Can the performance of a computer system be increased through overclocking such that the percentage gain of work performed is greater than the percentage increase of electricity consumed?

1. Can a modern computer be overclocked while increasing its energy efficiency?
By Josh Mullis
Project Design Plan
Computers are used in a variety of applications to organize data, store data, and present data in
a way that is easy for people to understand and recognize. Computers help people get value
and meaning from their data in a much more efficient way than pen and paper alone can do.
Today’s computers are designed with both performance and energy efficiency in mind. The
processor is designed to idle along, then throttle up to its manufactured peak speed only when
it is experiencing a high load. Overclocking of a processor, and the RAM that the computer uses,
is a method used to increase the speed at which they perform, allowing for faster performance
and more work to be done. When overclocking is performed, more electricity is used.
Can the performance of a computer system be increased through overclocking such that the
percentage gain of work performed is greater than the percentage increase of electricity
consumed? If so, the efficiency of the system will be increased and the value of overclocking
will be a plausible solution. This can have great benefits if done on a large scale in a data-center
environment using hundreds of computer systems acting in a server role. If overclocking
doesn’t make the computer more efficient then this is not a plausible solution.
Literature Review
There are many different examples of overclocking computers on the internet. There are far
fewer examples of measuring the efficiency of overclocking. Two examples are cited in this
literature review.
Gavrichenkov tested nine different processors and system configurations, overclocked to
different levels, to determine any gain or loss experienced in efficiency. It was determined that
due to differences in processor architecture, motherboard architecture and system
configuration there is not a definite set of rules to measure efficiency. Due to the differences,
an actual measure of the electricity used must be performed while the system is under load. It
was noted that the test session would be very interesting, due to the popularity of
overclocking. It was also noted that most of today’s modern motherboards allow for simplified
overclocking. Due to changes in manufacturing processes, the advertised limit of a processor is
now mostly determined by the amount of heat generated, rather than the actual limit imposed
by the physics and engineering of the given processor. Because of this, it is much easier to
overclock a system, provided ample consideration is given to cooling, and an adequate cooling
system is used. The article states that with the current technology, the wattage measured will
only rise slightly when the core voltage of the processor is not altered. When the core voltage is
kept at a stock setting, but the frequency of the processor is increased, overclocking yields only

2. a small increase in consumed power. If, however, the core voltage is increased, this can make
overclocking less efficient. (Gavrichenkov 2010)
Schmid and Roos overclocked an Intel i7-3770K, which uses a new 22nm manufacturing design
(much smaller than previous generations), the same processor as the one being used in this
experiment; and noted that the new architecture allows for increased overclocking headroom
without having to raise the core voltage supplied to the processor very much. This behavior
allows for overclocking, without having to significantly raise the power, thus making the
potential for improved energy efficiency more realistic and achievable.
Experimental Design Steps
Performance and peak energy consumption (peak watts) will be measured four different times
as follows:
1. A computer system is custom built using the components listed in the ‘tools and
technologies’ section, item 1.
2. Windows 7 Professional is installed on the test system.
3. A plug power meter is purchased.
4. PassMark® Performance Test™ software is downloaded and installed.
5. Intel® Extreme Tuning Utility (Intel® XTU) version 4.2 is downloaded and installed.
6. One test of the computer is run to obtain a performance baseline and energy
consumption baseline.
7. Three different tests of the computer are run, using incrementally increased levels of
overclocking.
8. The results are measured using the plug power meter and test software.
9. The peak watts and performance are graphed to give a visual representation of the
results.
Reasoning
Plugging the computer’s power cord into the power meter allows for measuring the overall
wattage peak of the computer system. This is one of the measurements a typical data center is
concerned about. This also greatly simplifies the experiment because the need to measure the
individual power draw of every component (processor, RAM, power supply, motherboard, etc.)
is eliminated. By performing four different tests, the performance experienced and watts used
can be reasonably compared. This allows for the first test to act as a baseline for performance
and energy consumption, the second test to measure the effect of overclocked RAM, the third
test to measure the effect of overclocked RAM and processor, and the fourth test to measure
the effect of overclocked RAM and a max overclocked processor.

