Description
LU measures the floating point rate of execution for solving a linear system of equations.
The parameters of the implemented programs are:
- M the size of the matrix, that is, the matrix is M by M.
- B the block size,
- px*py is the processor grid
Performance
P = N * T |
M |
B |
px |
py |
GFlops |
GFlops |
|
|
|
|
|
2009 |
2008 |
16 = 1 * 16 |
45600 |
200 |
4 |
4 |
92.07 |
67.45 |
32 = 2 * 16 |
64000 |
200 |
4 |
8 |
173.11 |
123.08 |
64 = 4 * 16 |
91200 |
200 |
8 |
8 |
354.32 |
263.56 |
128 = 8 * 16 |
128000 |
200 |
8 |
16 |
666.03 |
529.79 |
256 = 16 * 16 |
184320 |
256 |
16 |
16 |
1268.00 |
1004.85 |
- P = number of processes
- N = number of nodes
- T = number or processes per node
Results obtained on 10-25-2009 using:
- X10 11743
- LU 11751
- SMT on, binding on
LU Source Files
LU Log Files
LU Code
LU.x10
Labels