Dell PowerEdge 1655MC - Manual (Page 21)

Optimal BIOS settings for HPC with Dell PowerEdge 12 th generation servers
21
Results are shown in Figure 7. The bar graph plots the performance of the application with Node
Interleaving on relative to Node Interleaving off. A value higher than one indicates that the
application performed that much better with Node Interleaving enabled. The text in the bar
indicates the percentage of additional power consumed with Node Interleaving enabled. A value
that is positive indicates that when Node Interleaving was enabled, the cluster consumed more
power than it did when Node Interleaving was disabled. The secondary y-axis plots the associated
energy efficiency. A marker higher than one indicates the energy efficiency of the application was
that much better with Node Interleaving enabled when compared to Node Interleaving disabled.
The results show that enabling Node interleaving does not benefit any of the applications. The
actual impact is application dependent, but all applications do better with Node Interleaving
disabled. LU showed a 1% improvement with Node Interleaving enabled, but that is statistically too
small an improvement to be conclusive.
Based on these results it is recommended that Node Interleaving be set to disabled. However, if
the cluster is running applications that benefit from interleaved memory or if the situation is such
that interleaved memory is beneficial (as described in Section 3.3), then this option should be
tuned as needed.
5.5. Logical Processor
The impact of Logical Processor is presented in Figure 8.
Figure 8. Performance and Energy Efficiency of Logical Processor
* Data collected on a 16-server PowerEdge M620 cluster. Dual Intel Xeon E5-2680 @ 2.7GHz, 8*8GB 1600MT/s
memory per server. Mellanox InfiniBand FDR. Node Interleaving disabled. DAPC System Profile.
+
2%
pow
er
+
7%
po
w
er
+
5%
po
w
er
+
10%
pow
er
+
11
%
p
o
w
er
+
11
%
pow
er
+
6
%
Pow
er
+
2%
power
+
17
%
power
+
10%
power
+
8%
power
+
20%
power
+
5%
power
+
22
%
Power
0.97
0.92
0.95
0.90 0.89
0.85
0.94
0.53
0.93
1.01
0.65
0.44
1.13
1.10
0.00
0.20
0.40
0.60
0.80
1.00
1.20
0.20
0.40
0.60
0.80
1.00
1.20
1.40
HPL Fluent
truck_poly_14m
Fluent
truck_111m
WRF
conus 12km
NAMD
stmv
MILC
Intel input file
LU
class D
Pe
rf
or
manc
e/
W
at
t r
elative
to L
og
ic
al
Pr
oc
es
sor
off
(hig
her
me
ans
HT
on
is
bet
ter
)
Pe
rf
or
manc
e
with
Log
ic
al
Pr
oc
es
sor
on
on
relative
to of
f
(hig
her
me
ans
HT
on
is
bet
ter
)
Impact of Logical Processor
Performance-256 thr Performance-512 thr Performance/Watt-256 thr Performance/Watt-512 thr