More Resources
- 12 NPB benchmarks:
Table 1 NPB benchmarks Benchmark
Full Name
Description
IS
Integer Sort
It is used to solve two-dimensional large integer sorting based on the bucket sort, and requires full-switching communication.
EP
Embarrassingly Parallel
It is used to calculate the Gauss pseudo random number and mainly performs floating-point number calculation. The prominent feature of EP is that no communication between processors is performed. Therefore, different results obtained through interconnection show the same execution characteristics.
CG
Conjugate Gradient
It is used to solve the approximation of the minimum eigenvalue of a large sparse symmetric positive definite matrix.
MG
MultiGrid
It is used to monitor short- and long-distance communication. MG is a simplified multi-grid core.
FT
Fast Fourier Transform
It is used to detect long-distance communication and solve 3-dimensional partial differential equation using fast Fourier transform.
BT
Block Tri-Diagonal
It is used to solve the tri-diagonal equation set.
BT-IO
Block Tri-Diagonal-IO
It is used to test different parallel I/O technologies.
SP
Scalar Penta-diagonal
It is used to solve the penta-diagonal equation set.
LU
Lower Upper Triangular
It is used to solve sparse block equations based on symmetric super-relaxation method.
UA
Unstructured Adaptive
It is used to test unstructured adaptive networks, dynamic and irregular memory access.
DC
Data Cube
It is used to test data movement.
DT
Data Traffic
- Eight NPB problem sizes:
Table 2 NPB problem sizes Problem Size
Description
S
The S (sample) class is for sample programs.
W
The W (workstation) class is usually used for workstations.
A-F
The size of A class is the smallest and increases in ascending order.