Performance of Various Computers Using Standard Linear Equations

Performance of Various Computers Using Standard

Linear Equations Software

Jack J Dongarra

Computer Science Department

UniversityofTennessee

Knoxville TN

and

Mathematical Sciences Section

Oak Ridge National Lab oratory

Oak Ridge TN

Novemb er

Electronic mail address dongarracsutkedu

An uptodate version of this rep ort can b e found at httpwwwnetliborgb enchmarkp erformanceps

This work was supp orted in part by the Applied Mathematical Sciences subprogram of the Oce of

Energy Research US Department of Energy under Contract DEACOR and in part

by the Science Alliance a state supp orted program at the UniversityofTennessee

Performance of Various Computers Using Standard Linear

Equations Software

Jack J Dongarra

Computer Science Department

UniversityofTennessee

Knoxville TN

and

Mathematical Sciences Section

Oak Ridge National Lab oratory

Oak Ridge TN

Novemb er

Abstract

This rep ort compares the p erformance of dierent computer systems in solving dense systems

of linear equations The comparison involves approximately a hundred computers ranging from

aCray YMP to scientic workstations such as the Ap ollo and Sun to IBM PCs

Intro duction and Ob jectives

The timing information presented here should in no way b e used to judge the overall p erformance of

a computer system The results reect only one problem area solving dense systems of equations

This rep ort provides p erformance information on a wide assortment of computers ranging from

the homeused PC up to the most powerful sup ercomputers The information has been collected

over a period of time and will undergo change as new machines are added and as hardware and

software systems improve The programs used to generate this data can easily be obtained over

the Internet While wemakeevery attempt to verify the results obtained from users and vendors

errors are b ound to exist and should b e brought to our attention We encourage users to obtain the

programs and run the routines on their machines rep orting any discrepancies with the numb ers

listed here

The rst table rep orts three numbers for each machine listed in some cases the numb ers are

missing b ecause of lack of data All p erformance numbers reect arithmetic p erformed in full

precision usually bit unless noted On some machines full precision may b e single precision

such as the Cray or double precision suchasthe IBM The rst numb er is for the LINPACK

b enchmark program for a matrix of order in a Fortran environment The second numb er is for

solving a system of equations of order with no restriction on the metho d or its implementation

The third numb er is the theoretical p eak p erformance of the machine

LINPACK programs can be characterized as having a high p ercentage of oatingp oint arith

metic op erations The routines involved in this timing study SGEFA and SGESL use column

oriented algorithms That is the programs usually reference array elements sequentially down a

This work was supp orted in part by the Applied Mathematical Sciences subprogram of the Oce of Energy

Research US Department of Energy under Contract DEACOR and in part by the Science Alliance a

state supp orted program at the UniversityofTennessee

An uptodate version of this rep ort can b e found at httpwwwnetliborgb enchmarkp erformanceps

November

column not across a row Column orientation is imp ortant in increasing eciency b ecause of the

wayFortran stores arrays Most oatingp oint op erations in LINPACK take place in a set of subpro

grams the Basic Linear Algebra Subprograms BLAS which are called rep eatedly throughout

the calculation These BLAS referred to nowasLevel BLAS reference onedimensional arrays

rather than twodimensional arrays

In the rst case the problem size is relatively small order and no changes were made

to the LINPACK software Moreover no attempt was made to use sp ecial hardware features

or to exploit vector capabilities or multiple pro cessors The compilers on some machines may of

course generate optimized co de that itself accesses sp ecial features Thus many highp erformance

machines maynothavereached their asymptotic execution rates

In the second case the problem size is larger matrix of order and mo difying or replac

ing the algorithm and software was p ermitted to achieve as high an execution rate as p ossible

Thus the hardware had more opp ortunity for reaching nearasymptotic rates An imp ortantcon

straint however was that all optimized programs maintain the same relative accuracy as standard

techniques such as Gaussian elimination used in LINPACK

Furthermore the driver program supplied with the LINPACK benchmark had to be run to

ensure that the same problem is solved The driver program sets up the matrix calls the routines

to solve the problem veries that the answers are correct and computes the total number of

op erations to solve the problem indep endent of the metho d as n n wheren

The last column is based not on an actual program run but on a pap er computation to determine

the theoretical p eak Mops rate for the machine This is the number manufacturers often cite it

represents an upp er b ound on p erformance That is the manufacturer guarantees that programs

will not exceed this ratesort of a sp eed of light for a given computer

The theoretical p eak p erformance is determined by counting the number of oatingp oint ad

ditions and multiplications in full precision that can be completed during a p erio d of time

usually the cycle time of the machine As an example the Cray YMP has a cycle time of

ns During a cycle the results of both an addition and a multiplication can be completed

op erations cycle

Mops on a single pro cessor On the Cray YMP there are

cycle

pro cessors thus the p eak p erformance is Mops

The information in this rep ort is presented to users to provide a range of p erformance for the

various computers and to show the eects of typical Fortran programming and the results that can

b e obtained through careful programming The maximum rate of execution is given for comparison

The column lab eled Computer gives the name of the computer hardware on which the program

was run In some cases we have indicated the number of pro cessors in the conguration and in

some cases the cycle time of the pro cessor in nanoseconds

The column lab eled LINPACK Benchmark gives the op erating system and compiler used The

run was based on two routines from LINPACK SGEFA and SGESL were used for single precision

and DGEFA and DGESL were used for double precision These routines p erform standard LU

decomp osition with partial pivoting and backsubstitution The timing was done on a matrix of

order where no changes are allowed to the Fortran programs

The column lab eled TPP Toward Peak Performance gives the results of hand optimization

the problem size was of order

The nal column lab eled Theoretical Peak gives the maximum rate of execution based on the

cycle time of the hardware

The same matrix was used to solve the system of equations The results were checked for

accuracy by calculating a residual for the problem jjAx bjjjjAjjjjxjj

The term Mops used as a rate of execution stands for millions of oatingp oint op erations

completed p er second For solving a system of n equations n n op erations are p erformed

November

wecount b oth additions and multiplications

The information in the tables was compiled over a p erio d of time Subsequent systems software

and hardware changes may alter the timings to some extent

One further note The following tables should not b e taken to o seriously In multiprogramming

environments it is often dicult to reliably measure the execution time of a single program We

trust that anyone actually evaluating machines and op erating systems will gather more reliable and

more representativedata

A Lo ok at Parallel Pro cessing

While collecting the data presented in Table wewere able to exp eriment with parallel pro cessing

onanumb er of computer systems For these exp eriments we used either the standard LINPACK

algorithm or an algorithm based on matrixmatrix techniques In the case of the LINPACK

algorithm the lo op around the SAXPY can b e p erformed in parallel In the matrixmatrix imple

mentation the matrix pro duct can be split into submatrices and p erformed in parallel In either

case the parallelism follows a simple forkandjoin mo del where each pro cessor gets some number

of op erations to p erform

For a problem of size we exp ect a high degree of parallelism Thus it is not surprising

that we get such high eciency see Table The actual p ercentage of parallelism of course

dep ends on the algorithm and on the sp eed of the unipro cessor on the parallel part relative to the

sp eed of the unipro cessor on the nonparallel part

Highly Parallel Computing

With the arrival of massively parallel computers there is a need to benchmark such machines on

problems that make sense The problem size and rule for the runs reected in the Tables and

do not p ermit massively parallel computers to demonstrate their p otential p erformance The basic

aw is the problem size is to o small Toprovide a forum for comparing suchmachines the following

b enchmark was run on a numb er of massively parallel machines The b enchmark involves solving a

system of linear equations as was done in Tables and However in this case the problem size

is allowed to increase and the p erformance numb ers reect the largest problem run on the machine

The ground rules are as follows Solve systems of linear equations bysome metho d allow the

size of the problem to vary and measure the execution time for each size problem In computing

the oatingp oint execution rate use n n op erations indep endent of the actual metho d

used If you cho ose to do Gaussian Elimination partial pivoting must be used Compute and

rep ort a residual for the accuracy of solution as jjAx bjjjjAjjjjxjj

The columns in Table are dened as follows

R the p erformance in Gops for the largest problem run on a machine

max

N the size of the largest problem run on a machine

max

N the size where half the R execution rate is achieved

max

R the theoretical p eak p erformance in Gops for the machine

peak

In addition the numb er of pro cessors and the cycle time is listed

Obtaining the Software and Running the Benchmarks

The software used to generate the data for this rep ort can b e obtained by sending electronic mail to

November

netlibornlgov

LINPACK Benchmark

The rst results listed in Tableinvolved no hand optimization of the LINPACK b enchmark

