkit is 5.5. The operating system is Windows 7 64. Both solvers use default parameters.
For large matrices need long time computing, GSS 2.4 is Nearly 3 times faster than PARDISO. For matrices need short time computing, PARDISO is faster than GSS. One reason is that complex synchronization between CPU/GPU do need some extra time.
Table 1 numerical factorization time between GSS and PARDISO
|
Matrix |
Description |
Pattern |
dimension |
Non-zero |
GSS |
PARDISO |
|
msdoor |
Parasol matrices: medium size door |
|
415863 |
20240935 |
1.061 |
0.424 |
|
inline_1 |
structural problem: stiffness matrix |
|
503712 |
36816342 |
3.385 |
2.605 |
|
apache2 |
structural problem: SPD matrix (finite difference 3D) from APACHE small |
|
715176 |
4817870 |
3.518 |
4.386 |
|
boneS10 |
model reduction problem 3D trabecular bone |
|
914898 |
55468422 |
5.585 |
4.636 |
|
audikw_1 |
structural problem: symmetric rb matrix |
|
943695 |
77651847 |
45.534 |
122.832 |
|
bone010 |
model reduction problem: 3D trabecular bone |
|
986703 |
71666325 |
30.591 |
66.892 |
|
StocF-1465 |
computational fluid dynamics problem: flow in porous medium with stochastic permeabilies |
|
1465137 |
21005389 |
40.3 |
106.417 |
|
nd24k |
2D/3D problem: ND problem set |
|
72000 |
28715634 |
15.880 |
85.519 |
|
Fault_639 |
structural problem: contact mechanics for model of a faulted gas reservoir |
|
638802 |
28614564 |
43.633 |
186.405 |
|
Flan_1565 |
Finite element simulations: gas reservoir and structural problems. 3D model of a steel flange, hexahedral finite elements |
|
1564794 |
117406044 |
38.266 |
58.004 |
|
Emilia_923 |
Finite element simulations: gas reservoir and structural problems. geomechanical model for C02 sequestration |
|
923136 |
41005206 |
82.181 |
247.17 |
|
nd12k |
ND problem set. 3D mesh problems. |
|
36000 |
14220946 |
5.210 |
18.954 |
|
sum |
|
|
|
|
315.114 |
904.244 |