scieee Science in your language
[en] (orig)

DNA Sequences Alignment in Multi-GPUs: Energy Payoff on Speculative Executions

Pérez-Serrano, Jesús,Sandes, Edans,Melo, Alba,Ujaldon-Martínez, Manuel · 2017

Abstract

We present a performance per watt analysis of CUDAlign 4.0, a parallel strategy to obtain the optimal alignment of huge DNA se- quences in multi-GPU platforms using the exact Smith-Waterman method. Speed-up factors and energy consumption are monitored on different stages of the algorithm with the goal of identifying advantageous sce- narios to maximize acceleration and minimize power consumption. Ex- perimental results using CUDA on a set of GeForce GTX 980 GPUs illustrate their capabilities as high-performance and low-power devices, with a energy cost to be more attractive when increasing the number of GPUs. Overall, our results demonstrate a good correlation between the performance attained and the extra energy required, even in scenarios where multi-GPUs do not show great scalability.

Read accessible full text

DNA Sequences Alignment in Multi-GPUs: Energy Payoff on Speculative Executions

Author: Pérez-Serrano, Jesús,Sandes, Edans,Melo, Alba,Ujaldon-Martínez, Manuel
Year: 2017
Source: https://riuma.uma.es/xmlui/bitstream/10630/13777/1/CUDAlign-speculative-v3-final-and-best.pdf
DNA Sequences Alignmen in Mul i-GPUs:
Ene gy Payo on Specula i e Execu ions
Au ho s:
J. Pé ez, M. Ujaldón E. Sandes, A. Melo
Compu e A chi ec u e Depa men Compu e Science Depa men
Uni e si y o Malaga (Spain) Uni e si y o B asilia (B azil)
P esen e :
M. Ujaldón
Full P o esso @ Compu e A chi ec u e: >100 esea ch pape s published
CUDA Fellow @ NVIDIA: >100 alks/cou ses o e he pas 5 yea s
M. Ujaldón
Talk ou line [20 slides]
1. Backg ound and mo i a ion [4]
2. DNA sequence compa ison [3]
3. CUDAlign [3]
4. Expe imen al se up [3]
5. Expe imen al esul s [4]
6. Specula i e execu ions [3]
I. Backg ound and mo i a ion
M. Ujaldón
The Smi h-Wa e man algo i hm (SW)
SW is a well known biomedical applica ion o compu e:
1. The exac pai wise compa ison o DNA/RNA sequences.
2. A p o ein sequence (que y) o a genomic da abase.
Fine-g ained pa allelism applies be e o 1.
Al eady po ed o mul i-GPUs and Xeon Phis.
Huge da a olume (“big da a”):
Tens-Hund eds o Million Base Pai s
each sequence in ou s udy.
Se e al Pe a-cells (250) o he
dynamic p og amming ma ix used.
4
M. Ujaldón
P ima y goals o his wo k
We p esen a pe o mance pe wa analysis o SW using
CUDAlign 4.0, iden i ying ad an ageous scena ios o maxi-
mize speed-up and minimize powe consump ion on GPUs.
We e alua e:
1. Speed-up, scalabili y and powe on mul i-GPU sys ems.
2. How he wo kload size in luences ene gy in da a-in ensi e applics.
3. The ene gy o e head on specula i e execu ions.
5

M. Ujaldón
GPU accele a ion when ene gy ma e s:
Iden i ying good and bad scena ios
Example: Accele a ion e sus uel consump ion in my ca .
When is i wo h o inc ease 10 MPH?
6
D i ing a 60 MPH: 16.66% mo e speed. 10% mo e gas.
D i ing a 100 MPH: 10% mo e speed. 25% mo e gas.
You can sa e uel up o 4x i you p ess he h o le wisely.
M. Ujaldón
Specula i e execu ions: E olu ion
Back in he 80’s and 90’s:
Lo s o success ul s o ies
(b anch p edic ion, p e e ching,
look-ahead).
Because you gamble wi hou
isk, you a e agg esi e on be s.
7
Now ha ene gy ma e s:
Miss-p edic ions cause powe
consump ion and no execu ion
gains.
So you a e undecided o play
e en wi h good ca ds.
II. DNA Sequence Compa ison
M. Ujaldón
DNA Sequence Compa ison
A DNA sequence is ep esen ed by an o de ed lis o
nucleo ide bases, s ings o he {A, C, G, T} alphabe .
Sco e unc ion: Example:
+1 o a ma ch.
-1 o a misma ch.
-2 whene e you ind a gap.
Smi h-Wa e man algo i hm ob ains he op imal pai wise
local alignmen in quad a ic ime and space. Two phases:
Calcula e he dynamic p og amming ma ix.
Ob ain he alignmen ( aceback).
9
IV. Expe imen al se up