To receive the singleprecision software for this b enchmark in the mail message to netlibornlgov

typ e

send linpacks from benchmark

To receive the doubleprecision software for the LINPACK Benchmark typ e

send linpackd from benchmark

To run the timing programs one must supply a real function SECOND which returns the time

in seconds from some xed starting time

There is only one ground rule for running this b enchmark

No changes are to b e made to the Fortran source co de not even changes in the comments

The compiler and op erating system must b e generally available Results from a b eta version of

a compiler are allowed however the standard compiler results must also b e listed

Toward Peak Performance

The second set of results listed in Table reected user optimization of the software

To receive the singleprecision software for the column lab eled Toward Peak Performance in

the mail message netlibornlgov typ e

send s from benchmark

To receive the doubleprecision software typ e

send d from benchmark

The ground rules for running this b enchmark are as follows

Replacements or mo dications are allowed in the routine LU

The user is allowed to supply any metho d for the solution of the system of equations

The Mops rate will b e computed based on the op eration count for LU decomp osition

In all cases the main driver routine with its test matrix generator and residual check must

b e used

This rep ort is up dated from time to time Afaxcopy of this rep ort can b e supplied for details

contact the author To obtainaPostscript copy of the rep ort send mail to netlibornlgov and in

the message typ e

send performance from benchmark

Tohave results veried please send the output of the runs to

Jack Dongarra

Computer Science Department

UniversityofTennessee

Knoxville TN

Email dongarracsutkedu

There is a Frequently Asked Questions le for the Linpack b enchmark and Top at

httpwwwnetliborgutkp eopleJackDongarrafaqlinpackhtml

November

Table Performance in Solving a System of Linear Equations

Computer LINPACK Benchmark TPP Theoretical

n Best Eort Peak Mops

OSCompiler Mops n Mops

Intel Pentium GHz ifc O xW ip o ip align

ifc O xW ip o ip align Intel Pentium GHz

NEC SX pro c ns

NEC SX pro c ns R pi Wfprob use

Fujitsu VPP pro cns frt Wvr Of KA

IBM eServer pSeries

pro c MHz

IBM eServer pSeries O qarchpwr qtunepwr

pro c MHz Pv Wpeag

Cray T pro c ns

Cray T pro c ns f Oinline

IBM eServer pSeries M

pro c MHz

IBM eServer pSeries M

pro c MHz

IBM eServer pSeries M

pro c MHz

IBM eServer pSeries M O qarchpwr qtunepwr

pro c MHz Pv Wpeag

hp rx Itanium pro c GHz

hp rx Itanium pro c GHz f DSmckinley O

Oinline budget

Ono ptrs to globals

hp rx Itanium pro c GHz

hp rx Itanium pro c GHz f DSmckinley O

Oinline budget

Ono ptrs to globals

hp zx Itanium pro c GHz

f DSmckinley O hp zx Itanium pro c GHz

Oinline budget

Ono ptrs to globals

IBM eServer pSeries Turb o

pro c MHz

IBM eServer pSeries Turb o

pro c MHz

IBM eServer pSeries Turb o O qarchpwr qtunepwr

pro c MHz Pv Wpeag

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Intel P MHz ifc O xW align ip o

Zp r

hp AlphaServer ES pro c

hp AlphaServer ES pro c v O

f Oinline Cray T pro c ns

hp zx Itanium MHz f DSmckinley O

Oinline budget

Ono ptrs to globals

Fujitsu Siemens hp cLineXeon GHz fc O align r ip o xW

Cray SVex pro cMHz

Cray SVex pro cMHz f Oinline

Cray T pro c ns f Oinline

IBM eServer pSeries

pro c MHz

IBM eServer pSeries O qarchpwr qtunepwr

pro c MHz Pv Wpeag

Cray SVex pro cMHz

Cray SVex pro cMHz f Oinline

Compaq ES pro c MHz

NEC SX pro c ns R pi wfprob use

IBM eServer pSeries C O qarchpwr qtunepwr

pro c MHz Pv Wpeag

IBM eServer pSeries C O qarchpwr qtunepwr

pro c MHz Pv Wpeag

IBM eServer pSeries E O qarchpwr qtunepwr

pro c MHz Pv Wpeag

IBM eServer pSeries E O qarchpwr qtunepwr

pro c MHz Pv Wpeag

AMD Athlon MP pro c MHz ifc O tpp ip o

Compaq ES pro c MHz kf fkapargsinlinedaxpy

ur ur arch host

assume nounderscore

NEC SX pro c ns

frt Wvr Of KA Fujitsu VPP pro c ns

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Intel P MHz ifc O xW align r ip o

