Comparing the Performance of Arm Based and Traditional Computers For Drug Discovery by Forrest Kamperman
1. Comparing the Performance of ARM-Based
and Traditional Computers for Drug Discovery
Forrest Kamperman, Centre College
Abstract:
This summer I did computer hardware testing at Centre
with Dr. David Toth. We tested two different types of
computers, ARM based systems and x86/x64 systems.
We used molecular docking software as a benchmark to
test their performance, and also looked at the prices of
the systems and their power consumption to find the
system that gave the best performance per dollar, and
the system that consumed the least energy. This
information would be useful for anyone looking to do
large scale calculations, especially those doing drug
discovery work. This summer also gave me a good
opportunity to view if hardware testing is the kind of work
I would like to do in the future, and into the different
aspects of writing a professional scientific paper.
$45.03 $52.35
$510.36
$141.23
$32.98
$0
$100
$200
$300
$400
$500
$600
8-core System (x64) 64-core System (x64) Raspberry Pi (ARM) Cubieboard2 (ARM) Odroid U3
(ARM)
Cost
Computer
Hardware Cost To Screen 10,000 Compounds in 24 Hours
0.28
0.15
0.60
0.11
0.06
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
8-core System (x64) 64-core System (x64) Raspberry Pi (ARM) Cubieboard2 (ARM) Odroid U3
(ARM)
Power(kWh)
Computer
Power Consumed Per 10,000 Compounds Screened
Raspberry Pi Model
B
Cubieboard2 Odroid U3
Background:
There has been a recent push in computer science
towards finding more energy efficient ways of doing
large numbers of calculations, in order to lower power
consumption at large data centers. However, people
do not want to sacrifice the performance that they are
used to. One field in particular where high
performance is critical is virtual screening, where the
goal is to go through as many compounds as you can
in order to narrow it down to those most likely to work.
We decided to test out ARM based systems, which
are less powerful than the traditonal systems, but are
also much cheaper, and consume less power.
Devices:
Most of my testing was on the three ARM based mini-computers we were testing.
All three are available for under $100, and are much smaller than traditional
systems.
Cubieboard2:
This device costs around $75, contains two cores and 1 GB of RAM.
Raspberry Pi:
This is the best known device, but we found it to be the worst of the three. It costs
around $50, contains only a single core and has 512 MB of RAM.
Odroid U3:
This device cost a little over $75, and was the best system in our testing. It has
four cores, and has 2 GB of RAM.
Future Work:
After having worked at the Oregon Governor’s office last year,
my work this year was completely different. This year I worked
on research, mostly on my own, compared to working with the
public last summer. What I found is that I prefer research and
the computer science field in general to government work.
Next summer I hope to find work an internship in the private
sector, hopefully working in software development. While I do
prefer hardware testing to government work, I prefer working
with software to working with hardware.
System Cost
Compounds
Screened in
24 hours
Power (KWH)
consumed in
24 hours
Compounds
Screened
per KWH
Compounds
Screened per
Dollar in 24
Hours
64-Core
Server $5,015.01 960,116 14.91 64,394 191
8-core x86 $650.00 144,390 4.04 35,740 222
Odroid $76.50 23,853 0.14 170,379 312
CubieBoard2 $78.50 5,555 0.06 92,585 71
Raspberry Pi $51.00 999 0.06 16,655 20
Findings:
After all of our testing was finished, we had a clear winner.
In terms of both cost and power efficiency, the Odroid U3
was the clear winner. It required about half as much power
as the next closest device to screen 10,000 compounds,
and cost more than $10 less per 10,000 compounds
screened. Even more clear was which device was the
worst. The Raspberry Pi clearly was the least efficient
device, costing more than three times as much as the next
device to produce the same number of compounds, and
required more than double the power of the next device.
The Odroid also clearly was better than the traditional
systems, costing much less, and providing much more bang
for your buck. To build off of this, we hope to test other ARM
based systems that are coming out soon, to see if they are
just as efficient.