M. Ujaldón
Pla o ms used
17
Ha dwa e esou ces
N idia GPUs used
N idia GPUs used
In el CPU
Comme cial model
Numbe o co es
Co es equency
Memo y size and amily
Memo y equency
Memo y wid h
Memo y bandwid h
So wa e ins alled
GeFo ce GTX 980
Ti an Pascal
Xeon E5-2620
2048
3584
8
1126 MHz
1531 MHz
2100 MHz
4 Gby es GDDR5
12 Gby es GDDR5X
64 Gby es DDR4
7 GHz
10 GHz
2.4 GHz
384 bi
384 bi s
256
336 GB/s.
480 GB/s.
76.8 GB/s.
CUDA 8.0
CUDA 8.0
Ubun u 14.04 LTS 64 bi s
M. Ujaldón
Inpu da a se
Real DNA sequences coming om he Na ional Cen e o
Bio echnology (NCBI) da abase. Compa ison all human
(GRCh37) and Chimpanzee (panT o4) homologous ch omo-
somes. Among he 25 pai s o sequences, we ha e chosen:
18
Inpu
seq.
Inpu
seq.
Size
Size
Pe a
Cells
Sco e
Leng h
Co e age
Ma ches
Misma ches
Gaps
Human
Chimp.
Pe a
Cells
Sco e
Leng h
Co e age
Ma ches
Misma ches
Gaps
ch 22
ch 21
47M
ch Y
51M
50M
2.55
31.510.791
51.929.087
98.9%
88.5%
3.8%
7.7%
48M
46M
2.24
36.006.054
48.579.349
99.0%
91.9%
1.1%
7.1%
47M
33M
1.54
27.206.434
33.583.457
70.5%
94.4%
1.5%
4.1%
59M
26M
1.56
1.394.673
2.283.191
6.0%
88.1%
2.0%
10.0%
M. Ujaldón
Moni o ing ene gy
Beagleblone Black open-sou ce ha dwa e.
Accelpowe module wi h 8 senso s:
19
V. Expe imen al esul s
M. Ujaldón
Powe , ime and ene gy on ou GTX980 GPUs
21
Sequence
A e age powe (wa s pe GPU)
S age 1
S age 2
S age 3
A e age powe (wa s pe GPU)
A e age powe (wa s pe GPU)
A e age powe (wa s pe GPU)
ch 22
ch 21
47M
ch Y
101.11 W.
116.26 W.
77.27 W.
102.11 W.
116.47 W.
78.89 W.
104.37 W.
117.12 W.
76.33 W.
103.25 W.
119.63 W.
0.00 W.
Execu ion ime (seconds)
Execu ion ime (seconds)
Execu ion ime (seconds)
Execu ion ime (seconds)
To al ime
ch 22
ch 21
47M
ch Y
11161.92 s.
185.20 s.
14.25 s.
11361.38 s.
9687.36 s.
61.49 s.
11.03 s.
9759.89 s.
6694.95 s.
88.25 s.
9.05 s.
6792.26 s.
6798.12 s.
3.99 s.
0.00 s.
6802.11 s.
Ene gy consump ion (kilojules pe GPU)
Ene gy consump ion (kilojules pe GPU)
Ene gy consump ion (kilojules pe GPU)
Ene gy consump ion (kilojules pe GPU)
To al ene gy
To al cos
ch 22
ch 21
47M
ch Y
1128.63 kJ.
21.53 kJ.
1.10 kJ.
1151.27 kJ.
0.1660 €
989.26 kJ.
7.16 kJ.
0.87 kJ.
997.29 kJ.
0.1440 €
698.82 kJ.
10.34 kJ.
0.69 kJ.
709.85 kJ.
0.1024 €
701.94 kJ.
0.48 kJ.
0.00 kJ.
702.42 kJ.
0.1012 €

M. Ujaldón
Powe , ime and ene gy o ch 22 on mul i-GPU
22
No. GPUs
A e age powe (wa s pe GPU)
S age 1
S age 2
S age 3
A e age powe (wa s pe GPU)
A e age powe (wa s pe GPU)
A e age powe (wa s pe GPU)
4
3
2
1
101.11 W.
116.26 W.
77.27 W.
101.53 W.
108.16 W.
78.79 W.
100.30 W.
114.68 W.
76.74 W.
102.95 W.
114.44 W.
81.27 W.
Execu ion ime (seconds)
Execu ion ime (seconds)
Execu ion ime (seconds)
Execu ion ime (seconds)
To al ime
4
3
2
1
11161.92 s.
185.20 s.
14.25 s.
11361.38 s.
14719.32 s.
253.72 s.
17.70 s.
14990.76 s.
22080.04 s.
159.77 s.
23.17 s.
22262.99 s.
22302.24 s.
291.50 s.
46.65 s.
22640.40 s.
Ene gy consump ion (kilojules pe GPU)
Ene gy consump ion (kilojules pe GPU)
Ene gy consump ion (kilojules pe GPU)
Ene gy consump ion (kilojules pe GPU)
To al ene gy
To al cos
4
3
2
1
1128.63 kJ.
21.53 kJ.
1.10 kJ.
1151.27 kJ.
0.1660 €
1494.60 kJ.
27.45 kJ.
1.40 kJ.
1523.44 kJ.
0.1650 €
2214.77 kJ.
18.32 kJ.
1.78 kJ.
2234.88 kJ.
0.1614 €
2296.22 kJ.
33.36 kJ.
3.79 kJ.
2333.37 kJ.
0.0842 €
M. Ujaldón
Powe consump ion on 4 GTX 980 GPUs
s age by s age
23
M. Ujaldón
Time sa ings and ene gy penal ies (ch 22)
24
No.
GPUs
S age 1
S age 1
S age 2
S age 2
S age 3
S age 3
To al
To al
Sa ings
( ime)
Penal y
(ene gy)
Sa ings
( ime)
Penal y
(ene gy)
Sa ings
( ime)
Penal y
(ene gy)
Sa ings
( ime)
Penal y
(ene gy)
Fou
Th ee
Two
49.96%
96.60%
36.47%
158.15%
69.46%
6.09%
49.82%
97.35%
34.01%
95.26%
12.97%
146.85%
62.06%
0.81%
33.79%
95.86%
1.00%
92.90%
45.20%
9.83%
50.34%
-6.07%
1.67%
91.55%
VI. Specula i e execu ions