Cray SV pro c MHz

f O inline Cray SV pro c MHz

Cray SV pro c MHz f O inline

Tyan SAMD Athlon XP MHz pro c

ifc tpp O Tyan SAMD Athlon XP MHz pro c

Intel P ACERVeriton MHz ifc O xW align r ip o

Cray SV pro c MHz f O inline

Cray T pro c ns f Oinline

AMD Athlon Thunderbird GHz ifc O tpp align

r ip o

Compaq DSL MHz pro c

Compaq DSL MHz kf fkapargsinlinedaxpy

ur ur archhost

assume nounderscore

Fujitsu Siemens Celsius Intel fortran

P GHz O xW align r

HP Sup erDome pro c MHz

HP Sup erDome pro c MHz f O

Oinlinebudget

Ono dataprefetch

Olo op unroll

hp server rppro c MHz

hp server rp pro c MHz

hp server rp pro c MHz f O

Oinlinebudget

Ono dataprefetch

Olo op unroll

hp server rp pro c MHz

f O hp server rp pro c MHz

Oinlinebudget

Ono dataprefetch

Olo op unroll

Compaq ES MHz cpu

assume nounderscore O Compaq ES MHz cpu

November

LINPACK Benchmark TPP Theoret Computer

n Best Eort Peak M

OSCompiler Mops n Mops

Dell PE ItaniumMhz pro c

Dell PE ItaniumMhz pro c e Ox Ob Ot

Cray TE pro c MHz

Cray TE pro c MHz f ver Oinline

Cray SV pro c MHz f O inline

IntelHP Itanium MHz f Ofast O Ono dataprefetch

HP i Itanium MHz pro c

HP i Itanium MHz pro c f Ofast O Ono datapretch

NEC SX pro c ns

NEC SX pro c ns R fopp fx inline

AMD Athlon Thunderbird Mhz g s static O fomitframep ointer

Wall mp entiumpro marchp entiumpro

malignfunctions funrolllo ops

fexp ensiveoptimizations

maligndouble fscheduleinsns

mwidemultiply

Compaq Server DSe pro c MHz

Compaq Server DSe MHz

Compaq Server ES pro c MHz

Compaq Server ES pro c MHz kf fkapargsinlinedaxpy

ur ur O tune ev

assume nounderscore

Cray SV pro c MHz

Cray SV pro c MHz f O inline

NEC SXBpro cns

NEC SXBpro cns R fopp fx inline

Tyan SPentiumI I I MHz pro c

Tyan SPentiumI I I MHz pro c ifc tpp O

IBM RS P

pro c MHzMBL

IBM RS P

pro c MHzMBL

IBM RS P O Q qoathst

pro cMHzMBL qarchpwr qtunepwr

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Pv Wpeag

IBM eServer pSeries Mo del B

pro c MHz MB L

IBM eServer pSeries Mo del B

pro c MHz MB L

O Q qoathst IBM eServer pSeries Mo del B

pro c MHz MB L qarchpwr qtunepwr

Pv Wpeag

IBM eServer pSeries Mo del E

pro c MHz MB L

IBM eServer pSeries Mo del E O Q qoathst

pro c MHz MB L qarchpwr qtunepwr

Pv Wpeag

IBM eServer pSeries Mo del C

pro c MHz MB L

IBM eServer pSeries Mo del C O Q qoathst

pro c MHz MB L qarchpwr qtunepwr

Pv Wpeag

IBM RS P MHz O Q qoathst

qarchpwr qtunepwr

Pv Wpeag

NEC SXCe pro c R fopp fx inline

AMD Duron MHz ifc static O tpp ip o

Fujitsu Siemens Celsius

P GHz pgf fast

Cray C pro c ns CF Zp Wde

HP Sup erDome pro c MHz

HP Sup erDome pro c MHz f O Oinlinedaxpy

Cray C pro c ns CF Zp Wde

HP N pro c MHz

HP N pro c MHz f O Oinlinedaxpy

NEC SXApro cns

NEC SXApro cns R fopp fx inline

NEC SXBA pro c ns

HP V pro c MHz

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

HewlettPackard V MHz f O Oinlinedaxpy

Compaq pro c ns

kf fkapargsinlinedaxpy Compaq pro c ns

ur tune ev O

NEC SXBe pro c ns R fopp fx inline

kf fkapargsinlinedaxpy Compaq Alpha Server DSMHz

ur tune ev O

Compaq pro c ns

Compaq pro c ns kf fkapargsinlinedaxpy

ur tune ev O

NEC SXBA pro c ns R fopp fx inline

IBM RSK P pro c MHz

IBM RSK P pro c MHz O Q qoathst

qarchpwr qtunepwr

Pv Wpeag

IBM RSK B pro c MHz

IBM RSK B pro c MHz O Q qoathst

qarchpwr qtunepwr

Pv Wpeag

IBM eServer pSeries

pro cMHzMB L

IBM eServer pSeries

pro cMHzMB L

IBM eServer pSeries

pro cMHzMB L qarchpwr qtunepwr

Pv Wpeag

IBM RSK P

pro cMHzMB L

IBM RSK P

pro cMHzMB L

IBM RSK P O Q qoathst

pro cMHzMB L qarchpwr qtunepwr

Pv Wpeag

IBM eServer pSeries

pro cMHzMB L

IBM eServer pSeries

pro cMHzMB L

IBM eServer pSeries

pro cMHzMB L qarchpwr qtunepwr

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Pv Wpeag

IBM RSK P

pro cMHzMB L

IBM RSK P

pro cMHzMB L

O Q qoathst IBM RSK P

pro cMHzMB L qarchpwr qtunepwr

Pv Wpeag

IBM RSK SP Power pro c MHz

IBM RSK SP Power pro c MHz O Q qoathst qarchpwr

qtunepwr Pv Wpeag

Cray pro c ns CSOS level

CompaqDEC Alpha EV MHz O arch host tune host

Hitachi S pro c ns

Hitachi S pro c ns OSF MJ FORTRANV

IBM RSK SP pro c MHz

IBM RSK SP pro c MHz xlf O Q qoathst

qarchpwr qtunepwr Pv

Wpeag

Cray pro c ns CSOS level

Cray C pro c ns CF Zp Wde

HP N pro c MHz

HP N pro c MHz f O Oinlinedaxpy

HP V pro c MHz

HP V pro c MHz f O Oinlinedaxpy

NEC SXR pro c ns

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

NEC SXR pro c ns

NEC SXR pro c ns fsx R pi

Intel P MHz g O fomitframep ointer

funrolllo ops

IBM eServer pSeries F

CPUs MHz MB L

IBM eServer pSeries F

CPUs MHz MB L

IBM eServer pSeries F

CPUs MHz MB L

IBM eServer pSeries F xlf O Q qoathst

qarchpwr qtunepwr Pv

CPU MHz MB L Wpeag

IBM eServer pSeries H

CPUs MHz MB L

IBM eServer pSeries H

CPUs MHz MB L

IBM eServer pSeries H

CPUs MHz MB L

IBM eServer pSeries H xlf O Q qoathst

qarchpwr qtunepwr Pv

CPU MHz MB L Wpeag

Sun UltraSparc I I I MHz fast native xsafemem

dalign xO xarchvplusa

xchipultra

Cray pro c ns CSOS level

Intel P MHz g O fomitframep ointer

funrolllo ops

Gigabyte GAVXAMD Athlon MHz ifc tpp O

IBM RS MHz ThinNo de qarchpwr qhot O Pv

Wpeag

Compaq XP MHz kf tune ev O

fkapargsinlinedaxpy ur

NEC SX pro c ns

fsx R pi NEC SX pro c ns

NEC SX pro c ns fsx R pi

DEC pro c MHz

DEC pro c MHz f O fast

CF Zp Wde Cray YMP pro c ns

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Compaq Alpha Server dsMHz fast O arch ev tune ev

DEC pro c MHz

DEC pro c MHz f O fast

qarchpwr qhot O Pv IBM RSK MHz WideNo de

Wpeag

IBM RSK SP PowerSMP Pro c MHz

AMD Athlon Mhz g O s funrolllo ops

fomitframep ointer

IBM RSK SP PowerSMP Pro c MHz O Q qoathst

qarchpwr qtunepwr bnso

bIlibsyscallsexp Pv

Fujitsu VP ns FORTRAN EXVP VL

DEC pro c MHz kf inlinedaxpyur

fast O tune ev

Intel Pentium I I I MHz g O fomitframep ointer

funrolllo ops

IBM PSC MHz Thin No de qarchpwr qhot O Pv

Wpeag

Apple PowerPC G MHz g O fomitframep ointer

funrolllo ops

DEC PersonalWorkstation O fast tune ev

inline all sp eculate all

Cray YMP pro c ns CF Zp Wde

Sun Ultra pro c MHz

Sun Ultra MHzMB L fast xO xarchvplusa

xchipultra

Fujitsu VPP pro c ns FORTRANEXVP VL

DEC pro c MHz kf inlinedaxpyur

fast O tune ev

DEC pro c MHz kf inlinedaxpyur

fast O tune ev

DEC pro c MHz kf inlinedaxpyur

fast O tune ev

DEC pro c MHz kf inlinedaxpyur

fast O tune ev

Cray YMP M pro c ns CF Zp Wde

Fujitsu VX pro c ns FortranVP VL

Fujitsu VPP pro c ns FortranVP VL

FortranVP VL Fujitsu VPP pro c ns

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Fujitsu VP ns FORTRAN EXVP VL

HP Exemplar VClass pro c MHz O Oinlinedaxpy

O Oinlinedaxpy HP Exemplar VClass pro c MHz

HP Exemplar VClass pro c MHz O Oinlinedaxpy

O Oinlinedaxpy HP Exemplar VClass pro c MHz

HP Exemplar VClass pro c MHz HPUX

O Oinlinedaxpy

Cray S pro c ns CSOS level

NEC SXR pro c ns fsx R pi

NEC SXLR pro c ns fsx R pi

HewlettPackard C MHz O Oinlinedaxpy

Intel Pentium I I I MHz g O

DEC pro c MHz kf inlinedaxpy ur

fast O tune ev

DEC pro c MHz kf inlinedaxpy ur

fast O tune ev

DEC pro c MHz kf inlinedaxpy ur

fast O tune ev

DEC pro c MHz kf inlinedaxpy ur

fast O tune ev

DEC A pro c MHz kf inlinedaxpy ur

fast O tune ev

Sun HPC MHz pro c

Sun HPC MHz MB L fast xO xarchvplusa

xchipultra

Cray YMP pro c ns CF Zp Wde

Cray XMP pro c ns CF Zp Wde

Cray YMP M pro c ns CF Zp Wde

Sun UltraSparc I I MHz fast native xsafemem

dalign xO xarchvplusa

xchipultra

SGI Origin Mhz pro c

SGI Origin Mhz f IPA O n mips

r call shared TENVX

OPTIEEE arithmeticroundo

LNOblo ckingoou maxpf

INLINEarray bounds

fsx R pi NEC SX pro c ns

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Sun UltraSparc I I MHz fast native xsafemem

