Information Technology Reference
In-Depth Information
14
np=7
12
np=6
10
np=5
np=4
8
np=3
6
np=2
4
np=1
2
0
1
2
3
4
nt
Fig. 8.
Speedup for Loop/Task Partitioning
data bus can be accessed by one core only at any given moment, the other cores
must wait for their turns (the sequential section
S
in the upper right half of the
figure.) For this reason, thread concurrency decreases in this section. Neverthe-
less, the measurement error in the performance increases with the number of
processing units. Logical cores always exhibited a lower performance than phys-
ical cores and can not be compared. Operating Systems always show these cores
(logical and physical cores) as comparable and the general consumers usually
get awry judgements about their real performance.
Fig. 9.
Serialization effects
Regarding the parallelization of the density matrix, the use of hybrid models
exhibited a good performance where the speedup increased until 10
.
5
X
(Task
Partitioning) to 12
.
2
X
(Loop Partitioning) using the available cluster platform.
