Con en s lis s a ailable a ScienceDi ec
Jou nal o Pa allel and Dis ibu ed Compu ing
jou nal homepage: www.else ie .com/loca e/jpdc
The Eu opean mas e o HPC cu iculum
Pascal Bou ya, Ma s B o sson a, Ramon Canalb, A yan E ekha i c, Sieg ied Höfinge d,
Didie Sme s e, Ha ald Kös le , Tomáš Kozubekg, Ezhilma hi K ishnasamy a,, Josep Llosa b,
Alexand a Lukas-Ro he , Xa ie Ma o ellb, Di k Plei e h, ,∗, Ana P oyko a i,
Ma ia-Ribe a Sancho b, Ola Schenkc, C is ina Sil anoj
aPa allel Compu ing & Op imisa ion G oup and FSTM-DCS/SnT, Uni e si y o Luxembou g, Esch-su -Alze e, 4364, Luxembou g
bUni e si a Poli ècnica de Ca alunya, Ba celona, 08034, Spain
cIns i u e o Compu ing, Facul y o In o ma ics, Uni e si à della S izze a i aliana, Lugano, 6962, Swi ze land
dTechnische Uni e si ä Wien, Wien, 1040, Aus ia
eLabo a oi e Jacques-Louis Lions, So bonne Uni e si é, Pa is, 75005, F ance
Depa men o Compu e Science, F ied ich-Alexande -Uni e si ä E langen-Nü nbe g, E langen, 91058, Ge many
gIT4Inno a ions, VSB-Technical Uni e si y o Os a a, Os a a, 70800, Czech Republic
hDi ision o Compu a ional Science and Technology, KTH Royal Ins i u e o Technology, S ockholm, 10044, Sweden
iHPC labo a o y, Facul y o Physics, Uni e si y o Sofia, Sofia, 1164, Bulga ia
jDipa imen o di Ele onica, In o mazione e Bioingegne ia, Poli ecnico Milano, Milano, 20133, I aly
A R T I C L E I N F O A B S T R A C T
Keywo ds:
Compu ing educa ion
High-pe o mance compu ing
Mas e in HPC
Model cu icula
The use o High-Pe o mance Compu ing (HPC) is c ucial o add essing a ious g and challenges. While
significan in es men s a e made in digi al in as uc u es ha comp ise HPC esou ces, i s ealisa ion, ope a ion,
and, in pa icula , i s use c i ically depends on sui ably ained expe s. In his pape , we p esen he esul s o
an effo o design and implemen a pan-Eu opean e e ence cu iculum o a mas e ’s deg ee in HPC.
1. In oduc ion
The use o High-Pe o mance Compu ing (HPC) is c i ical o ad-
d ess many o he g and scien ific, enginee ing, and socie al challenges.
The scien ific challenges equi ing HPC ange om sol ing undamen al
open ques ions o physics a he la ges and smalles scales o challenges
in li e sciences like he decoding o he human b ain. One o he mos
p ominen challenges whe e HPC plays a c ucial ole is clima e adap-
a ion. Model simula ions using HPC sys ems ha e played a key ole
in p edic ing clima e change. Des ina ion Ea h is cu en ly de eloping
la ge-scale clima e and wea he digi al wins, which will push he use
o HPC o he nex le el [20]. The impo ance o HPC is also eflec ed
in he g owing olume o he ma ke . Analys s om Hype ion Resea ch
*Co esponding au ho .
E-mail add ess: [email p o ec ed] (D. Plei e ).
1h ps://eu ohpc-ju.eu opa.eu/.
p edic he annual g ow h o he HPC ma ke o 8.2 % o e he nex
fi e
yea s [6].
The ole o HPC has also been ecognized a he poli ical le el. Fo
example, in Eu ope, a new o ganisa ion has been c ea ed called Join
Unde aking Eu oHPC,1which es ablishes an exascale HPC in as uc-
u e and suppo s ela ed esea ch. The s a egy is o coo dina e effo s
and pool esou ces in o de o acili a e a globally leading ole o Eu-
opean s akeholde s. Simila effo s a e ongoing in o he egions o he
wo ld, mos no ably in China, India, Japan, and he US.
The success o hese effo s c i ically depends on he a ailabili y o
app op ia ely ained esea che s, enginee s, and o he expe s. Ensu -
ing he a ailabili y o such expe s is challenging due o he b ead h o
equi ed skills coming om a ious science and enginee ing domains.
HPC is a e y in e disciplina y a ea, which needs o be eflec ed in he
h ps://doi.o g/10.1016/j.jpdc.2025.105081
Recei ed 1 June 2024; Recei ed in e ised o m 22 Decembe 2024; Accep ed 1 Ap il 2025
Jou nal o Pa allel and Dis ibu ed Compu ing 201 (2025) 105081
A ailable online 3 Ap il 2025
0743-7315/© 2025 The Au ho s. Published by Else ie Inc. This is an open access a icle unde he CC BY license ( h p://c ea i ecommons.o g/licenses/by/4.0/ ).
P. Bou y, M. B o sson, R. Canal e al.
educa ion. This anges om a ious domains o compu a ional science
and enginee ing, om compu e science o compu e and elec ical en-
ginee ing.
To add ess he challenge o he u u e wo k o ce, Eu oHPC has p o-
ided unding o he EUMas e 4HPC p ojec .2The p ojec conso ium
includes dozens o uni e si ies and educa ional ins i u ions h oughou
Eu ope, HPC cen es, and p i a e companies ha a e ac i e in he a ea
o HPC. A p ima y objec i e o he p ojec is he design o a common Eu-
opean mas e e e ence cu iculum in HPC. I c ea ed an oppo uni y o
o mula e a holis ic cu iculum ha d aws on he knowledge and expe-
ience o a wide ange o s akeholde s, expands exis ing cu icula wi h
unique cha ac e is ics, and p omo es new eaching and aining me h-
ods. Examples a e he aining o ans e sal skills o he implemen a ion
o challenges. The cu iculum is designed in a modula way o acili a e
i s up ake by a la ge numbe o uni e si ies h oughou Eu ope. This
makes i easie o implemen i wi hin exis ing p og ams. Fu he mo e,
he p ojec implemen s a pilo p og am in ol ing abou 150 s uden s,
who benefi om a mobili y p og am o ecei e he mos ad anced ed-
uca ion and aining om mul iple o ganisa ions.
Wi h his pape , we make he ollowing con ibu ions:
1. Documen a ion o a holis ic cu iculum o a mas e ’s in HPC edu-
ca ion, wi h a ocus on i s unique ea u es.
2. Discussion o he me hodology ha led o he o mula ion o his
cu iculum.
3. E alua ion o ea ly expe iences wi h he implemen a ion o he cu -
iculum and he aining o EUMas e 4HPC s uden s.
This pape is o ganised as ollows: In Sec ion 2, we discuss a selec-
ion o ela ed wo k. The nex Sec ion 3, documen s he me hodology
ha esul ed in he cu iculum, which is documen ed in Sec ion 4. In
Sec ion 5, we epo on diffe en aspec s o he ini ial implemen a ion
o he cu iculum. In Sec ions 6and 7, we p o ide a fi s e alua ion o
he use o he cu iculum based on a case s udy and a su ey. We end
wi h a summa y and conclusions in Sec ion 8.
