BenchADM Benchmarking Application
Description
The BenchADM benchmark
application is the computational kernel representative of
many applications in numerical relativity. It is
characterised by being compute intensive, with around 600
Floating point operations per grid point and with six grid
functions synchronised each timestep.
Information
- ThornList for BenchADM: BenchADM.th
- Parameter file for BenchADM: In CactusBench arrangement,
BenchADM/par/BenchADM_40l.par
Floating Point Operation Count
Floating point operations are counted using perfex
on our Origin 2000 (using perfex -e 21). The origin
counts a multiply and add as a single operation, so we
list numbers with this switched both on and off
(compile with -TARG:madd=OFF to switch it off).
|
|
40-cubed
|
50-cubed
|
60-cubed |
70-cubed |
80-cubed |
|
Number of grid points
|
64000
|
125000
|
216000
|
343000
|
512000
|
|
Standard optimisations and madd=OFF
|
4,197,057,059
|
8,473,076,926
|
14,049,337,624
|
18,702,999,402
|
28,219,247,382
|
|
Standard optimisations and madd=ON
|
3,280,705,059
|
6,584,368,126
|
11,649,097,035
|
24,065,484,192
|
36,309,100,232
|
|
No optimisations and madd=OFF
|
6,689,714,370
|
13,476,602,909 |
23,769,050,354
|
38,296,994,985
|
57,790,370,995
|
|
![[MAINIMAGE]](../images/development.jpg)