dalign xO xarchvplusa

xchipultra

NEC SXL pro c ns fsx R pi

SGI Octane MHz IP f O

FORTRAN EXVP VL Fujitsu VP ns

HP Exemplar VClass pro c MHz HPUX

HPUX HP Exemplar VClass pro c MHz

HP Exemplar VClass pro c MHz HPUX

O Oinlinedaxpy

SGI Octane R IP MHz O OPTOlimit

TARGplatformIP

LNOblo ckingOFF

Compaq Alpha EV MHz g O fomitframep ointer

funrolllo ops

Cray S pro c ns CSOS level

HewlettPackard C MHz O Oinlinedaxpy

DEC pro c MHz kf fkapargsinlinedaxpy

ur tune ev O

assume nounderscore

DEC pro c MHz

Cray YMP pro c ns CF Zp Wde

ASUS PBFCeleron MHz Pro c ifc tpp O

Convex CXA pro c fc tm c O ds

ns ep is

HewlettPackard KEG MHz Oall Oinlinedaxpy

HewlettPackard CXP MHz Oall Oinlinedaxpy

HP Exemplar SClass pro c SPPUX

SPPUX HP Exemplar SClass pro c

HP Exemplar SClass pro c SPPUX

HP Exemplar SClass pro c SPPUX Oall Oinlinedaxpy

Sun UltraSPARC I I pro cMHz

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Sun UltraSPARC I I pro cMHz

Sun UltraSPARC I I pro cMHz fast xO xarchvplusa

xchipultra o

Cray YMP M pro c ns CF Zp Wde

DEC AlphaStation MHz fkapargsinlinedaxpy

ur tune ev O

Convex CXA pro c fc tm c O ds

ns ep is

Cray YMP M pro c ns CF Zp Wde

Cray XMP pro c ns CF Zp Wde

IBM RSR MHz v xlf Pv Wpmeew

O qarchpwrx qtunepwrx

qhotqhst qnosave

DEC Alphastations MHz f O

HewlettPackard C MHz Oall Oinlinedaxpy

IBM POWER MHz OPvWpeagqarchpwrx

DEC pro c MHz kf inlinedaxpy ur

fast O tune ev

DEC pro c MHz kf inlinedaxpy ur

fast O tune ev

DEC pro c MHz kf inlinedaxpy ur

fast O tune ev

DEC pro c MHz

DEC pro c MHz inlinedaxpy ur fast

O tune ev

DEC pro c MHz

inlinedaxpy ur fast DEC pro c MHz

O tune ev

Pentium I I I MHz gnufO

SGI Octane R IP MHz O OPTOlimit

TARGplatformIP

LNOblo ckingOFF

g v Apple PowerBo ok G MHz

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

IBM RSH MHz v xlf Pv Wpmeew

O qarchpwrx qtunepwrx

qhotqhst qnosave

IBM POWER mo del MHz OPvWpeagqarchpwrx

Convex CXA pro c fc tm c O ds

ns epis

Cray J pro c ns CF Zp Wde

CF Zp Wde Cray J pro c ns

Cray J pro c ns CF Zp Wde

Fujitsu VP ns FORTRAN EXVP VL

SGI Octane MHz IP f O

Cray J pro c ns cf Zp Wd

SGI POWER CHALLENGE

MHz pro c

SGI POWER CHALLENGE

MHz pro c

SGI POWER CHALLENGE

MHz pro c

SGI POWER CHALLENGE

MHz pro c

SGI POWER CHALLENGE non shared OPT

MHz pro c IEEE arithmeticroundo

TENVX col WKur

ur WKsoroo

WKinlinedaxpydscalidamax

SWPmax pair candidates

SWPstrict ivdepfalse

Cray J pro c ns CF Zp Wde

Cray XMP pro c ns CF Zp Wde

Cray S pro c ns CSOS level

DEC pro c MHz

DEC pro c MHz inlinedaxpy ur fast

O tune ev

Cray J pro c ns cf Zp Wd

ASUS PBDPentiumI I I MHz pro c

ASUS PBDPentiumI I I MHz pro c ifc tpp O

IBM RS F MHz pro c

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

IBM RS F MHz pro c

IBM RS F MHz pro c O qhot qarchpp c

qoathst Pv Wpea

g bnso bIlibsyscallsexp

bno delcsect

SGI Origin

MHz pro c

SGI Origin

MHz pro c

SGI Origin

MHz pro c

SGI Origin

MHz pro c

SGI Origin n mips Ofastip TENVX

MHzpro c LNOblo ckingoou maxpf

Sun UltraSparc I I MHz fast native xsafemem

dalign xO xarchvplusa

xchipultra

Fujitsu VP ns FORTRAN EXVP VL

Cray J pro c ns CF Zp Wde

Sun Ultra HPC MHz p

Sun Ultra HPC fast nativexarchvplusa

MHzMB L xsafemem dalign libmil xO

fsimple stackvar xarchvplusa

xcache xchipultra

xdep end xlibmil xlibmopt xsafemem

Qoption cg Qms pip eoat lo op ld

xcrossle

Cray J pro c ns cf Zp Wd

Hitachi S ns FORTHAP VC

Cray J pro c ns CF Zp Wde

Cray J pro c ns cf Zp Wd

Intel v O G QxM Dell Dimension XPS T MHz

Qip Qauto Qrcd

Cray S pro c ns CF Zp Wde

IBM POWER mo del H MHz OPvWpeagqarchpwrx

SGI POWER CHALLENGE

MHz pro c

SGI POWER CHALLENGE

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

MHz pro c

SGI POWER CHALLENGE

MHz pro c

SGI POWER CHALLENGE

MHz pro c

SGI POWER CHALLENGE

MHz pro c

SGI POWER CHALLENGE

MHz pro c

SGI POWER CHALLENGE

MHz pro c

SGI POWER CHALLENGE

MHz pro c

SGI POWER CHALLENGE

MHz pro c

SGI POWER CHALLENGE non shared OPT

MHz pro c IEEE arithmeticroundo

TENVX col WKur

ur WKsoroo

WKinlinedaxpydscalidamax

SWPmax pair candidates

SWPstrict ivdepfalse

Convex CXA pro c ns fc tm c O is

Intel Pentium I I Xeon MHz g funrollalllo ops O

ETAGproc ns ETAVFTN

Convex C pro c ns fc tm c O ep ds is

IBM ES VF pro c ns VASTVS Fortran VR

Dell Dimension XPS T MHz WinSE Intel Fortran

Pentium I I I O G QaxK Qip o

Intel Pentium I I I MHz g O fomitframep ointer

funrolllo ops

HALstation mo del MHz Kfast Keval KGREG Kgs

KVPLUS X Kpreex

Kpreload Kfuse x FLDFLAGS dn

Dell Pentium I I I MHz f O

SUNUltra mo d f v fast O

Convex C pro c ns fc tm c O ep ds is

Intel Pentium I I I MHz g O

Kfast Keval KGREG Kgs HALstation mo del MHz

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

KVPLUS X Kpreex

Kpreload Kfuse x FLDFLAGS dn

SGI CHALLENGEOnyx

ns pro c

SGI CHALLENGEOnyx

ns pro c

SGI CHALLENGEOnyx

ns pro c

SGI CHALLENGEOnyx

ns pro c

SGI CHALLENGEOnyx

ns pro c

SGI CHALLENGEOnyx

ns pro c

SGI CHALLENGEOnyx

ns pro c

SGI CHALLENGEOnyx

ns pro c

SGI CHALLENGEOnyx

ns pro c

SGI CHALLENGEOnyx IRIX fOmipsWo

ns pro c lo opunrollOlimitWf

dchacheoptjmp optnon shared

pfa keepWK WK

ipadaxpysaxpyurmc

Convex C pro c ns

fc tm c O ep ds is

Sun UltraSPARC pro cMHz

Sun UltraSPARC pro cMHz V fast native dalign

libmil xO xsafeDmem

Qoption cg

Qms pip eoat lo op ldD

onetrip xcrossle

IRIX fOmipsWo SGI CHALLENGEOnyx

ns pro c lo opunrollOlimitWf

dchacheoptjmp optnon shared

pfa keepWK WK

ipadaxpysaxpyurmc

Intel Pentium I I MHz g O funrollalllo ops

Convex SPP pro c MHz

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Convex SPP pro c MHz

Convex SPP pro c MHz fc fc is

Dell Pentium I I I MHz f O

SUNUltra mo d f v fast O

Convex C pro c ns

fc tm c O ep ds is