2. Rela ed wo k
The impo ance o quali y educa ion in HPC o basic sciences and
enginee ing has long been ecognised (see, o example, he ea ly epo
by Long e al. [7]). An in e es ing newe de elopmen a e he offe -
ings o s uden s who do no majo in compu e science, bu ins ead a e
in oduced o co e concep s, he eby s essing he new objec i e o com-
pu a ional hinking [21].
EUMas e 4HPC has ex ended he scope u he by also conside ing
HPC sys em adminis a o s as a possible u u e ole, which is nei he
co e ed by educa ion o sys em adminis a ion no he adi ional HPC
educa ion. Such an ex ension has ea lie been p oposed in [24].
One pa icula challenge o educa ion in he a ea o HPC is he as
e olu ion o he compu a ional me hods as well as he a chi ec u es and
echnologies. One no able example is he in oduc ion o he e ogeneous
compu ing pla o ms. Qasem e al. sugges ed add essing he g owing
need o handling such pla o ms by defining modules ha in oduce
undamen al concep s o he e ogeneous (o accele a ed) compu ing in o
unde g adua e educa ion in compu e science and compu a ional engi-
nee ing [13]. Ano he s a egy is he s onge in ol emen o supe com-
pu ing cen es. A p ac ical example is gi en by Chen e al. [3] epo ing
on effo s a China’s Na ional Uni e si y o De ense Technology (NUDT)
ha has hos ed he Tianhe supe compu e s.
O e he yea s, nume ous s udies ha e been published in oduc-
ing new me hods o eaching HPC. One example is he o ganisa ion
o hacka hons as pa o uni e si y educa ion, o which he B azilian
2h ps://eumas e 4hpc.uni.lu/.
Pa allel P og amming Ma a hons [8] is an example. The au ho s con-
side hacka hons as a good way o helping s uden s wi h a pa adigm
shi om adi ionally sequen ial p og amming echniques o pa allel
and dis ibu ed compu ing. I also ela es o ou effo s o in ol e s u-
den s om diffe en uni e si ies in ob aining p ac ical expe iences on
HPC sys ems (see Sec ion 5.4.3), which o he s ha e p oposed as an
educa ional ool (see, e.g., [5]). Ano he example o pape s desc ib-
ing he success ul adop ion o new ways o eaching HPC is he use o
Jupy e no ebooks [9]. Building small-scale, low-cos compu ing clus-
e s based on simple de ices, e.g. edge de ices like Raspbe y Pis, is
no new (see, e.g., [4]) bu s ill has a limi ed up ake. The in eg a ion o
such a me hodology wi hin cou ses is discussed in [19]. Mo e ecen ly,
Xu Zhiguang combined he app oach o building clus e s on-si e using
low-cos de ices wi h he use o high-end compu ing esou ces in he
cloud le e aging Google collabs [23].
Va ious wo ks documen he de elopmen o specific cou ses. One
nice ecen example is Va banescu e al. [18] who de eloped a ull pe -
o mance enginee ing cou ses and documen ed he se up. Such cou ses
fill an impo an gap as he e is a need o me hodology educa ion in
he a ea o pe o mance analysis and enginee ing [1]. EUMas e 4HPC
add esses his by o eseeing a specialisa ion as Pe o mance Analys and
Ad iso (see Sec ion 4.2.2).
A b oade scope ini ia i e in cu iculum de elopmen o Pa allel
and Dis ibu ed Compu ing has been launched wi h he guidelines pub-
lished by he NSF/IEEE-TCPP Cu iculum Wo king G oup [11]. An anal-
ysis o cu en ends in HPC educa ion and co esponding challenges o
u u e di ec ions has been epo ed by he Wo king G oup on Inno a-
ion and Technology in Compu e Science Educa ion [14]. Compe ency-
based lea ning was iden ified as a needed ans o ma ion om he mo e
p e alen knowledge-based lea ning in HPC educa ion [2]. A good and
con inuous sou ce o publica ions and p esen a ions ela ed o each-
ing me hods and new cou ses in he a ea o HPC educa ion a e se e al
wo kshop se ies:
•The EduPa wo kshops ake usually place as pa o he IEEE In e -
na ional Pa allel & Dis ibu ed P ocessing Symposia (IPDPS) [10],
•EduHPC is a se ies o wo kshops in he con ex o he In e na ional
Con e ences o High Pe o mance Compu ing (SC).3
•Edu-HiPC wo kshops ha e been o ganised o a ew yea s on an
annual basis in India.4
In he con ex o Eu oHPC, a ela ed effo is o define a ain-
ing baseline ha is no limi ed o uni e si y educa ion a he mas e ’s
le el. This effo is being ca ied ou by he CASTIEL2 and Eu oCC2
p ojec s.5The aining baseline o HPC, High-Pe o mance Da a Analy -
ics (HPDA), and A ificial In elligence (AI) is defined a ound six diffe en
ca ee pa h p ofiles. One o he main design goals o he aining base-
line was ha each o he building blocks emain modula , allowing o
u he adap a ions in u u e eleases. The a ge sec o s comp ise “Re-
sea ch/Academic + + ”, “SMEs” and “Big Indus y” and co esponding
ca ee pa h p ofiles a e “HPC applica ion de elope ”, “Sys em Admin-
is a o ”, “HPC applica ion use ”, “Da a scien is / Analys ”, “Execu i e
/ Technical lead” and “Sys em designe ”. Each ca ee pa h p ofile is as-
signed a specific lis o aining uni s, p o iding a s epwise acquisi ion
o skills needed in he espec i e a ge p o ession.
A no able diffe ence o he EUMas e 4HPC cu iculum is he signifi-
can ly educed wo kload o a pa icula ca ee pa h p ofile. Ca ee pa h
p ofiles a e conside ed mo e as addi ional aining wi hin he ealm o
li elong lea ning [15] o offe in e es ed candida es a non- adi ional
3Fo he 2024 edi ion, see h ps:// cpp.cs.gsu.edu/cu iculum/?q=
eduhpc24.
4Fo he 2024 edi ion, see h ps:// cpp.cs.gsu.edu/cu iculum/?q=
eduHiPC24.
5h ps://www.eu occ-access.eu/.
Jou nal o Pa allel and Dis ibu ed Compu ing 201 (2025) 105081
2
P. Bou y, M. B o sson, R. Canal e al.
ou e o specializa ion. P elimina y discussions had iden ified mic o-
c eden ials [16] as a iable way o ce ifica ion o he success ul com-
ple ion o a pa icula ca ee pa h p ofile — albei a he lowe end
o app oxima ely 3-6 ECTS. Gi en he explici emphasis on indus y,
he aining baseline could offe in e es ing ex ensions o he EUMas-
e 4HPC cu iculum.
