HP ProLiant BL620c G7, User Manual

Page: 39 / 123

Hardware options installation 39
•
AMP modes Advanced ECC, DDDC, HP Memory Quarantine, Online Spare, and Mirrored Memory
have further requirements beyond the ones listed here. For additional memory configuration
requirements, see the corresponding AMP sections:
o
Advanced ECC memory population guidelines (on page 44 )
o
Double Device Data Correction (on page 44 )
o
HP Memory Quarantine (on page 45 )
o
Online Spare memory population guidelines (on page 45 )
o
Mirrored Memory population guidelines (on page 46 )
Memory subsystem architecture
The Intel® Xeon® E7 family and 6500/7500 series processor memory architecture is designed to take
advantage of multiple stages of memory interleaving to reduce latency and increase bandwidth.
Each Intel Xeon E7 family and 6500/7500 series processor contains two memory controllers as shown in the
illustration below. Each memory controller has two SMI buses operating in Lockstep mode. Each SMI bus
connects to an SMB or buffer as shown in the illustration below. The buffer converts the SMI to DDR3 and
expands the memory capacity of the system. Each buffer has two DDR3 channels and can support up to four
DIMMs for a total of 16 DIMMs per processor, or 32 DIMMs per HP ProLiant BL620c G7 Server Blade with
both processors installed.
Memory speed is not affected by number of DIMMs, ranks, or voltage. All DIMMs run at the highest possible
speed for a given processor.
DDR3 memory speed is a function of the QPI bus speed supported by the processor:
•
Processors with a QPI speed of 6.4 GT/s run memory at 1066 MT/s.
•
Processors with a QPI speed of 5.6 GT/s run memory at 978 MT/s.
•
Processors with a QPI speed of 4.8 GT/s run memory at 800 MT/s.
Successive cache lines are interleaved between the DIMMs and the Lockstep SMI channels of the two
memory controllers in the processor such that adjacent cache lines reside on different memory controllers,
SMIs, DIMMs, and DIMM ranks for better performance. To take advantage of this feature, DIMMs should be
populated evenly between all SMI channels. If an SMI channel pair has more DIMMs than others, the extra
memory on that SMI channel pair does not benefit from the interleaving mechanism across memory
controllers.
Memory architecture for processor 1
Channel Slot Slot number
1
A
E 4
3
2
C
G
2
1
3
A
E 5
6
4
C
G 7
8
5
B
F 12
11
6
D 10