3. Sequence of Events
1. Plug the Weanas™ Plug Power Meter into the surge protecting outlet.
2. Reset the power meter to zero using a paperclip.
3. Set the power meter to show “Hi” watts by pushing the “function” button.
4. Plug the main power cord for the computer into the power meter.
5. Start the computer and logon to Windows.
6. Download the PassMark® Performance Test™ software from their website and install it.
7. Run the Performance Test™ software using the computer’s stock, non-overclocked
speeds to determine your performance score.
8. Take a screenshot of the completed test score using OneNote, PrtScn or Snipping Tool
and save it.
9. Record the “Hi” watts reading on the power meter.
10. Restart the computer and press the ‘delete’ key to enter the Asus UEFI BIOS.
11. Go to the tuning tab and set the memory to use Intel XMP Profile 2.
12. Press F10 to save the settings and boot to Windows.
13. Reset the power meter to zero using a paperclip.
14. Run the Performance Test™ software a 2nd time.
15. Take a screenshot of the completed test score using OneNote, PrtScn or Snipping Tool
and save it.
16. Record the “Hi” watts reading on the power meter.
17. Download and install the Intel® Extreme Tuning Utility (Intel® XTU) version 4.2.
18. Launch the Intel® Extreme Tuning Utility
19. Change the stock processor multiplier from 39x to 40x on all four cores and press the
apply button.
20. Reset the power meter to zero using a paperclip.
21. Run the Performance Test™ software a 3rd time.
22. Take a screenshot of the completed test score using OneNote, PrtScn or Snipping Tool
and save it.
23. Record the “Hi” watts reading on the power meter.
24. Open the Intel® Extreme Tuning Utility
25. Change the stock processor multiplier from 40x to 43x on all four cores, then increase
the core voltage on the processor 1 step up so it can run stable, and press the apply
button.
26. Reset the power meter to zero using a paperclip.
27. Run the Performance Test™ software a 4th time.
28. Take a screenshot of the completed test score using OneNote, PrtScn or Snipping Tool
and save it.
29. Record the “Hi” watts reading on the power meter.
30. Compare the Performance Test™ score and “Hi” watts reading from all four tests.

4. Tools, Technologies, and Measurement Units
1. Windows 7 Professional workstation that is custom-built to allow overclocking.
a. Rosewill BLACKHAWK Gaming ATX mid-tower case
b. CORSAIR HX750 power supply
c. ASUS P8Z77-V LGA 1155 motherboard
d. Intel i7-3770K CPU
e. Cooler Master Hyper 212 EVO - CPU Cooler
f. 32GB G.SKILL F3-1866C10Q RAM
g. XFX Double D FX-785A-CDFC Radeon HD 7850 2GB video card
h. OCZ Technology 256GB Vertex 4 Solid State Drive (SSD)
i. LG WH14NS40 Blu-Ray burner
2. Weanas™Plug Power Meter http://www.amazon.com/Weanas-Energy-Voltage-
Electricity-Monitor/dp/B00DTMQ1S6/
a. Peak or ‘Hi’ watts will be measured
3. PassMark® Performance Test: http://www.passmark.com/products/pt.htm
a. The Performance Test software runs through a series of tests:
i. CPU tests: Mathematical operations, compression, encryption, SSE,
3DNow! instructions and more
ii. 2D graphics tests: Drawing lines, bitmaps, fonts, text, and GUI elements
iii. 3D graphics tests: Simple to complex DirectX 3D graphics and animations
iv. Disk tests: Reading, writing and seeking within disk files
v. Memory tests: Allocating and accessing memory speed and efficiency
b. A proprietary performance score will be used to determine the computers
overall performance of all the tests run
4. Intel® Extreme Tuning Utility (Intel® XTU) version 4.2:
https://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&ProdId=3483&DwnldID=23
084&keyword=%22%22extreme+tuning+utility%22%22&lang=eng
Variables
Independent variable: The level of overclocking performed on the computer.
Dependent variables: The performance measured as a proprietary test score, and the peak
watts measured using a power meter.
Controlled variables: The same computer is used and the power meter, which is plugged into
the same surge protector and outlet, doesn’t change.

5. Threat Reduction to Internal Validity
Systematic error is reduced through controlled variables, by using the same power meter,
plugged into the same surge protector and outlet. The dependent variables, the performance
test and power meter, are reset to zero before each test to assure an accurate measurement.
The independent variables are written as part of the experiment and delineated so that each
test can be reproduced with the exact hardware and settings. This allows for independent
verification of the findings.
Hypothesis
The hypothesis of this experiment is that the efficiency of an Intel i7-3770K, LGA 1155 computer
system can be increased, while also increasing the system’s speed and ability to do work
through overclocking. This hypothesis was based on the statement from the Schmid and Roos
article: “This behavior allows for overclocking, without having to significantly raise the power,
thus making the potential for improved energy efficiency more realistic and achievable.”
Process of Data Collection
The initial test, test 1, was used as a baseline for performance and electricity usage. This test
measured a peak of 185.9 watts and a Performance Test score of 4708.6.
Peak Watts Performance Test
185.9 4708.6