3. Me hodology
The p ocess o de eloping he cu iculum has been a mul is age
effo . I s a ed wi h an analysis o he need o expe s ained in HPC-
ela ed opics in he p i a e and academic sec o s. This esul ed in a se
o skill p ofiles. Ini ial esul s o his effo ha e been published ea lie
[12]. Fo he design o he cu iculum, a balance had o be main ained
be ween fixed cha ac e is ics and a iable elemen s. The fixed cha ac-
e is ics include a du a ion o 4 semes e s. To ob ain a mas e ’s deg ee,
s uden s a e expec ed o ake cou ses ha ansla e in o 30 ECTS pe
semes e and a o al o 120 ECTS o ob ain a mas e ’s deg ee. To keep
he cu iculum flexible, he eaching con en has been defined in e ms
o small modules wi h 1 o 3 ECTS pe module and an effo o app oxi-
ma ely 10 o 30 eaching hou s. Al hough he se o modules, p esen ed
in he nex Sec ion, is based on cu en p ac ices, i has been augmen ed
and adjus ed o ake in o accoun new ends and upcoming needs in o
accoun .
To assess he upcoming p i a e sec o equi emen s, a dialogue has
been conduc ed wi h ele an companies h ough ques ionnai es, s uc-
u ed in e iews, and discussions du ing con e ences and o he mee ings
ela ed o HPC. The in e ac ion was guided by an analysis o job ad e -
isemen s. As hese eflec only cu en needs, he dialogue ocused on
u u e ends. The esul o his effo has been he iden ifica ion o six
diffe en skill p ofiles:
•Compu a ional scien is , HPC applica ion suppo specialis ,
•Applica ion so wa e de elope ,
•HPC sys em so wa e de elope , HPC so wa e enginee ,
•HPC ha dwa e de elope ,
•HPC sys em adminis a o , and
•HPC a chi ec .
A simila app oach has also been aken o he academic sec o . The
esul s ha e been he ollowing ou p ofiles:
•Applica ions
•Pa allel p og amming and ools suppo
•De Ops (sys em suppo and de elopmen )
• Sys em a chi ec
The assessmen o he skill equi emen s in bo h sec o s also con-
fi med he need o offe educa ion and aining in ans e sal and p o es-
sional skills. These skills a e o en no ye add essed in exis ing cu icula
o HPC s udies. We conside ed a wide ange o skills, including hose
ela ed o esea ch echniques, opics ela ed o en ep eneu ship, and
skills o add ess legal opics. Fo de eloping his pa o he cu iculum,
he ollowing s eps ha e been pe o med:
•Analysis o exis ing cu icula and ini ia i es ac oss he o ganisa-
ions in ol ed in he EUMas e 4HPC p ojec .
•Selec ion o ele an ans e sal skills and espec i e lea ning ou -
comes.
•Explo a ion o he educa ional app oaches o achie e he expec ed
lea ning ou comes.
Fo an o e iew o his pa o he cu iculum, see Sec ion 4.3.
Fo each o he modules, lea ning ou comes ha e been defined, ak-
ing Bloom’s axonomy in o accoun . The main s eng h o ou pe spec-
i e is he adop ion o a s uc u al me hodology o offe a comp ehensi e
Fig. 1. P oposed undamen al modules and cu iculum concep in EUMas-
e 4HPC.
app oach o cu iculum design. This me hodology enables uni e si y
p og ams o conside s uden equi emen s, e alua e he easibili y o
eaching ma e ials and pedagogical app oaches, and de elop assess-
men s wi h conc e e lea ning objec i es. Bloom’s axonomy p o ides
a amewo k o unde s anding he lea ne ’s needs, which is use ul a
he g adua e le el, as s uden s may be accep ed o easons o he han
abili y.
A new ype o module, balancing modules, has been in oduced o
offe mo e flexibili y in e ms o s udy pa hs. These modules suppo
he in ake o s uden s wi h bachelo ’s deg ees o he han compu e sci-
ence. An o e iew o he p oposed balancing modules is p o ided in
Sec ion 4.4.
4. The cu iculum
In his sec ion, he cu iculum is in oduced. A isual o e iew is
p o ided in Fig. 1I shows he diffe en domains o he undamen als,
which a e in oduced in Sec ions 4.1 and 4.3. The specialisa ions will
be discussed in Sec ion 4.2. Finally, he balancing modules will be p e-
sen ed in Sec ion 4.4.
4.1. Fundamen als
We ha e iden ified fi e domains o specific a eas o he undamen-
als:
•Ma hema ics and S a is ics
•So wa e Enginee ing
•Pa allel P og amming
•Compu e A chi ec u e
•T ans e sal Skills
The app oach ha we ollowed was o define small modules ha
can be combined o p o ide a numbe o cou ses, see Fig. 2. T ans e -
sal skills add ess issues like di e si y, inclusion, and gende bias as a
c oss-cu ing opic o all modules. Modules a e se o ha e 3 ECTS so
ha by combining hem i is easy o build egula cou ses o 6 ECTS,
o la ge ones, up o 9 ECTS. Each uni e si y impa ing he mas e can
decide o combine he modules as hey p e e , p o ided ha he final
con en is compa able. Some o he modules in each domain a e “ba-
sic”, in he sense ha hey o e lap wi h con en s po en ially done in a
p e ious bachelo ’s deg ee. Those modules a e only assigned o s uden s
wi h diffe en backg ounds. Fo example, s uden s coming om a non-
echnical bachelo can be assigned modules such as “Basic Compu e
A chi ec u e”, “Basic Pa allel P og amming”, o “Ope a ing Sys ems”.
Table 1shows he weigh o each a ea in he cu iculum, as well as
he amoun o knowledge implemen ed as Balancing Modules, and he
amoun con ibu ing o he co e o Fundamen als.
The ollowing subsec ions show he o ganiza ion o he modules o
each domain and hei lea ning ou comes:
Jou nal o Pa allel and Dis ibu ed Compu ing 201 (2025) 105081
3
P. Bou y, M. B o sson, R. Canal e al.
Fig. 2. P oposed undamen als in EUMas e 4HPC (Yea One).
Table 1
P oposed ECTS o undamen als and balancing modules in yea one.
A ea Balancing ECTS Fundamen als ECTS To al
Ma hema ics and S a is ics 6 6 12
So wa e Enginee ing 6 – 6
Pa allel P og amming 9 24 33
Compu e A chi ec u e 3 15 18
T ans e sal Skills – 9 9
To al 24 54 78
4.1.1. Ma hema ics and s a is ics
This domain ge s he s uden s o a common le el, by allowing hose
wi h no o low exposu e o Algeb a o S a is ics o ake he Balancing
Modules o Nume ical Me hods o Linea Sys ems and In oduc ion o
S a is ics and Lea ning.
Once all s uden s ha e been b ough o a common le el, he mas e
p og am includes Nume ical Me hods o Pa ial Diffe en ial Equa ions,
including he applica ion o nume ical analysis, conside ing s abili y,
consis ency, and a e o con e gence, and he me hods o fini e ele-
men s, Gale kin, Fou ie and spec al me hods. On he p ac ical side,
s uden s de elop and compa e he pe o mance o diffe en nume ical
me hods.
La e in he domain, Nume ical Op imiza ion includes he knowledge
o ma hema ical aspec s o op imiza ion, smoo h and non-smoo h op i-
miza ion echniques, s ochas ic g adien and Limi ed-memo y B oyden-
Fle che -Gold a b-Shanno (L-BFGS) algo i hms. On he p ac ical side,
s uden s de elop sample applica ions using hose op imiza ion me hods.
