Why matlab performs much slower on a cluster with 144 cores than on a desktop with only 6 cores when solving a large system
2 次查看(过去 30 天)
显示 更早的评论
Hello, everyone! Recently I'm trying to solve a quite large system Ax = b, with A an 353344*353344 sparse matrix and number of nonzeros is 22548736. The structure of the matrix is shown in Fig. 1.
![](https://www.mathworks.com/matlabcentral/answers/uploaded_files/702892/image.png)
Fig. 1 Structure of the sparse matrix A.
At present, only direct solver is adopted. I solve this system on both desktop and a cluster.
On desktop, the details are as follows:
Matalb version: R2017 a;
CPU: Intel(R) Xeon(R) CPU E5-2620 v3 @ 2.40GHz;
Cores: 6 cores;
Time elapsed: 20 seconds on average
On Cluster, the details are as follows:
Matlab version: R2017 b;
CPU: Intel(R) Xeon(R) CPU E7-8870 v3 @ 2.10GHz
Cores: 72 cores;
Time elapsed: 232 seconds on average
OMP_NUM_THREADS: 72
As far as I know, when solving such a problem, the Intel Math Kernel Library with openmp-parallelized LAPACK and BLAS is adopted. If so, the program on cluster should be faster than that on desktop even though the frequency is a little lower than that of the desktop.
Since this system need to be solved thounds of times, I urgently need to accelarete this process. Can anybody give any advices on this?
0 个评论
回答(0 个)
另请参阅
类别
在 Help Center 和 File Exchange 中查找有关 Introduction to Installation and Licensing 的更多信息
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!