AMD KI I MHz BCM ATX g O o gldst

Cray pro c ns cf

ETA E pro c ns ETAVFTN

GatewayGPentiumPro MS Fortran NT G Oxb

IBM ES VF pro c ns VASTVS Fortran VR

SGI CHALLENGEOnyx IRIX fOmipsWo

ns pro c lo opunrollOlimitWf

dchacheoptjmp optnon shared

pfa keepWK WK

ipadaxpysaxpyurmc

Cray XMPse ns cf

DEC pro c ns

DEC pro c ns O fast

IBM RS MHz v xlf Pv Wpfzmeew

O Q qstrict qarchpwrqtune

pwrx qhot qhst qnosave

DEC AMP pro c

DEC AMP pro c O fast

DEC pro c ns O fast

AMD KI I Mhz g O

Convex SPP pro csMHz

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Convex SPP pro csMHz

Convex SPP pro cs MHz

Convex SPP pro c MHz fc fc is

cf Cray pro c ns

IBM ES pro c ns VASTVS Fortran VR

DEC pro c ns O fast

O fast DEC pro c ns

CompaqDEC Alpha EV MHz g O

Convex C pro c ns fc tm c O is

DEC pro cs ns

DEC pro c ns O fast

DEC Alpha AXP ns O fast

DEC AMP pro cs ns O fast

DEC AMP pro c ns O fast

DEC Alpha AXP pro c

DEC Alpha AXP MHz inldaxpyurur

NEC SX ns FORTRAN SX

Cray YMP EL pro c ns CF Zp Wde

HP MHz OP WlaarchiveWPnv

w ConvexMLIB

Compaq Proliant MHz MS Power Stat Full Opt

Cray YMP EL pro c ns CF Zp Wde

Cray SMPv pro c ns uf veccollapse pi

Cray YMP EL pro c ns CF Zp Wde

DEC pro cs ns

DEC pro cs ns O fast

DEC ns O fast

HPJ MHz O DC Odataprefetch

Cray pro c ns cf

IBM RISC Sys MHz v xlf O P Wpea

v xlf O P Wpea IBM RISC Sys MHz

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

IBM ES pro c ns VASTVS Fortran VR

SGI CHALLENGEOnyx IRIX fOmipsWo

ns pro c lo opunrollOlimitWf

dchacheoptjmp optnon shared

pfa keepWK WK

ipadaxpysaxpyurmc

DEC Alpha AXP MHz inldaxpyurur

Solaris GNU F v Pentium Pro Mhz

NEC SX FORTRAN SX

Cray YMP EL pro c ns CF Zp Wde

Apple Macintosh MF O Aschedtarg

Convex C pro c fc fc O ep ds is

Cray YMP EL pro c ns CF Zp Wde

ETA Q pro c ns ETAVFTN

Cray YMP EL pro c ns CF Zp Wde

DEC Alpha AXP ns O fast

Cray SMPMCP pro c ns

Cray SMPMCP pro c ns MCP Release

DEC pro c ns O fast

FPS S MCP pro c ns

FPS S MCP pro c ns pgf O Minline

CDC Cyb er V Fortran V

Convex C pro c fc fc O ep ds is

Macintosh MHz Absoft Corpc O o

NEC SXE FORTRAN SX

SGI Indigo RMHz mips Olimit Wo

lo opunroll Wfdcacheopt

Wfdcacheoptx O non shared

Alliant FX pro c fortran O inline

IBM RISC Sys MHz v xlf O P Wpea

IBM RS Cluster pro c MHz

IBM RISC Sys MHz v xlf O P Wpea

VASTVS Fortran VR IBM ES VF pro c ns

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

IBM ES VF pro c ns VASTVS Fortran VR

inldaxpyurur DEC Alpha AXP MHz

Hitachi SR pro c MHz f PVECOPTSFOLD

SGI CHALLENGES Mhz RSC IRIX f O mips

Alliant FX pro c fortran O inline

HP MHz HPUX f OP

IBM R VASTVS Fortran

Sun Sparc MHz pro c Sun fast unroll O

Alliant FX pro c fortran Ovg inline

Alliant FX pro c fortran O inline

ETA P pro c ns ETAVFTN

Convex C pro c fc fc O ep ds is

CrayS ns cf

Convex C pro c fc O ep uo ppfcpp is

Convex C pro c O ep uo ppfcpp is

Convex C pro c fc O ep uo ppfcpp is

Convex C pro c O ep uo ppfcpp is

DEC Alpha AXP ns O fast

DEC Alpha AXP MHz inldaxpyurur

IBM RISC Sys MHz v xlf O P Wpea

SGI CHALLENGEOnyx IRIX fOmipsWo

ns pro c lo opunrollOlimitWf

dchacheoptjmp optnon shared

pfa keepWK WK

ipadaxpysaxpyurmc

Alliant FX pro c fortran O inline

NAS ASEX VPF pro c

NAS ASEX VPF VASTVS opt

HP MHz OP WlaarchiveWPnvw

VASTVS Fortran VR IBM ES Mo del VF

IBM ES VF ns VASTVS Fortran VR

IBM ES VF pro c ns VASTVS Fortran VR

Meiko CS pro c

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Meiko CS pro c

Meiko CS pro c dalign O XTsshunroll

Fujitsu M EX VL frt Of Ne

Sun Sparc pro c Sun fast O

unroll Bstatic

DEC VAX VP pro c ns HPO VV DXML

VASTVS Fortran VR IBM ES VF pro c ns

IBM ES VF pro c ns VASTVS Fortran VR

Apple Macintosh MF O Ashedtarget

IBM ES VF pro c ns VASTVS Fortran VR

IBM RISC Sys MHz v xlf O P Wpea

MultiowTRACE Fortran

Convex C pro c fc O ep uo ppfcpp is

Convex C pro c O ep uo ppfcpp is

Alliant FX pro c fortran O inline

Siemens VPEX ns Fortran VP VL

IBM ES ns VASTVS Fortran

Apple Macintosh MF O Aschedtarg

Apple Power Mac Absoft Power PC v O U

FPS Mo del F

Fujitsu VP Fortran VL

IBM RISC SysH MHz v xlf O P Wpea

IBM RSC MHz v xlf Pv Wpfzme

ew O qarchpp c qhot

qhst qnosave qnofold

IBM ESR ns VASTVS Fortran

Siemens VPEX ns Fortran VL

Amdahl VP VL

Apple Power Mac Absoft Power PC v O U

Convex C pro c fc fc O is

GatewayP MS PS NT G Oxb

IBM ES Mo del VF ns VASTVS Fortran VR

IBM RISC SysL MHz v xlf O P Wpea

v xlf O P Wpea IBM RISC Sys MHz

IBM RISC Sys MHz v xlf O P Wpea

IBM RISC SysH MHz v xlf O P Wpea

IBM ES VF pro c ns VASTVS Fortran VR

Cray SMP pro c ns uf Oc a

MS Power Stat Full Opt Compaq Deskpro MHz

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Fujitsu VP Fortran

HP MHz HPUX f OP O

IBM ES ns VASTVS Fortran

NAS ASEX VPF VASTVS

SGI D pro c MHz f O mp

Fortran VP VL Siemens VPEX ns

Sun MP Ross Hyp ersparcMhz cg dalign libmil O

Apple PowerMacintosh Motorola MF O

Apple Power Mac Absoft f v O

Alliant FX pro c fortran O inline

Amdahl VP VL

CDC CYBER pip e FTN

Convex C pro c fc O uo ppfcpp is

Convex C pro c O uo ppfcpp is

Cray XMS ns cf Zp Wde

Hitachi S FORTHAP

IBM ES Mo del VF ns VASTVS Fortran VR

Siemens VPEX ns Fortran VP VL

MultiowTRACE Fortran

Amdahl VP VL

Fujitsu VP Fortran

Hitachi MHvector Fort E VI

Hitachi S HAP V

IBM J VF pro c ns

IBM J VF pro c ns VS Fortran VR

PowerPC MHz LS Fortran prerelease

SGI Crimson pro c MHz R O mips G

SGI D pro c MHz f O mp

SGI Indigo ExtremeRMHz O mips G

FPS Mo del F

Hitachi MH Fort E VI

IBM RISC Sys MHz v xlf O P Wpea

v xlf O P Wpea IBM RISC Sys MHz

IBM RISC Sys MHz v xlf O P Wpea

IBM ES pro c ns VASTVS Fortran VR

Kendall Square pro c

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Kendall Square pro c ksrf O r inline auto