4.1.2. So wa e enginee ing
So wa e enginee ing echniques and ools a e al eady common o
s uden s coming om CS- ela ed backg ounds. Fo his eason, we con-
side hese opics o belong o Balancing Modules. Two modules a e
defined: So wa e De elopmen Techniques and Tools including he de-
sc ip ion and use o he so wa e li e cycle models, s anda d p ocedu es
(like ISO/IEC/IEEE 24765), equi emen s, use-cases, po abili y and
pe o mance po abili y; So wa e Quali y, Valida ion and Tes ing in-
cludes he conside a ion and implemen a ion o so wa e sus ainabili y,
echnical dep h assessmen , so wa e quali y c i e ia, es ing me hods,
and hei p ac ical applica ion.
4.1.3. Pa allel p og amming
The domain o Pa allel P og amming includes 3 modules om he se
o Balancing opics ha a e usually o e lapping wi h CS- ela ed bach-
elo s: Ope a ing Sys ems, Basic Pa allel P og amming, and Algo i hms
and Da a S uc u es.
On op o hose, he mas e cu iculum p o ides 4 addi ional a eas
o knowledge:
Pa allel p og amming ools The a ea includes Compile s and Op imiza-
ions, and Pe o mance Analysis and Models. In his a ea he s uden s ge
o know, he abili ies o explain and use he concep s ela ed o he com-
pila ion p ocess, p og amming languages, code op imiza ions, exploi a-
ion o he memo y hie a chy, main echniques o ob aining pa allelism
(ins uc ion-le el pa allelism, ec o iza ion, and mul i- h eading), pe -
o mance analysis me ics and laws (Amdahl’s and Gus a son’s law),
pe o mance analysis models, and ools. On he mo e p ac ical side, use
p ofiling-based op imiza ions, and de ec o e head and bo lenecks on
pa allel applica ions.
Sha ed-memo y en i onmen s and p og amming The a ea co e s he de-
sc ip ion and use o single-node sha ed-memo y compu e s, applying
pa alleliza ion echniques using well-known, s anda d, and s a e-o - he-
a pa allel p og amming models (cu en ly p oposed a e MPI, OpenMP,
SYCL, e c.). Use o au oma ic pa alleliza ion, and combine pa alleliza-
ion and ec o iza ion. On he p ac ical side, apply pa allel cons uc s,
asking, and ec o iza ion o pa icula benchma ks and code s uc u es.
He e ogeneous en i onmen s and p og amming The a ea includes he de-
sc ip ion and use o single-node sha ed-memo y sys ems wi h accele a-
o s (cu en ly GPUs, TPUs, DPUs and FPGAs), applying offloading ech-
niques and managing da a ans e s wi h s anda d and s a e-o - he-a
he e ogeneous p og amming ools (cu en ly CUDA, OpenCL, OpenACC,
OpenMP a ge , SYCL, e c.), imp o e he use o he accele a o s o
pa icula algo i hms. On he p ac ical side, be able o analyse he pe -
o mance o he e ogeneous pa allel applica ions and de ec d awbacks.
Dis ibu ed-memo y en i onmen s and p og amming The a ea co e s he
desc ip ion and use o clus e , g id and cloud en i onmen s, compila ion
ools, o ganisa ion o applica ions based on wo kflows and pipelines, mi-
c ose ices, and pa alleliza ion o applica ions wi h s anda d, s a e-o -
he-a dis ibu ed p og amming ools (cu en ly MPI, e c.). On he p ac-
ical side, use he echniques o exploi in e ope abili y ac oss diffe en
p og amming models (cu en ly MPI + OpenMP, MPI + CUDA, e c.).
4.1.4. Compu e a chi ec u e
The domain o Compu e A chi ec u e includes one module om he
se o Balancing opics ha usually o e laps wi h CS- ela ed bachelo s:
Basic Compu e A chi ec u e.
Jou nal o Pa allel and Dis ibu ed Compu ing 201 (2025) 105081
4
P. Bou y, M. B o sson, R. Canal e al.
On op o hose, he mas e cu iculum p o ides 4 addi ional a eas
o knowledge:
P ocesso a chi ec u e The a ea co e s basic and ad anced p ocesso a -
chi ec u e. In his a ea, he s uden s ge o know he diffe en ypes o
p ocesso a chi ec u es, he undamen al mechanisms o s a ic ILP p o-
cesso s, so wa e-le el ins uc ion scheduling echniques (lis schedul-
ing, un olling, so wa e pipelining), he memo y hie a chy and i s com-
ponen s, cache design and cache s a egies and p o ec ion and secu i y
issues ela ed o he p ocesso a chi ec u e. On he mo e ad anced side,
s uden s lea n and ge o use he undamen al mechanisms o dynamic
ILP p ocesso s (ou -o -o de execu ion, b anch p edic ion, excep ion
handling, ins uc ion and da a p e e ch, e c.) and conside p o ec ion
and secu i y issues in he p ocesso design. On he p ac ical side, analyse
he pe o mance and assess he impac on powe and ene gy o single-
h eaded p og ams on in-o de and ou -o -o de p ocesso s, and apply
simple op imiza ions o code agmen s.
He e ogeneous compu e a chi ec u e The a ea co e s accele a o s (e.g.,
GPUs, FPGAs), hei a chi ec u e and mic o-a chi ec u e, ha dwa e
echniques o h ead scheduling and synch oniza ion, accele a o s
memo y hie a chy, FPGA building blocks and FPGA in e connec s. On
he p ac ical side, explo e accele a o pe o mance capabili ies o syn-
he ic benchma ks, p og am sample applica ions wi h FPGA p og am-
ming wo kflows and explo e he pe o mance o selec ed compu a ional
mo i s on accele a o s (e.g., s uc u ed g ids, FFT, dense linea algeb a).
Mul ip ocesso a chi ec u e The a ea co e s he implemen a ion o di -
e en models o h ead-le el and da a-le el pa allelism, such as co e
mul i- h eading, many-co e p ocesso s, ec o uni s, mul i-chip mod-
ules, he impac o mul ip ocessing on he p ocesso ISAs, implemen-
a ion o ha dwa e-le el synch oniza ion mechanisms, he ope a ion o
he memo y hie a chy on mul ip ocesso s: snooping and di ec o y-based
cohe ence p o ocols, cohe ency, and consis ency. Design a he a chi ec-
u al le el o ha dwa e o sha ed-memo y, basic componen s o mode n
pa allel sys ems, including mul iple p ocesso s, cache hie a chies and
ne wo ks. And quan i y pe o mance me ics o pa allel sys ems and
e alua e pa allel compu e o ganiza ions. On he p ac ical side, de elop
small benchma ks e alua ing he pe o mance o ha dwa e synch oniza-
ion mechanisms, he o e head o h ead and p ocess managemen and
e alua e he mechanisms implemen ing da a cohe ence and consis ency.
Ne wo king The a ea co e s ne wo k opologies, ou ing algo i hms,
flow con ol, conges ion con ol mechanisms, e o de ec ion/co ec ion
mechanisms, ISO/OSI model, ne wo k p o ocols (TCP/IP, RDMA, e c.)
and ne wo k pe o mance me ics (e.g., la ency, bandwid h, and qual-
i y o se ice). On he p ac ical side, de elop benchma ks implemen ing
communica ions wi h commonly used p o ocols (TCP/IP, RDMA, e c.),
and analyse he pe o mance o a ne wo k conside ing la ency, band-
wid h and quali y o se ice.
