A Power Model Combined of Architectural Level
and Gate Level for Multicore Processors
Manman Peng
Key Laboratory for Embedded and Network Computing
of Hunan Province
Hunan University
Changsha, China
pengmanman@hnu.edu.cn
Yan Hu
Key Laboratory for Embedded and Network Computing
of Hunan Province
Hunan University
Changsha, China
hnu.huyan@gmail.com
Abstract—Low power consumption is becoming a critical
factor for multicore processors. As the multicore processor
design complexity increases, power estimation for multicore
processors has gained more importance. This paper presents a
new power model combined of architectural level and gate
level for multicore processors. The model maps the multicore
processors to a combination of building blocks, and estimates
the gate-level power of these blocks using parameterized RTL.
Then, the power numbers are made in the form of look-up
tables, and integrated in architecture simulators. The
experiments show that for peak power estimation, an excellent
accuracy has been reached and simulation performance is
greatly improved compared to the gate level.
Keywords-power modeling; gate level; architectural level;
Multicore processors
I. INT ROD UC ION
As VLSI technology is developing rapidly in
complexity and density, power consumption of chips has
become a major concern in the state of the art of high-
performance CPU design. To consume less power but still
get better overall performance, Industry has already shifted
gears to deploy architectures with multiple cores [2] and
large last-level caches [3], so the power consumption of
multi-core and many-core processors deserves to get more
attention.
According to the different design phases of the chip,
the analysis methods of overall power consumption is
divided into the following categories: architecture-level,
RTL-level, gate-level (netlist-level) and transistor-level. As
the power model is refined from the highest level of
abstraction to the lowest, the accuracy and detail of
functional and timing information increase. At the two
extremes, there are many power models/tools [4, 5, 6] have
been proposed. Architecture-level power models [4, 5] are
fast, but ignore the impact of specific circuit implementation
of various factors on power consumption. Gate-level power
models [6] can give precise power dissipation of a circuit
driven by certain input vectors. However, due to the nature
of its simulation process, gate-level estimators always have
the slowest speed, and the circuit netlists must be known
before any simulation could be performed.
In this paper, we introduce a new power model for
multicore processors which is a combination of gate level
and architectural level. Our model uses a series of EDA
tools and DesignWare Library [8] which is a collection of
the industry's most widely used, silicon-proven reusable
intellectual property blocks to get the gate-level power.
Then, these power numbers are made in the form of a
lookup table, and integrated in Gem5 [12] to provide power
estimates. For the components that we can’t get the RTL
hardware description, we still use analytical methods.
Dynamic, short-circuit, and leakage power are all modeled.
The rest of the paper is organized as follows: section 2
discusses related work, section 3 presents building blocks of
a multicore processor, section 4 provides a detailed
description of our power estimation methodology and its
flow, section 5 describes the validation experimental results
and section 6 discusses our conclusions.
II.
RELATED WORK
Wattch [4] is a widely-used architecture-level power
analysis and optimization tool. The power estimation of
Wattch is that the tool integrates parameterized power
models of common structures present in modern superscalar
microprocessors into Simplescalar. Wattch only models
dynamic power consumption and fall the main processor
units into four categories: array structures, fully associative
content-addressable memories, combinational logic and
wires and clocking. The limitation of Wattch is that they do
not necessarily model all of the miscellaneous logic present
in real microprocessors, and use simple linear scaling
models based on 0.8um technology that are inaccurate to
make predictions for current and future deep-submicron
technology nodes.
CACTI [13] is an integrated cache and memory access
time, cycle time, area, leakage, and dynamic power model.
It uses device models based on the industry-standard ITRS
2013 12th IEEE International Conference on Trust, Security and Privacy in Computing and Communications
978-0-7695-5022-0/13 $26.00 © 2013 IEEE
DOI 10.1109/TrustCom.2013.204
1652