Datasheet

Ultrascalability combined with extreme availability
features and industry-leading performance
Please see the Legal Information section for important notices and information.
5
the bandwidth of all the memory channels is utilized to support the capability of the processor.
So, as the channels are decreased, the burden to support the requisite bandwidth is increased
on the remaining channels, causing them to become a bottleneck.
For peak performance:
• Always populate processors with equal amounts of memory to enable a balanced NUMA
system.
• Always populate both memory controllers with equal memory capacity.
• Ensure an even number of ranks are populated per channel.
• Use dual-rank DIMMs whenever appropriate.
Power guidelines:
• Fewer larger DIMMs (for example 8 x 4GB DIMMs vs. 16 x 2GB DIMMs will generally have
lower power requirements
• x8 DIMMs (x8 data width of rank) will generally draw less power than equivalently sized x4
DIMMs
Reliability guidelines:
• Using fewer, larger DIMMs (for example 8 x 4 GB DIMMs vs. 16 x 2GB DIMMs is generally
more reliable
• Xeon 7500 Series memory controllers support IBM Chipkill memory protection technology
with x4 DIMMs (x4 data width of rank) and x8 DIMMs
For increased availability beyond Chipkill error correction, the eX5 servers offer additional levels
of IBM Active Memory protection:
Memory ProteXion technology provides multichip error protection and works in conjunction with
Chipkill technology—which provides multibit protection per chip—and standard ECC protection,
to provide multi-level memory correction.
With memory rank sparing, two unused ranks per memory card are configured as spares.
These spare ranks are used in the event of a memory failure on the same memory card.
Memory mirroring works much like disk mirroring. The total memory is divided into two
channels. Data is written concurrently to both channels. If a DIMM fails in one of the DIMMs in
the primary channel, it is instantly disabled and the mirrored (backup) memory in the other
channel becomes active (primary) until the failing DIMM is replaced. With mirroring enabled,
one-half of total memory is available for use. (Note: Due to the double writes to memory,
performance is affected.) Mirroring is handled at the hardware level; no operating system
support is required.
Memory scrubbing is an automatic daily test of all system memory. It detects and reports
memory errors that might be developing before they cause a server outage. Memory scrubbing
and Memory ProteXion work together. When a bit error is detected, memory scrubbing
determines whether the error is recoverable. If the error is recoverable Memory ProteXion
technology will write the data to new location; if it is not recoverable, scrubbing sends an alert to
light path diagnostics, which then notifies IBM Systems Director.
Memory is available in 2GB, 4GB, 8GB, or 16GB DIMMs.
High-Performance Scalability
Scalability can be achieved in two different ways. You can choose to scale “natively” by simply
cabling two chassis together via the Quick Path Interconnect (QPI) ports, for 8 sockets and 128
DIMMs. Or, you can scale via MAX5, adding memory capacity for increased performance. This
configuration cables a 4-socket chassis to a MAX5 containing 32 DIMM slots, for a total of 4
sockets with up to 96 DIMMs.
Up to four QPI cables connect the MAX5 to the host x3850/x3950 X5 at speeds up to 6.4Gbps
(processor-specific), providing exceptional interchassis throughput (100GBps aggregate).
Because of this design, an x3850/x3950 X5 server has the incredible flexibility of starting out as
a 4-socket server with up to 1TB of memory, seven PCIe adapter slots, and eight HDD bays or
16 SSD bays, and then doubling the processors, cache, slots, and HDD bays—or by increasing
only the memory capacity by 50%—while maintaining best-in-class performance.
The MAX5 scalability and memory expansion feature is a unique IBM enhancement, not offered
by other x86 server architectures.