4.1.5. T ans e sal skills
The T ans e sal Skills a e dis ibu ed ac oss he diffe en modules o
he cu iculum. In his pape , we collec ed all he in o ma ion, and i is
p esen ed in Sec ion 4.3.
4.2. Specialisa ions
Specialisa ions a e ypically offe ed in he 3 d semes e o he Mas-
e ’s deg ee. Wi hin he EUMas e 4HPC p ojec , we ha e iden ified he
ollowing ou specialisa ions as he mos significan h ough he anal-
ysis o he job ma ke a ound he p i a e and academic sec o s:
•Nume ical and Da a Specialis o Science Domains
•Pe o mance Analys and Ad iso
• Sys em De elopmen and Suppo
Table 2
P oposed ECTS o specialisa ions and modules in yea wo.
Specialisa ion Modules ECTS
Nume ical and Da a Specialis o Science Domains 20 66
Pe o mance Analys and Ad iso 15 51
Sys em De elopmen and Suppo 17 57
Sys em A chi ec 16 54
• Sys em A chi ec
Fo each specialisa ion, se e al eaching modules ha e been iden-
ified, including all ele an and ac ual opics in he o m o lea ning
ou comes ha will allow s uden s o ga he addi ional knowledge and
expe ise wi hin he selec ed specialisa ion. Each module has i s p e-
equisi es, he con en s he s uden s will lea n, and a ious sugges ed
p ac ical opics. The in oduced specialisa ions complemen he unda-
men als p esen ed abo e and oge he a e he basis o building he
gene al cu iculum o he mas e p og am.
The numbe o modules cu en ly iden ified and espec i e ECTS o -
e ed by each specialisa ion a e shown in Table 2. Each specialisa ion
should con ain a minimum o 30 ECTS o modules. P ac ically, hey ha e
mo e ECTS o allow mo e eedom o he implemen ing ins i u ions. The
numbe o ECTS o pass specialisa ion is 30. Each specialisa ion con ains
egula modules o 3 ECTS, a p ojec module o a maximum o 9 ECTS,
and a lis o modules sha ed wi h o he specialisa ions. All modules a e
g ouped in o hema ic a eas (see Fig. 3), which p o ide he s uden s
wi h he lea ning ou comes iden ified o each specialisa ion and u -
he desc ibed in he ollowing subsec ions.
The a ailabili y o cou ses in diffe en implemen a ions a diffe en
ins i u ions may esul in sligh changes in he modules and ECTS.
4.2.1. Nume ical and da a specialis o science domains
This specialisa ion ocuses on g adua e expe s in scien ific compu -
ing, ma hema ical modelling, he de elopmen o nume ical algo i hms
o supe compu e s, pa allel p og amming, and knowledge o some ap-
plica ion domains, such as enginee ing, physics, and biomedical. Fu -
he mo e, AI modelling wi h HPC capabili y and high-pe o mance da a
analy ics ha a e a ailable o a ge ing esea ch and applica ions is
ending in AI / ML. They will be able o in e ac in in e disciplina y
eams wi h expe s and scien is s in he applica ion domain on he de-
elopmen o efficien pa allel implemen a ions, pe o mance analysis,
la ge da a analysis, and isualisa ion. The ollowing will discuss mo e
de ails abou he lea ning ou comes o his specialisa ion.
Fig. 4and Table 3show an o e iew o he a eas a ge ed and he
expec ed ECTS wi hin his specialisa ion. A de ailed desc ip ion o each
a ea ollows.
Scien ific compu ing The opic will ocus on efficien nume ical algo-
i hms and analysis o diffe en ial equa ions, especially o pa ial di -
e en ial equa ions. In pa icula , his a ea con ains modules such as
Compu a ional Nume ical Linea Algeb a, Pa allel Nume ical Me hods,
and Implemen a ion o Pa ial Diffe en ial Equa ion (PDE) Sol e s. They
include algo i hms o linea algeb a: o hogonalisa ion, leas -squa es
me hods, sol ing la ge linea sys ems, and s udying hei s abili y and
communica ion equi emen s. Fo example, how good is he s abili y o
he algo i hm wi h espec o he le el o pa allelism? Fu he mo e, di -
e en domain decomposi ion me hods will be s udied in he con ex o
pa allel nume ical me hods, such as mul ig id and Schwa z me hods. Fi-
nally, we in oduce diffe en well-known nume ical me hods o sol ing
PDE ha will be s udied in de ail, such as fini e diffe ence, fini e olume
and fini e elemen me hods. The s udy g oup is expec ed o implemen
any o hose me hods om sc a ch, and also o ha e hands-on expe-
ience wi h handling (o using) lib a ies used wi hin hose nume ical
me hods. Impo an ly, all o hose me hods and echniques will always
be discussed in he con ex o pa allel a chi ec u e machines, including
GPU-accele a ed sys ems.
Jou nal o Pa allel and Dis ibu ed Compu ing 201 (2025) 105081
5
P. Bou y, M. B o sson, R. Canal e al.
Fig. 3. P oposed specialisa ions in EUMas e 4HPC (Yea Two).
Fig. 4. P oposed modules (s udy a eas) wi hin Nume ical and Da a Specialis s
o Science Domains.
Ad anced HPC opics o science domains I in oduces ad anced opics
such as quan um compu ing and domain-specific p og amming mod-
els. Rega ding quan um compu ing, i co e s quan um compu e a chi-
ec u e ( o example, qubi s) and p og amming models ( o example,
Qiski , Ocean, Ci q, e c.) and applying he quan um compu ing ech-
nique (using he eal machine o simula o ) o sol e applica ions om
science and enginee ing. Fu he mo e, wi hin domain-specific p og am-
ming models, especially ocusing on embedded sys ems and hei p o-
g amming models. Fo example, p og ammabili y and unning nume i-
cal o AI algo i hms such as in FPGAs and IPU a chi ec u es. Finally, i
in oduces how o un a la ge science and enginee ing simula ion ( o
example, medical enginee ing, fluid dynamics, solid mechanics and dy-
namics, li e sciences, ma e ial sciences, and chemis y) using la ge HPC
machines, in pa icula , by s udying so wa e packages o simula ion,
pa allelisa ion s a egies, me hods, ma hema ical p ope ies, and imple-
men a ion app oaches on HPC sys ems.
AI/ML I co e s he essen ial opics o expe ise in machine lea ning,
a ificial in elligence, and compu a ional s a is ics. Machine lea ning
will ocus on opics such as model e alua ion and selec ion, anomaly
de ec ion, con o mal lea ning (p edic ion wi h gua an ees o accu acy),
causal in e ence (iden ifica ion o causal ela ionships), and hyb id ap-
p oaches such as in eg a ing nume ical me hods wi h a ificial neu al
Table 3
P oposed modules (s udy a eas)
wi hin Nume ical and Da a Spe-
cialis o Science Domains.
