Family paper

Better Energy Efficiency through Advanced
Power and Thermal Management
Power consumption is increasingly important in today’s data centers,
and considerable effort was devoted to optimizing performance
per Watt in the Intel Itanium processor 9300 series. The previous-
generation processor supported Demand Based Switching (DBS),
in which the OS monitors activity and throttles frequency when
workloads are light to improve energy efficiency.
The Intel Itanium processor 9300 series provides enhanced power
and thermal management. This includes an enhanced form of DBS,
in which both voltage and frequency are modulated to enable even
better energy efficiency. It also includes Intel® Turbo Boost Technology,
4
which can automatically increase frequency and voltage settings when
workloads are high to provide the best possible performance without
exceeding the processor’s thermal design power (TDP) envelope.
Performance is maximized by monitoring 120 events in each of the
four cores and adjusting the core voltage and frequency every 6μs.
This function can be used in conjunction with DBS to fine-tune perfor-
mance versus power consumption and to meet different needs at
different times (e.g., peak performance for end-of-month financial
closes and peak energy efficiency when workloads are relatively light).
Improvements in Reliability, Availability
and Serviceability (RAS)
With the ongoing move toward real-time business models, data
integrity and uninterrupted operation are becoming more important
than ever in mission-critical computing environments. Data errors can
be propagated quickly across multiple applications and downtime can
have a major impact on revenue, customer satisfaction, vendor
relationships and brand perception.
Consolidation of multiple applications per system further magnifies
the potential impact of system downtime. A fatal error in a virtualized
server could potentially bring down all hosted applications. Systems
must therefore be highly resilient in order to maintain service levels
while capturing the full benefits of high consolidation ratios (lower
costs, simpler management and less drain on data center resources).
To address growing needs, the Intel Itanium processor 9300 series
extends the mainframe-class Reliability, Availability and Serviceability
(RAS) features that were integrated into previous-generations of
Intel Itanium processors.
In any computing system, errors in data or processing can impact
data reliability, system availability or both. There are basically
two kinds of errors:
• Soft errors are typically caused by an alpha particle or other micro
event that changes the logic state of one or more silicon gates.
These are transient errors and can be fixed simply by correcting
the logic state.
• Hard errors are more persistent. They imply a failure at the
hardware level and require a permanent fix to ensure correct
and uninterrupted operation.
The Intel Itanium processor 9300 series incorporates extensive
capabilities for detecting, correcting and reporting both kinds of
errors on the processor die, in attached components and along
the pathways that connect those components (Figure 2 on
the next page).
Processor RAS Features
An extensive set of RAS features are supported in the processor
die itself.
• Error Avoidance, Detection and Correction Across All Major Core
Structures: Single bit errors are avoided, detected and corrected
throughout each core using a variety of mechanisms, including:
− Soft error (SE) hardened latches and registers
5
: These new
circuit topologies were designed by Intel to improve resistance to
soft errors (see the sidebar, Reducing Soft Errors in Latches by
Up to 100x).
− Error correcting code (ECC) or parity: These widely used
algorithms are implemented in hardware to monitor data and
identify errors that occur during transmission.
− Intel® Cache Safe technology: Heuristics are used to monitor the
number of errors per cache index and map out bad cache lines if
error frequencies reach a specified threshold. Cache data is also
automatically scrubbed to correct single bit errors that might
otherwise accumulate and create uncorrectable problems. Intel®
Cache Safe technology has been extended in the Intel Itanium
processor 9300 series to cover the second-level and directory
cache arrays. In the previous generation, only the third-level
cache was covered.
6
White Paper: The Intel® Itanium® Processor 9300 Series