NAS ASEX Fortran

SGI D pro c MHz f O mp

Siemens HF Fortran

Apple Power Mac Absoft f v O

Power Computing Absoft f Power PC v

Cydrome CYDRA Fortran Rel

Fujitsu VP Fortran

VASTVS Fortran VR IBM ES Mo del VF ns

IBM ES ns VASTVS Fortran

Apple Power Mac Absoft f Power PC v

DELL XMT Pentium MHz PS NT V G Oxb

IBM POWERPC MHz OPvWpeagqarchpwrx

IBM E VF VS opt

SGI D pro c MHz f O mp

Apple Power Mac Absoft f Power PC v

CDC CYBER E FTN V VLHIGH

CrayS ns run CFT

Gateway PXL MS PS G Ox D NDEBUG

IBM VF VS Fortran V

IBM RISC SysH MHz v xlf O P Wpea

SGI Indigo MHz O mips G sopt

Stardent inline nmax

Stardent Stellar GS f O is R

Stardent Stellar GS f O is re R

CDC InfoServer MHz f O mips Wbr

Cray SMPMCP pro c ns MCP Release

FPS S MCP pro c ns pgf O

IBM ES Mo del VASTVS Fortran VR

IBM ES pro c ns VASTVS Fortran VR

Meiko Comp Surface pro c

Meiko Comp Surface pro c O Mvectsmallvect

Meiko Comp Surface pro c Minlinedaxpy

Windows NT G Oxb Gateway P MHz Pentium

SGI Power Series MHz R O mips G sopt

Stardent inline nmax

Sp erry ext wISP UCS level

MultiowTRACE Fortran

Alliant FX pro c fortran O inline

O DAS inline Alliant FX pro c

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

IBM J VS Fortran VR

Intel Paragon pro c O Mvectsmallvect

Minlinedaxpy Knoieee

MIPS RC MHz f O

Fortran MultiowTRACE

Stardent inline nmax

Stardent Ardent Titan

Stardent Ardent Titan f O inline

Sun Sparc MHz pro c

Sun Sparc MHz pro c fast O unroll Bstatic

Intel iPSCDelta pro c

Intel iPSCDelta pro c if O Mvectsmallvect

Intel iPSCDelta pro c Minlinedaxpy Knoieee

Intel iPSC d pro c

if O Mvectsmallvect Intel iPSC d pro c

Intel iPSC d pro c Minlinedaxpy Knoieee

SGI D pro c MHz f O mp

Apple Power Mac Absoft f Power PC v

Apple Power Macintosh Absoft v F O

IBM S VS Fortran

O DAS inline Alliant FX pro c

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

CDC CYBER f o

IBM Power Vis Sys pro c

IBM Power Vis Sys pro c O Minlinedaxpy

NAS ASEX Fortran

Sun SPARCsystem MHz f O cg libmil native

SGI D pro c MHz f O mp

IBM RISC Sys MHz v xlf O P Wpea

VASTVS Fortran VR IBM ES VF ns

Solb ourne Viking sparc f O cg dalign

IBM RISC Sys MHz v xlf O P Wpea

DEC VAXvector pro c Fortran HPO V

Comparex Fujitsu M VSFORTRAN

DEC VAXstation V

Apple Power Macintosh Absoft v F O

IBM ES pro c ns VASTVS Fortran VR

IBM ES Mo del VASTVS Fortran VR

NAS AS VASTVS opt

Alliant FX pro c O DAS inline

IBM ES Mo del VASTVS Fortran VR

SCS CFT

SGI D pro c MHz f O mp

IBM ES Mo del VASTVS Fortran VR

IBM ES Mo del VSFORTRAN VR

IBM E VF VS opt

IBM E VS opt

Siemens F Fortran V

Convex C Fortran

Alliant FX pro c O DAS inline

DEC VAXvector pro c Fortran HPO V

CECpxXLPentium MHz wfc ldosg ox

Sun SPARCsystem MHz f native fast O Bstatic

Stardent Ardent Titan f O inline

IBM VS opt

Alliant FX pro c O DAS inline

IBM RSNPowerPC MHz xlf O P Wpea

IBM RISC SysM MHz v xlf O P Wpea

IBM RISC SysMA MHz v xlf O P Wpea

IBM ES Mo del VASTVS Fortran VR

fc Convex C

IBM RISC Sys MHz v xlf O P Wpea

Alliant FX pro c O DAS inline

Alliant FX pro c fortran O inline

Apple PowerBo ok PBcs MHz MF O Aschedtarg

Fujitsu M Fortran opt

DEC VAX V

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

MultiowTRACE Fortran

SGI D pro c MHz f O

Apple Performa CD Absoft f Power PC v

Siemens G Fortran V opt

IBM E VS opt

F APFTN OPT FPS M

Alliant FX pro c O DAS inline

SGI D pro c MHz f O mp

f Ap ollo DN

DEC VAX opt DEC Fortran V

HPJ MHz fort o

Alliant FX pro c O DAS inline

Gateway P MHz Pentium FLEM Z

DEC RISC Ultrix

DEC Ultrix

Gateway P MHz Pentium mfdwarnfpot

Alliant FX pro c O DAS inline

CDC MHz f O

Numb erSmasher MHz NDP vastinlineonOLMfdiv

VAXstation DEC FORTRAN V

DEC VAX pro c VMS V

Intel iPSC dVX pro c

SGI D pro c MHz f O

Honeywell DPS ES FV

Siemens D Fortran V

IBM ES Mo del ns VASTVS Fortran VR

CDC CYBER FTN opt

Numb er Smasher i MHz on OLM fdiv inline

CDC CYBER FTN opt

MIPS RC MHz f O

Alliant FX pro c O DAS inline

AMD DX fc and gcc

Alliant FX pro c O DAS inline

NAS ASEX VS opt

SGI D f O

Sun MP f O cg dalign

IBM ES ns VASTVS Fortran

Sun SPARCstation IPX f O cg dalign

Sun IPX f O cg dalign

CDC CYBER f o

Sun SPARCstation f O cg dalign

SGI Indigo MHz R O G sopt

Amdahl HSFPF H enhanced opt

MIPS M MHz f O

f O MIPS RC MHz

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Alliant FX pro c O DAS inline

SGI D pro c MHz f O

Amdahl HSFPF VS opt

CDC f opt

DEC station Mhz MIPS f

MIPS RS MHz f O

DEC VAXvector pro c Fortran HPO V

f O Sun

CDC f opt

Apple Power Macintosh Absoft F SDK

NAS wHSA VS opt

CDC FTN

Sun Sparc ELC dalign xcg fsimple O

CDC CYBER NOSVE FTN

Gould NP Fortran

IBM E VS opt

MIPS RC MHz f O

Tadp ole SPARCb o ok f O

CDC CYBER f o

Convex CXP Fortran

DEC VAX VMS

FPS M F APFTN OPT

Harris Nighthawk f

Convex CXL Fortran

IBM ES Mo del VS Fortran VR

NAS ASEX VS opt

Solb ourne f Sun O dalign

Sun f O dalign

Sun f O cg dalign

CDC CYBER FTN opt

Cyb erPlus CPFTN

IBM H enhanced opt

Sun SparcServer f O dalign

Alliant FX pro c O DAS inline

Gateway MHz DX FLEM Z

Apple Mac Quadra AV Absoft w v O f s N

HPAPOLLO e f O rev

NAS ASEX VS opt

Fujitsu AP pro c

Fujitsu AP pro c Sun f O dalign

f O rev HPAPOLLO t

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Alliant FX pro c O DAS inline