A ea Modules ECTS
Scien ific
Compu ing
3 9
Ad anced
HPC Topics
3 9
AI/ML 3 9
HPDA 3 9
P ojec 1 9
Sha ed
Modules
7 21
In To al 20 66
ne wo ks o science and enginee ing p oblems. Thei HPC implemen-
a ion uses exis ing so wa e amewo ks, o example, Tenso Flow and
PyTo ch, in pa allel a chi ec u e, using CPU and GPUs.
High-pe o mance da a analy ics I a ge s big da a, high-dimensional
da a analysis, and da a isualisa ion. Big da a analysis includes file sys-
ems and da abases o big da a and machine lea ning o big da a
connec ed wi h i s analysis using ools and p og amming models, such
as Apache Spa k and he Hadoop Map-Reduce ecosys em. Wi h espec
o high-dimensional da a analysis, mo e explana ion will be gi en o he
geome y o high-dimensional da a se s, nonlinea dimension educ ion
and mani old models, linea dimension educ ion, p incipal componen
analysis and ke nel me hods, ma hema ical dis ance and dimension e-
duc ion, singula alue decomposi ion, and p incipal componen anal-
ysis. Finally, we in oduce da a handling and isualisa ion ools like
Pa aView and VisI .
P ojec I encou ages he s udy g oup o ocus ei he on ma hema ical
modelling (nume ical me hods, op imisa ion, and s a is ical modelling)
o pa allel implemen a ion and analysis o ma hema ical algo i hms on
HPC machines, especially conside ing eal-wo ld applica ions such as
ae odynamics, machine lea ning, ma e ial science, e c.
Sha ed modules This will help he s udy g oup in eg a e hei knowl-
edge wi h o he beneficial specialisa ion opics, conside ing he a ea
o scien ific compu ing and, o example, co e ing ad anced opics in
la ge-scale compu e a chi ec u e ( om “Sys em A chi ec ”), ope a ing
sys ems, s o age, Vi ualisa ion & Con aine isa ion ( om “Sys em De el-
opmen and Suppo ”) and pe o mance analysis ools o he e ogeneous
HPC machines ( om “Pe o mance Analys and Ad iso ”). Essen ially,
Jou nal o Pa allel and Dis ibu ed Compu ing 201 (2025) 105081
6
P. Bou y, M. B o sson, R. Canal e al.
Fig. 5. P oposed modules (s udy a eas) wi hin Pe o mance Analys and Ad iso .
Table 4
P oposed modules (s udy a eas)
wi hin Pe o mance Analys and
Ad iso .
A ea Modules ECTS
Pe o mance
Analysis
2 6
Pe o mance
Modelling
2 6
Pe o mance
Op imisa ion
2 6
P ojec 1 9
Sha ed
Modules
8 24
In To al 15 51
i will gi e a u he be e unde s anding o he de ailed s udy o com-
pu a ional science.
4.2.2. Pe o mance analys and ad iso
This specialisa ion a ge s g adua e expe s in pe o mance analy-
sis and uning o complex applica ion codes o mode n compu e a -
chi ec u es and supe compu e s. I ollows a s uc u ed pe o mance
enginee ing app oach consis ing o pe o mance analysis based on mea-
su emen s, pe o mance modelling o es ima e pe o mance limi s and
bo lenecks on a ce ain pla o m and pe o mance op imisa ion us-
ing hand-c a ed o au oma ically gene a ed a chi ec u e-specific imple-
men a ions.
Fig. 5and Table 4show an o e iew o he a eas a ge ed and he
expec ed ECTS wi hin his specialisa ion. A de ailed desc ip ion o each
a ea ollows.
Pe o mance analysis The con en includes c i e ia o selec ing he
igh ool o specific pe o mance analysis asks and in oduces a i-
ous pe o mance ools such as p ofile s, ace s, debugging ools, and
moni o ing ools. S uden s will lea n o p ofile pa allel applica ions
o iden i y pe o mance bo lenecks, pe o m eal- ime debugging and
acing o mul i h eaded and dis ibu ed applica ions, and moni o sys-
em pe o mance and esou ce u iliza ion. The modules p o ide an in-
dep h explo a ion o p ofiling ools like N idia Insigh , In el VTune,
and likwid, and debugging ools such as To alView and DDT. Pe o -
mance moni o ing and acing ools including In el® T ace Analyze ,
Scalasca, Sco e-P, Vampi , TAU, and Pa a e , as well as moni o ing
ools like Ganglia, Nagios, and P ome heus a e also co e ed. S uden s
will c ea e cus om moni o ing sc ip s and dashboa ds o HPC sys ems
and de elop op imisa ion s a egies based on insigh s om pe o mance
ools, ocusing on code es uc u ing, load balancing, and da a manage-
men . Pe o mance-d i en decision-making o ha dwa e upg ades o
configu a ions is also add essed. P ac ical exe cises in ol e hands-on
expe ience wi h p ofiling ools o analyse compu e and memo y pe o -
mance, using pe o mance analysis ools o de e mine he pe o mance
o pa allel applica ions, and de ec ing o e head and bo lenecks in sam-
ple pa allel applica ions.
Pe o mance modelling I co e s he p inciples o pe o mance models,
compa ing echniques o alida ing models agains exis ing implemen-
a ions, and calib a ing models o ensu e accu acy and eliabili y. I also
includes pe o mance modelling o so wa e applica ions such as web
applica ions, da abases, and pa allel p og ams, and in ol es analysing
wo kload pa e ns, esponse ime, and h oughpu o iden i y pe o -
mance bo lenecks using models. S uden s will apply pe o mance mod-
elling a he node and sys em le els, including o dis ibu ed sys ems,
cloud se ices, and HPC clus e s. I also co e s analysing sys em-le el
pe o mance me ics and in e ac ions, as well as scaling and load balanc-
ing in dis ibu ed sys ems. Pe o mance models will be used o iden i y
op imisa ion s a egies, and s uden s will apply model-d i en op imi-
sa ion echniques o imp o e sys em pe o mance and pe o m cos -
benefi analysis o op imisa ion decisions. P ac ical exe cises include
c ea ing basic pe o mance models, w i ing case s udies o model ali-
da ion, and p edic ing bo lenecks in pa allel applica ion samples.
Pe o mance op imisa ion Fo op imisa ion o a p o ided implemen a-
ion on a specific pla o m, one fi s needs o be able o compa e he
a chi ec u es o mode n supe compu e s and unde s and single-co e
a chi ec u e and op imisa ion s a egies. Memo y hie a chy and da a
access op imisa ions, efficien sha ed memo y pa alleliza ion, and an
e alua ion o diffe en pa alleliza ion app oaches o mul i-co e p oces-
so s, including GPUs, a e necessa y. I also co e s efficien dis ibu ed
memo y pa alleliza ion, ad anced pe o mance models, and he com-
pa ison o se ial and pa allel pe o mance modelling. Addi ionally, s u-
den s will e alua e ene gy-efficien implemen a ion and execu ion o
pa allel p og ams. P ac ical exe cises in ol e implemen ing basic nu-
me ical me hods wi h high ha dwa e efficiency on pa allel compu e s
and gaining insigh in o inno a i e p og amming echniques and al e -
na i e supe compu e a chi ec u es. This also includes ad anced ech-
niques like domain-specific languages and code gene a ion.