6. In test 2, the RAM was overclocked using XMP profile 2, while the CPU was not changed. This
test measured a peak of 186 watts and a Performance Test score of 4905.2. The result was a
4.01% increase in performance, and .05% increase in wattage over the baseline.
Peak Watts Performance Test Performance Increase Wattage Increase
186 4905.2 4.01% .05%
In test 3, the RAM was still overclocked using the XMP profile 2, and the CPU was overclocked
by changing the frequency multiplier from 39x to 40x. This changed the maximum frequency of
the CPU from 3.9 GHz to 4.0 GHz. This test measured a peak of 187.8 watts and a Performance
Test score of 4905.2. The result was a 7.18% increase in performance, and 1.01% increase in
wattage over the baseline.
Peak Watts Performance Test Performance Increase Wattage Increase
187.8 5073.2 7.18% 1.01%
In test 4, the RAM was still overclocked using the XMP profile 2, and the CPU was overclocked
by changing the frequency multiplier from 40x to 43x. This changed the maximum frequency of
the CPU from 4.0 GHz to 4.3 GHz. This test measured a peak of 188.5 watts and a Performance
Test score of 5256.2. The result was a 10.42% increase in performance, and 1.37% increase in
wattage over the baseline.
Peak Watts Performance Test Performance Increase Wattage Increase
188.5 5256.2 10.42% 1.37%

7. Appropriate Methods
All tests were performed using the same computer hardware and testing equipment. Peak
wattage is an effective way to measure the energy used since the voltage is constant at 120
volts. The amperage used can be determined by dividing the peak watts by the voltage as
follows:
Test 1 Test 2 Test 3 Test 4
Peak Watts 185.9 186 187.8 188.5
Peak Amps
(Watts/120)
1.54 1.55 1.56 1.57
Since the amperage increases are proportional to peak watts, it is sufficient to measure only the
peak watts. Four tests were used to get a good look at the increases over the baseline of the
first test.
Results
The graph below shows that the performance gained with each successive test was greater
than the energy increase observed. Test 4 produced the greatest delta between energy and
performance indicating it was the most efficient of the tests.
185.9 186 187.8 188.5
4708.6 4905.2 5073.2 5256.2
TEST 1 TEST 2 TEST 3 TEST 4
Energy and Performance
Peak Watts (Energy) Test Score (Performance)

8. The increase in efficiency is better illustrated by graphing the percentage change of each
increase, over test 1 baseline, as shown below. Test 2 had a 4.04% delta while test 4 had a
9.05% delta.
Conclusion
Confirmation of Hypothesis
The results of the experiment confirm and document that the efficiency of an Intel i7-3770K,
LGA 1155 computer system can be increased, while simultaneously increasing the system’s
speed and ability to do work through overclocking.
Experimental Design as a Key Factor
The design of this experiment was important in determining the reliability of the results. A
single independent variable, the level of overclocking performed, was used. If more than one
independent variable is used in an experiment, it alters more than one thing between test
groups, making it impossible to know what caused the measured outcome. A clear objective
was important in the design of the experiment. The objective was taken from the hypothesis, in
this case, to see if the efficiency could be increased. An experimenter must reliably determine
that the independent variable was the cause of the measured difference in results. Controlling
as many variables as possible makes the experiment’s results more reliable. If random factors
are allowed to enter the experiment, the accuracy of the experiment is reduced and the results
are unreliable.
0.05
1.01 1.37
4.01
7.18
10.42
TEST 2 TEST 3 TEST 4
Efficiency Increase
Peak Watts (% Increase) Test Score (% Increase)

9. Replication
Replicating this experiment can be achieved by purchasing the same tools and technologies,
such as computer components, and power meter; then using the same measurement tool, the
performance testing software. The experimenter will need knowledge of modern PC building
techniques to custom-build the computer, and then be able to overclock it.
Evaluation of Validity
As the experiment is replicated by independent parties, the outcome can be verified through
consistent results. Consistency in the results lends credibility to them. If the results are
inconsistent then the design of the experiment must be changed until consistent results can be
obtained.
Sources
1 - Gavrichenkov (Apr 12, 2010) CPU Overclocking vs. Power Consumption,
http://www.xbitlabs.com/articles/cpu/display/power-consumption-overclocking.html
2 - Schmid, Roos (May 23, 2012) Overclocking Core i7-3770K: Learning To Live With
Compromise, http://www.tomshardware.com/reviews/ivy-bridge-overclocking-core-i7-
3770k,3198.html