CDC CYBER FTN opt

CDC CYBER NOSVE OPTHIGH

FPSM APFTN OPT

IBM ES Mo del VS Fortran VR

IBM K pro c H enhanced opt

MIPS M UMIPS v f O

MIPS M MHz f O

Salford v optimise Prism MHz

Tadp ole SPARCb o ok MHz f O

Apple Macintosh QUADRA Absoft w v O f s N

CDC Lo cal

IBM K pro c VS opt

Culler PSC CSD Fortran

FPS M APFTN

Micronics MHz EISA NDP Fortran on

HP T

CDC CYBER FTN opt

HP Series fc O

Sp erry FTN optZEO

Sun SPARCstation f O cg dalign

ELXSI pro c

ELXSI pro c EMBOS optinlinevector

FPS M F APFTN OPT

Honeywell DPS FRX

IBM H enhanced opt

IBM VS opt

IBM D VS opt

MIPS RS MHz f O

Sp erry ext UFTN

HP Series wfp fc O

Amdahl V H enhanced opt

CDC CYBER FTN opt

CDC CYBER NOSVE OPTHIGH

DECstation VV O

DEC f O

Amdahl V VS opt

DEC VAXstation V

MIPS M MHz f O

NAS VS opt

Siemens P For A

ALR mb oard K cache Lahey FL v Z

Apple Mac Quadra Absoft w v O f s N

Compaq Deskpro l w Microway NDPF O OL on

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

NeXTCub e gcc O

Sun SPARCstation f O cg dalign

IBM VS Fortran opt

Compaq Deskpro l w Salford FTN optimized

Compaq Deskpro l w Watcom WFCP OL OT

ALR mb oard K cache Lahey FL v nZ

CDC CHAT No opt

CSPI MAP Fortran

DEC VAX pro c

DEC VAX pro c VMS V

ELXSI Fortran opt

Gateway Micronics DX fc emxgcc O m

GatewayPentium HHz Lahey F

IBM ES Mo del VS Fortran VR

IBM Fast Mult H Ext

IBM E VS Fortran opt

IBM VS opt

MIPS M MHz f O

Prime P f rev b opt

Siemans E BS

Amdahl V Hopt

Compaq Deskpro l w Lahey FL Z

Sun f O sysb eta

ES pro c ns Russian flike IBM VS OPT

SonyPlaystation gcc Linux

CDC CYBER NOSVE OPTHIGH

DEC VAX pro c

DEC VAX VMS v

Solb ourne f O

IBM VS Fortran opt

IBM MG VS Fortran opt

IBM VS Fortran opt

ICL wFPU FORTRAN PLUS V

IBM MHz Microsoft

Siemens E Fortran V

Concurrent XP Fortran VI IZ

MIPS M wR FP f

Gould PN VTX Fortran

VAXstation DEC FORTRAN V

IBM VS Fortran opt

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

nCUBE pro c

nCUBEproc

nCUBE pro c

nCUBE pro c Fortncc O

IBM Fast Mult HExt

Prime PI I f rev b opt

DEC VAX VMS v

HP Series fc O

DEC VAX VMS v

HPAp ollo DN FPA

Mentor Graphics Computer fortran

MIPS M HHz f O

Data General MV f

IBM VS Fortran opt

Sp erry wSAM FTN optZEO

CDC CYBER NOSVE

Russian PS FORTRANPS

GatewayDX HHz Lahey F

Harris H VOS opt g

HPAp ollo DN

HP Series fc O

HPAPOLLO t f O rev

Harris HCX hf O

Pyramid OSx

HP Series fc O

DEC VAX VMS v

Harris HCX wfpp f

CDC FTN opt

CDC CYBER FTN opt

CCI Power wfpa UNIX bsd f

IBM VS Fortran opt

Sp erry

Gould PN UNIX

SUN FPA f O pa

IBM MG VS Fortran opt

DEC VAX pro c VMS v

CDC CYBER FTN opt

HP Series HPUX

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

IBM RT AIX

Harris H VOS opt g

microVAX VMS v

Apple Macintosh I Ifx AUX f

Ap ollo DNxxT FPX DOMAINIX SR opt

microVAX ULTRIX VFU

IBM VS Fortran opt

Sun FPA f O pa

Prime P f rev b opt

Ridge Mo del ROSrf

IBM VS Fortran opt

Gould mult acc fort

NORSK DATA ND FortranE

Sp erry FTN optZEO

Apple Mac I Ifx Absoft w v O f s

CDC CYBER NOSVE OPTHigh

CSA wTC Fortran L

Inmos T MHz Fortran L o

Sequent Symmetry wfpa Fortran fpa O

CONCEPT UTX

Celerity C UNIX bsd f

IBM RT PC fpa f

IBM VS Fortran opt

CDC RUN

Gould PN UTX

Prime F

Opus Series pm MHz UNIX Greenhills

Masscomp MC wfpa f v O rtv v

Data General MV f opt level

IBM MG VS Fortran opt

DATEK w KB Cache MS Fortran Ox AH G

Inmos T MHz Fortran L o

Ap ollo DN FTN CPU opt

IRIS Turb oFPA f

CDC CYBER NOSVE OPTHIGH

Apple Macintosh PowerBo ok Absoft w v O f s

Gould PN VTX Fortran

Harris Fortran

IBM H opt

IBM VS Fortran opt

NORSK DATA ND Fortran

Celerity C UNIX bsd f

Honeywell DPS FRX

Denelcor HEP f UPX

VAX FPA VMS v

CDC CYBER FTN opt

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Apple Macintosh I Isi Absoft w v O f s