P ojec The p e equisi es o his (g oup) p ojec a e basic pe o mance
analysis and knowledge o pe o mance models and compu e a chi ec-
u e undamen als. I usually in ol es p ofiling and op imizing codes
on HPC sys ems o indus y- ela ed p ojec s in he a ea o pe o mance
enginee ing.
Sha ed modules Pe o mance enginee ing also equi es insigh in o
compu e a chi ec u e and compu a ional me hods. Thus, his speciali-
sa ion sha es se e al op ional subjec s wi h he specialisa ions “Sys em
A chi ec ” (He e ogeneous Sys ems and Accele a o s, HPC Sys em De-
sign, Design Tools and Simula o s, High-Le el Digi al Design), “Sys em
De elopmen and Suppo ” (Ad anced Ope a ing Sys ems, Compile De-
sign, Pa allel P og amming Models), and “Nume ical and Da a Specialis
o Science Domains” (Domain Specific P og amming).
4.2.3. Sys em de elopmen and suppo
This specialisa ion a ge s g adua e expe s in he de elopmen o
so wa e o supe compu e s, including he p og amming model, com-
pile , ope a ing sys em, pe o mance analysis ools, and middlewa e
echnologies. Mo e specifically, we selec ed he ollowing skills: he de-
sign and implemen a ion o p og amming models, compile s, compile
op imisa ions, oolchain, so wa e s acks ( o CPU, GPU, FPGA, CRGA),
pe o mance analysis ools, ope a ing sys ems, ke nel de elopmen , pa -
allel file sys ems, high-speed ne wo king, synch oniza ion, con aine
echnologies, i ualisa ion echnologies, in eg a ion o HPC and cloud,
Jou nal o Pa allel and Dis ibu ed Compu ing 201 (2025) 105081
7
P. Bou y, M. B o sson, R. Canal e al.
Fig. 6. P oposed modules (s udy a eas) wi hin Sys em De elopmen and Sup-
po .
Table 5
P oposed modules (s udy a eas)
wi hin Sys em De elopmen and
Suppo .
A ea Modules ECTS
Ope a ing
Sys ems and
Vi ualisa ion
2 6
S o age and
Sys em
Adminis a ion
2 6
Compile s and
Pa allel
P og amming
Models
2 6
P ojec 1 9
Sha ed
Modules
10 30
In To al 17 57
se e adminis a ion, in as uc u e se up and managemen , and secu-
i y.
Fig. 6and Table 5show an o e iew o he a eas a ge ed and he
expec ed ECTS wi hin his specialisa ion. A de ailed desc ip ion o each
a ea ollows.
Ope a ing sys ems and i ualisa ion This a ea includes Ad anced Op-
e a ing Sys ems and Vi ualisa ion and Con aine iza ion Suppo . They
p o ide he s uden s wi h he equi ed design and implemen a ion al e -
na i es a ound scheduling policies, memo y managemen , pa allel I/O,
de ice d i e s, and suppo o i ualised en i onmen s, in o de o
hem o design new app oaches o suppo ing hese and new ea u es
on upcoming sys ems. The p ac ical side includes he de elopmen o
OS d i e s, and e alua ing he suppo o i ualised en i onmen s.
S o age and sys em adminis a ion This a ea co e s he aspec s o Sys em
Adminis a ion on HPC En i onmen s, including p o ec ion and secu-
i y aspec s, applica ion schedule s and esou ce manage s (cu en ly
Slu m...), ins alla ion o applica ions, de elopmen ools, de elopmen
o plans o sys em main enance, use managemen and physical in-
eg a ion o he componen s o a supe compu e . Also, he S o age,
Pa allel File Sys ems and Da abases include he design, implemen a-
ion and managemen o pa allel file sys ems (cu en ly GPFS, Lus e,
Glus e , e c.), he use o ad anced s o age de ices, and pa allel high-
pe o mance da abases. P ac ical includes ins alla ion o he OS i sel ,
and configu a ion o se ices aking in o accoun he p o ec ion and se-
cu i y equi emen s, as well as he design and implemen a ion o pa s
o a pa allel filesys em.
Compile s and pa allel p og amming models This a ea co e s he design
and implemen a ion o Pa allel P og amming Models and Compile s
o HPC en i onmen s. I includes he design o pa allel languages and
models o p og amming HPC sys ems, design o high-le el abs ac-
ions o exp essing pa allelism (cu en ly MPI, OpenMP, Kokkos, Raja,
e c.), ask-based pa allel p og amming, in e ope abili y be ween mod-
els, compile design o HPC and he e ogeneous sys ems, and implemen-
a ion o compile op imisa ions. P ac ical co e s he implemen a ion o
compile passes o op imisa ions, code gene a ion, and language and
p og amming model ex ensions o imp o e exp essi eness o pe o -
mance po abili y.
P ojec In his module, s uden s wo k in g oups wi h he goal o p o-
iding a solu ion o a pa icula p oblem ela ed o he specialisa ion
a eas.
Sha ed modules Addi ionally, his specialisa ion sha es se e al op ional
subjec s wi h he specialisa ions “Sys em A chi ec ” and “Pe o mance
Analys and Ad iso ”. F om “Sys em A chi ec ” he modules on “De-
sign Tools and Simula o s”, “P ocesso Design”, and “Mul ip ocesso
Design”. F om “Pe o mance Analys and Ad iso ”, he modules on
“Tool-based Pe o mance Enginee ing” and “Model-based Pe o mance
Enginee ing”.
4.2.4. Sys em a chi ec
This specialisa ion a ge s g adua e expe s in he design and de el-
opmen o supe compu e s. Focus on p ocesso , mul ip ocesso , supe -
compu e a chi ec u es, memo y and I/O sys ems, ne wo king, ci cui
design, e ifica ion and es , low-powe echniques, and ab ica ion. The
nex subsec ions desc ibe he lea ning ou comes o each iden ified s udy
a ea.
Fig. 7and Table 6show an o e iew o he a eas a ge ed and he
expec ed ECTS wi hin his specialisa ion. A de ailed desc ip ion o each
a ea ollows.
Sys em and ha dwa e echnologies This s udy a ea includes con en s
ela ed o HPC sys em design, ha dwa e echnologies and sus ainable
nanoelec onic design. In de ail, HPC sys em design g oups he con-
en a ound HPC acili ies (e.g. powe deli e y, cooling solu ions, use o
esou ces), and HPC sys em componen s (e.g. CPUs, accele a o s, mem-
o ies, ne wo king, acks, blades). In he a ea o ha dwa e echnologies,
we included he p esen a ion o he unde lying silicon and pho onics
echnologies (MOS, CMOS, memo ies, pho onics), Moo e’s law and Den-
na d scaling, powe wall and da k silicon, sys em eliabili y (so e o s,
ha d e o s, solu ions) and Secu i y (Roo o us , enc yp ion p imi-
i es). Finally, in he Sus ainable nanoelec onic design, we include he
p inciples o sus ainabili y in in eg a ed ci cui s, sou ces o powe dissi-
pa ion, low-powe echniques, li e ime analysis and op imisa ion, CO2
oo p in ; and he socie al and economic impac .
