Copyright University of California, Irvine, 2004.
For problems or questions regarding this web, contact
Mahesh Mamidipaka.
Last updated: 10/06/04.
For a given set of cache parameters, cache size (C), block size (B),
associativity (A), and technology, eCACTI can be used to:
(1) estimate power dissipation (both static and dynamic) in caches
for both read and write operations
(2) estimate the cache read access time
(3) estimate the cache area
(4) find the optimal cache configuration in terms of the sextuplet
(Ndwl, Ndbl, Nspd, Ntwl, Ntbl, Ntspd)
eCACTI was developed on Sun OS 5.8. It can be run command line
in the following way:
eCACTI C B A TECH NSubbanks [paramFile]
(or)
eCACTI C B A TECH RWP ERP EWP NSubbanks [ParamFile]
where, C is the cache size, B is the block size, A is the
associativity, TECH is the technology in microns.
paramFile is an optional input file that can be used to specify few other parameters.
Parameters that can specified in paramFile are the following:
-verbose # turn verbose on; default is off.
-config < Ndwl >:< Nspd > :< Ndbl > :< Ntwl > :< Ntspd > :< Ntbl >
# find the area, power, and, access time for the specified configuration
# default is to explore all configurations and find optimal power configuration
-dualVt # specifies that the cache design is based on dual-Vt technology
# default is NO dual-Vt technology
A sample parameter file and command line arguments for running eCACTI are
indicated below.
For these specified inputs, eCACTI estimates the
power, area, and access time for cache of size
16 KB,
block size 32 B,
and associativity of 2 in 0.13u technology and for 1:1:2:1:1:4 configuration.
The input also requests verbose and the estimates for dualVt technology
based cache design.
Example command line arguments:
eCACTI 16384 32 2 0.13 1 params
Example parameter file (params) contents:
-verbose
-config 1:1:2:1:1:4
-dualVt
Following is the output of eCACTI for the following command line arguments.
Command: eCACTI 16384 32 2 0.13 1
Output:
--------------------------------
---------- eCACTI 1.0 ----------
--------------------------------
Cache Subarray Parameters (C, B, A): (16384 B, 32 B, 2)
Number of Subarrays: 1
Ports (RW, R, W): (1, 0, 0)
Technology: 0.13 um, Vdd: 1.1 V
#Cache configurations expressed in (Ndwl, Nspd, Ndbl, Ntwl, Ntspd, Ntbl) format
Optimal Power-AccessTime-Area Config: (1, 1, 1, 4, 1, 1)
Power Stats:
------------
Read hit power: 94.0955 mw
Write hit power: 92.8738 mw
Read miss power: 182.2022 mw
Write miss power: 102.2757 mw
Timing Stats:
-------------
Access time: 1.49748 ns
Cycle time (wave pipelined): 0.700554 ns
Area Stats:
-----------
Aspect ratio (height/width): 1.082291
Total area one subarray : 0.010465 cm^2
Following is the output of eCACTI for the following command line arguments
(file params has -verbose option).
Command: eCACTI 16384 32 2 0.13 1 params
Output:
--------------------------------
---------- eCACTI 1.0 ----------
--------------------------------
Cache Subarray Parameters (C, B, A): (16384 B, 32 B, 2)
Number of Subarrays: 1
Ports (RW, R, W): (1, 0, 0)
Technology: 0.13 um, Vdd: 1.1 V
#Cache configurations expressed in (Ndwl, Nspd, Ndbl, Ntwl, Ntspd, Ntbl) format
Area efficient configuration: (1, 1, 32, 16, 1, 1)
Area = 3.777677, Time = 5.488263 ns, Power = 10287.37 (6118.55033 4168.81488) mW; 6716.21 (6038.79672 677.40940) mW;
Optimal time configuration: (1, 1, 1, 2, 1, 8)
Area = 70.620714, Time = 0.841453 ns, Power = 236.50 (113.63572 122.86573) mW; 132.49 (112.54673 19.94412) mW;
Lowest power configuration: (1, 1, 4, 1, 1, 4)
Area = 66.791649, Time = 1.463368 ns, Power = 61.41 (42.96684 18.44211) mW; 35.73 (32.75304 2.97417) mW;
Optimal Power-AccessTime-Area Config: (1, 1, 1, 4, 1, 1)
Power Stats:
------------
Read hit power: 94.0955 mw
Write hit power: 92.8738 mw
Read miss power: 182.2022 mw
Write miss power: 102.2757 mw
Timing Stats:
-------------
Access time: 1.49748 ns
Cycle time (wave pipelined): 0.700554 ns
Area Stats:
-----------
Aspect ratio (height/width): 1.082291
Total area one subarray : 0.010465 cm^2
Power Components (dyn, lkg mW):
-------------------------------
#NOTE: The power values are for a cache read hit
Total Power all Banks : 76.3, 16.39
Data Array Split:
-----------------
decode : 0.587, 0.3759
wordline : 0.3529, 0.0007582
bitline : 52.34, 13.37
bitline : 3.242, 0.8357
sense_amp : 0.7294, 0.5283
read_control : 0.1605, 0.001591
write_control : 0, 0.001606
tag comparison : 0.3516, 0.01415
mux driver : 2.254, 0.02931
sel inverter : 0.008325, 0
data output driver : 6.199, 0.4265
Time Components:
----------------
data side (with Output driver) (ns): 0.836819
tag side (with Output driver) (ns): 1.49748
subbank address routing delay (ns): 0
decode_data (ns): 0.147709
wordline and bitline data (ns) : 0.458511
sense_amp_data (ns): 0.134875
decode_tag (ns): 0.700554
wordline and bitline tag (ns): 0.238861
sense_amp_tag (ns): 0.082875
compare time (ns): 0.165663
mux driver (ns): 0.165914
sel inverter (ns): 0.0478887
data output driver (ns): 0.095724
total_out_driver (ns): 0
total data path (without output driver) (ns): 0.741095
total tag path is set assoc (ns): 1.40176
Area Components:
----------------
Aspect Ratio Total height/width: 1.082291
Data array (cm^2): 0.008745
Data predecode (cm^2): 0.000043
Data colmux predecode (cm^2): 0.000010
Data colmux post decode (cm^2): 0.000000
Data write signal (cm^2): 0.000052
Tag array (cm^2): 0.001494
Tag predecode (cm^2): 0.000043
Tag colmux predecode (cm^2): 0.000010
Tag colmux post decode (cm^2): 0.000000
Tag output driver decode (cm^2): 0.000015
Tag output driver enable signals (cm^2): 0.000052
Percentage of data ramcells alone of total area: 69.940943 %
Percentage of tag ramcells alone of total area: 4.371309 %
Percentage of total control/routing alone of total area: 25.687748 %
Subbank Efficiency : 74.312252
Total Efficiency : 79.870352
Total area One Subbank (cm^2): 0.010465
Total area subbanked (cm^2): 0.009737
Copyright University of California, Irvine, 2004.
For problems or questions regarding this web, contact
Mahesh Mamidipaka.
Last updated: 10/06/04.