CPUs and he e ogeneous sys ems This s udy a ea includes con en e-
la ed o p ocesso design, mul ip ocesso design, high-le el digi al de-
sign, and he e ogeneous sys ems and accele a o s. Specifically, p oces-
so design includes CPU pipelining, memo y hie a chy, caches, b anch
p edic ion, ese a ion s a ions, eo de buffe , load-s o e queue and
ha dwa e pe o mance coun e s. Mul ip ocesso Design ocuses on he
logic needed o da a cohe ency and consis ency, he on-chip ne wo k
and he memo y hie a chy. Then, high-le el digi al design is de o ed
o he p inciples o digi al sys ems based on p og ammable and config-
u able componen s, he sys em b eakdown in o ha dwa e and so wa e
componen s and he design o p ocessing subsys ems (e.g., ideo, ec o
and ma ix ope a ions, so ing algo i hms). Finally, he e ogeneous sys-
ems and accele a o s include single chip, mul iple chip (chiple ) and
Jou nal o Pa allel and Dis ibu ed Compu ing 201 (2025) 105081
8
P. Bou y, M. B o sson, R. Canal e al.
Fig. 7. P oposed modules (s udy a eas) wi hin Sys em A chi ec .
Table 6
P oposed modules (s udy a eas)
wi hin Sys em A chi ec .
A ea Modules ECTS
Sys em and
Ha dwa e
Technologies
3 9
CPUs and
He e ogeneous
Sys ems
4 12
Chip Design
and Tes
3 9
P ojec 1 9
Sha ed
Modules
5 15
In To al 16 54
mul iple boa d a chi ec u es and accele a o s (e.g. Vec o uni , GPU,
FPGA, TPUs, and s encils).
Chip design and es This s udy a ea g oups he con en ela ed o Phys-
ical Design and Tes and he ela ed ools. In de ail, Sys em-on-Chip
Physical Design includes he iming and powe cons ain s o a complex
in eg a ed ci cui , he physical implemen a ion o a complex in eg a ed
ci cui , low powe design echniques, design, analysis and e alua ion
o elec onic sys ems in applica ions such as au oma ion, ene gy dis i-
bu ion and gene a ion, consume elec onics, biomedicine, e c. In he
module dedica ed o he es ing o digi al designs, we include aul
models, aul co e age, obse abili y, au oma ic es pa e n gene a o s,
scan-based es s and sel - es s. Finally, in he a ea o Design Tools and
Simula o s, s uden s will be p esen ed and use HDLs (e.g., Sys emC, Sys-
emVe ilog) o desc ibe sys ems, apply he VLSI design flow o ha dwa e
design and use physical design ools (pa i ioning, chip planning, place-
men , global ou ing, de ailed ou ing).
P ojec In his module, we expec s uden s o selec a ce ain opic e-
la ed o ongoing esea ch p ojec s, unning compu e- ime p ojec s on
HPC sys ems, indus y- ela ed p ojec s in he a ea o sys em a chi ec-
u e, o pe o ming hands-on sys em de elopmen , modelling o op i-
misa ion.
Sha ed modules Addi ionally, his specialisa ion sha es se e al op ional
subjec s wi h he specialisa ions “Sys em De elopmen and Suppo ”
and “Pe o mance Analys and Ad iso ”. F om “Sys em De elopmen
and Suppo ”, he modules on Ad anced Ope a ing Sys ems, S o age,
Pa allel File Sys ems and Da abases and Compile Design a e highly
in e es ing o his specialisa ion oge he wi h he Tool-Based Pe o -
mance Analysis in he “Pe o mance Analys and Ad iso ” specialisa-
ion.
4.3. T ans e sal skills
The EUMas e 4HPC cu iculum is designed o include in addi ion o
undamen al HPC skills also ans e sal skills. These ans e sal skills
a e essen ial o s uden s o succeed in hei ca ee s and o p epa e
hem effec i ely o he e ol ing job ma ke . We adop he ollowing
defini ion o ans e sal skills by he UNESCO In e na ional Bu eau o
Educa ion: “Skills ha a e ypically conside ed as no specifically ela ed
o a pa icula job, ask, academic discipline o a ea o knowledge and
ha can be used in a wide a ie y o si ua ions and wo k se ings ( o
example, o ganisa ional skills)” [17]. A subse o hese skills in ol es
so skills (some imes also called p o essional skills): “Te m used o in-
dica e a se o in angible pe sonal quali ies, ai s, a ibu es, habi s and
a i udes ha can be used in many diffe en ypes o jobs”.
Nowadays, many op echnology companies es applican s on hei
so skills du ing he in e iew p ocess. Taking in o accoun his a i-
ude, a se o ad anced skills is p oposed wi hin he cu iculum o ain
s uden s o add ess issues such as di e si y, inclusion, gende , and un-
conscious bias.
Based on he me hod desc ibed in Sec ion 3, six g oups o skills ha e
been iden ified, which a e lis ed below. Fo each o hese g oups, mod-
ules ha e been defined in e ms o con en opics:
•Resea ch Techniques: scien ific me hods, c i ical hinking, cu ios-
i y, a ificial in elligence in highe educa ion.
•Scien ific W i ing: w i ing pape s, publishing scien ific esul s, sci-
en ific publica ions da abases, echnical p esen a ions, and p o-
posal w i ing.
•Wo k En i onmen : e hics and in eg i y, eam managemen , hu-
man esou ces, ela ional skills (communica ions), p io i y manage-
men , gende , di e si y.
•En ep eneu ship: business o ganiza ion and managemen , business
plan de elopmen , ma ke iabili y, financial managemen , (pub-
lic) p ocu emen p ocedu es, in ellec ual p ope y, and iden i ying
cus ome needs.
•Legal Aspec s: licensing, open da a, in ellec ual p ope y, da a p o-
ec ion and p i acy (GDPR), EU egula ions in in o ma ion echnol-
ogy.
•O he : a en ion o de ail and aes he ics, cu iosi y, willingness o
aise ques ions and o ely on he expe ise o o he people, am-
bi ion o acqui e mo e skills and willingness o con inuously lea n
new hings.
4.4. Balancing modules
Mos o he ime, high-pe o mance compu ing is misin e p e ed as
ela ed o pu e compu e science. Howe e , his is no ue; om he
ea ly s ages o he p esen day, compu e s ha e been used o doing
a i hme ic compu a ions using la ge olumes o da a se s. Fo example,
sol ing a la ge numbe o sys ems o equa ions, AI compu a ions using
la ge olumes o da a se s, ma e ial science and biomedical simula ion-
based specific ma hema ical modelling and compu e g aphics.
All o hose examples equi e b oad knowledge o physics, ma he-
ma ics, enginee ing, and compu e science skills. The e o e, o become a
highly qualified pe son in high-pe o mance compu ing, one mus ha e
mo e han in e media e knowledge in physics, ma hema ics, and engi-
nee ing.
The e o e, i is essen ial ha we encou age s uden s wi h back-
g ounds in physics, enginee ing, and ma hema ics o s udy mas e ’s
in HPC. Howe e , hose who ha e bachelo ’s deg ees in enginee ing,
physics, and ma hema ics may lack knowledge o pu e compu e sci-
ence.
Jou nal o Pa allel and Dis ibu ed Compu ing 201 (2025) 105081
9