Itel AS mo d H

NORSK DATA ND FortranE

KONTRON KSM UNIX SVS F

Sun i MHz SunOS Sun O

CDC CYBER FTN opt

IBM MG VS Fortran opt

Ap ollo DN

Pyramid xe OSx

IBM VS Fortran opt

VAX FPA UNIX bsd f

DEC VAX UP VMS v

CDC CYBER FTN opt

Ridge ServerRT EFP ROSrf

CDC CYBER FTN opt

Ridge OS RISC

PC Craft MHz w PLI Fortran

Concurrent OS fortran z

Tandy MC MHz LPI Fortran

Tektronix w UTEK f

CDC CYBER NOSVE OPTHIGH

Prime P f rev b opt

Apple Macintosh I Ix AUX f

Concurrent OS v f

Compaq w Microsoft Fortran

Apple Macintosh I Icx Absoft w v O f s

Apple Macintosh I Ix Absoft w v O f s

DEC VAX VMS v

IBM PS MHz AIX

Apple Macintosh SE Absoft w v O f s

Ap ollo DN DOMAINIX SR opt

ICL f OPT

IBM VS Fortran opt

HP Vectra RSC MHz LPI Fortran

VAX FPA VMS v

Compaq w RMForrtan

microVAX I I VMS v

Prime P f rev b opt

Apple Macintosh I Isi Fortran

Apple Mac I I Mhz Mhz Absoft w v O f s

CDC FUN

CONCEPT UTX

IBM PS MHz AIX

IBM RT w f

VAX FPA VMS v

micro VAX I I ULTRIX VFU

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Concurrent OS fortran

Denicon DSI SVS Fortran MSDOS

ENCORE Multimax NS f

HP Series HPUX f

Northgate MHz Lahey F

Prime Primos f v

Sun MHz f O f

Tektronix w UTEK f

VAX FPA UNIX BSD f O

Sun MHz f O f

NCUBE pro c MHz Fortran

Apple Mac SE ABSOFT

Ap ollo DN DOMAINIX SR opt

Masscomp MC f v O rtv v

VAX FPA UNIX bsd f

Prime Primos

Sp erry FTN optZEO

Pyramid X FPA UNIX bsd f

Apple Mac I I Mhz Mhz Absoft

SUN MHz f O f

HP Series HPUX f

Apple Macintosh I I Absoft w v O f s

microVAX I I f Ultrix

Apple Mac SE MHz ABSOFT

Ridge ROS RISC

Data General MV f opt level

Apple MAC I I w

Prime P f rev b opt

Apple MacLevco Pro digy ABSOFT MacFort

Apple Mac I I w FORTRAN

HP Series HPUX f

Ap ollo DN DOMAINIX SR opt

Ap ollo DN AEGIS FTN

Masscomp MC wFPP Fortran

Harris HS wFPP Fortran

Sequent Balance DYNIX Fortran

Denicon DSI Greenhills f MSDOS

VAX VMS v

Encore Multimax f

HP Series Fortran

Opus UNIX f bsd

ATT B FP UNIX V

Acorn Cambridge fortran

IBM MG Hopt

Burroughs B Fortran ver

VAX FPA VMS v

November

LINPACK Benchmark TPP Theoretical Computer

n Best Eort Peak Mops

OSCompiler Mops n Mops

Masscomp MCS wFPB Fortran

IBM RTPCModel f

VAX FPA VMS

Prime Fortran

IBM PCAT VS Fortran opt

IBM PCXT H opt

VAX UNIX bsd f

Ap ollo DN AEGIS FTN

Sun SKY FFP f O fsky

Ametek S no de RM Fortran

Ap ollo DN FPA AEGIS FTN

AMSTRAC MHz MSFortran Ox AH

microVAX I VMS

Canaan VS

Chas River Data SKY SVS Fortran

Ap ollo DN PEB AEGIS FTN

IBM ATw PROFORT

IBM PC w PROFORT

Cadtrak DS Intel Fortran

Apple Mac Classic I I MHz Absoft

IBM PCAT w Microsoft

Chas River Data SVS Fortran

Ap ollo DN AEGIS FTN

Masscomp MC Fortran

IBM PC w Microsoft

Apple Mac I I ABSOFT

HP Series HPUX

Sun f O fsoft

Atari ST ABSOFT ACFortran v

Apple Macintosh ABSOFT b

Palm Pilot I I I

HP GX

November

Table A Lo ok at Parallel Pro cessing

Computer x Problem with Parallel Pro cessing

Time no of Time Sp eedup Eciency

unipro cessor pro cessors multipro cessors

Hitachi S

NEC SXR

NEC SX

NEC SXR

NEC SX

CrayC

NEC SX

NEC SXR

Convex CXA

IBM ES ns

Cray YMP

IBM ES ns

CrayS

Cray XMP

November

Computer x Problem with Parallel Pro cessing

Time no of Time Sp eedup Eciency

unipro cessor pro cessors multipro cessors

Cray XMP

Convex C

DEC Alpha

Convex SPP i

Cray SMPMCP

DEC Alpha

MeikoCS

Fujitsu AP

IBM J ns

IBM S VF

November

Computer x Problem with Parallel Pro cessing

Time no of Time Sp eedup Eciency

unipro cessor pro cessors multipro cessors

IBM S VF

Kendall Square Research

IBM ES ns

Intel Delta

IBM E VF

Sun Sparc MHz

Alliant FX

IBM PVS

IBM RS Cluster ns

nCUBE

November

Computer x Problem with Parallel Pro cessing

Time no of Time Sp eedup Eciency

unipro cessor pro cessors multipro cessors

nCUBE

Intel iPSC

Meiko Computing Surface i

IBM RS Cluster ns

Sun Sparc MHz

Convex C

Parsytec FT

Sun Sparc MHz

FPS Mo del

Suprenum SC

November

Computer x Problem with Parallel Pro cessing

Time no of Time Sp eedup Eciency

unipro cessor pro cessors multipro cessors

Suprenum SC

Alliant FX

Stardent Ardent Titan

SGI D MHz

Sun Sparc MHz

Alliant FX

SGI D MHz

Alliant FX

DEC VAX

ELXSI

DEC VAX

Sequent Balance

November

Table Highly Parallel Computing

Computer Number of R N N R

max max peak

Full Precision Pro cessors Gops order order Gops

Earth Simulator

ASCI Q AlphaServer EV GHz

ASCI WhitePacic IBM SP Power MHz

Compaq AlphaServer SC ESEV GHz

Linux NetworXQuadrics GHz Xeon wQuad

Compaq AlphaServer SC ESEVGHz wQuad

HPTi Intel Xeon Ghzdual wMyrinet

IBM p cluster Power GHz

IBM SP Power no des MHz

Compaq Alphaserver SC ESEVGHz wQuad

IBM eServer pSeries Turb o GHz POW

IBM SP Power no des MHz

ASCI Red Intel Pentium I I Xeon core MHz

IBM p cluster Power GHz

Atipa Tech Pentium GHz wMyrinet

ASCI BluePacic SST IBM SP E MHz

ASCI Red Intel Pentium I I Xeon core MHz

Compaq Alphaserver SC ESEV GHz

PowerEdge HPC Cluster GHz Xeon wGnet

IBM p cluster Power GHz

IBM SP no des MHz POWER High

HITACHI SRMPPMHz

IBM eServer pSeries Turb o GHz POW

HITACHI SRFMHz

ASCI Red Intel Pentium I I Xeon core Mhz

SGI ASCI Blue Mountain

IBM SP no des MHz POWER Thin

IBM eServer pSeries Turb o GHz POW

Intel ASCI Option Red MHz Pentium Pro

NEC SXMns

CRAY TE MHz

IBM eServer pSeries Turb o GHz POW

HITACHI SRFMHz

Intel ASCI Option Red MHz Pentium Pro

Legend DeepComp GHz Pent wMyrinet

Linux NetworXQuadrics GHz Xeon wMyri

NEC SXM

Linux NetworXQuadrics GHz Xeon wMyri

Selfmade Intel Pentium GHz wSCID

SP Power MHz Nighthawk

HITACHI SRFMHz

CRAY TEE MHz

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

IBM p cluster Power GHz

Fujitsu VPP nsec

HITACHI SRMHz

SGI Origin RA MHz

HP XC Itanium GHz wQuadrics

IBM eServer pSeries GHz Power

CRAY TE MHz

Compaq AlphaServer SC ESGhz

HITACHI SRGMHz

Compaq AlphaServer SC ESGHz

IBM eServer pSeries Turb o GHz POW

SelfMade MVSM EV MHz Myrinet

Fujitsu VPP nsec

IBM SP no des MHz

Presto I I I Athlon MP Ghz Myrinet

Compaq AlphaServer SC ESGHz

IBM eServer pSeries Turb o GHz POW

HITACHI SREMHz

NCSA Titan ClusterItanium MHz wMyrinet

Compaq AlphaServer SC ESGHz

LANL Space Simulator PGHzMbsw

NEC Magi Cluster PI I I MHz wMyrinet

Pentium GHzGHzGHz

IBM eServer pSeries Turb o GHz POW

Dell WindowsPentium GHz wgnet

RWC MHz dual Pent I I I wMyrinet

IBM SP no des MHz POWER

HITACHI SRFMHz

Linux Cluster UIUCNCSA GHz Pentium I I I

IBM eServer pSeries Turb o GHz POW

ASCI Red Intel Pentium I I Xeon core Mhz

HITACHI SRFMHz

Compaq Alpha AMHzdual wMyrinet

Fujitsu VPP nsec

IBM eServer pSeries Turb o GHz POW

IBM SP no des MHz POWER

HITACHI SREMHz

IBM pSeries Turb ox GHz wGigenet

RWC SCore Cluster I I I Pentium I I I MHz

IBM SP no des MHz e

IBM SP no des MHz POWER High

SGI Origin MHz

IBM SP no des MHz POWER Thin

ASCI Red Intel Pentium I I Xeon core Mhz

CPlantRossAlpha EV MHz wMyrinet

Compaq AlphaServer SC ESGhz

Compaq ESEV AlphaServer SC

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

NEC SXM

Fujitsu VPP nsec

IBM eServer pSeries GHz Power

Fujitsu VPP nsec

AMD Athlon MP clusterGHzwFenet

IBM SP no des MHz POWER High

HP Sup erdome MHz Hyp erPlex

IBM ASCI Option Blue Pacic MHz

SGI Origin RA MHz

HITACHI SRMHz

HP Sup erdome MHz bT

CRAY TE MHz

CRAY TEE MHz

IBM SP no des MHz POWER

CRAY TE MHz

NEC SXM

Sun HPC Cluster MHzMB L

CRAY TE MHz

SGI Origin MHz

HITACHI SRGMHz

SGI Origin MHz

IBM eServer pSeries Turb o GHz POW

NEC SXM

CPPACS MHz PARISC based CPU

NEC SXM nsec

Intel Pentium I I I GHz GHz

Fujitsu VPP nsec

IBM SP MHz Power no des

Compaq AlphaServer SC ES

PowerEdge PGHz PGHz wgnet

CRAY TE MHz

Compaq AlphaServer SC ESEV MHz

Athlon MP Ghz wMyrinet

PowerEdge HPC Cluster GHz Xeon wGnet

Compaq AlphaServerSC ES MHz Quadrics

XeniaIBM IntellistationXeon GHz wMyrinet

CRAY TE MHz

Fujitsu VPPE nsec

SGI Origin MHz CPU

HITACHI SRFMHz

IBM SP no des MHz e

NEC SXM

CPlantSib eriaAlpha EV MHz wMyrinet

Dell PowerEdge HPCP GHz Xeon wMyrinet

IBM SP no des MHz POWER

IBM pSeries Turb ox GHz wGigenet

SGI Origin MHz CPU

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

SGI Origin MHz

SGI Origin MHz CPU

Fujitsu VPP nsec

IBM SP no des MHz POWER

COMPASECCO Pentium I I I GHZ wMyrinet

Compaq AlphaServer SC ESGHz ev

Intel Paragon XPS MP MHz OSSUNMOS

IBM SP no des MHz POWER High

Dell PowerEdge HPC Cluster GHz Xeon wgenet

Sun HPC Cluster MHzMB L

Compaq Alphaserver SCMhz wQuadrics

Fujitsu VPPE nsec

Compaq ESEV AlphaServer SC

hp server rp MHz Hyp erPlex

IBM eServer pSeries GHz Power

Intel Paragon XPS MP MHz OSSUNMOS

IBM SP no des MHz POWER Thin

HITACHI SRMHz

NEC SXM

hp server rp MHz bT

HP Sup erdome MHz Hyp erPlex

NEC SXM nsec

HP Sup erdome MHz bT

SGI Origin RA MHz

SGI Origin MHz CPU

LosLob os Sup erclusterPI I I MHz wMyrinet

CRAY TE MHz

HELIX AMD GHz wgnet

HITACHI SRMHz

Numerical Wind Tunnel ns

HITACHI SRMHz

Intel Paragon XPS MP MHz OSSUNMOS

CRAY TE MHz

Korean Inst STPentium GHz wMyrinet

CLiC Pentium I I I MHZ

Titech Grid Cluster Pentium I I IS Ghz

Fujitsu VPPnsec

CRAY TEE MHz

Compaq SC MHz

SGI Origin MHz

IBM SP no des MHz e

HITACHI SRFMHz

Fujitsu VPPnsec

Self MadeP GHz P GHz wGenet

IBM ASCI Option Blue Pacic MHz

SGI Origin ClusterxRA MHz

HITACHI SRGMHz

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

SelfMade Intel Pentium XeonGHz wGigE

HPTi ACL Alpha MHz wMyrinet

SGI Origin MHz

IBM eServer pSeries Turb o GHz POW

Numerical Wind Tunnel ns

Intel Paragon XPS MP MHz OSSUNMOS

NEC SXM

Netnity XseriesX PI I I GHz

IBM x Cluster PI I I GHz wMb enet

Numerical Wind Tunnel ns

Compaq AlphaServerSC ESEV MHz

Sun Fire K MHzMB E

SGI Itanium MHz

Fujitsu VPPnsec

Intel Pentium I I I GHz w Mb enet

Sun Fire K MHzMB E

IBM Ss MHz SP switch

Compaq AlphaServer SC ESEV MHz

Self MadeP GHz P GHz wGenet

Origin MHz Clusterx

IBM eServer pSeries GHz Power

SGI Origin MHz CPU

IBM eServer pSeries xwGenet GHz

CRAY TE MHz

Intel P clusterGHzGHz wGenet

HITACHI SRFMHz

Pentium GHz wGiganet

IBM Ss MHz SP switch

Sun Fire K MHzMB E

IBM SP no des MHz POWER

Compaq AS SC MHz EV Quadrics sw

SGI Origin MHz CPU

IBM SP MHz

Intel Paragon XPS MP MHz OSSUNMOS

SGI Origin MHz x cpu

HITACHI SRGMHz

IBM SP no des MHz POWER

Sun Blade MHz Cluster wMyrinet

Compaq ESEV AlphaServer SC MHz

Fujitsu VPP nsec

Compaq AlphaServerSC ES MHz Quadrics

SGI Origin MHz

IBM Ss MHz SP switch

Sun Fire K MHzMB E

HITACHI SRMHz

Intel Paragon XPS MHz OSSUNMOS

SGI Origin MHz

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

HITACHI SREMHz

SGI Cluster Dual Pentium I I I GHz

IBM SP no des MHz POWER High

IBM Ss MHz SP switch

IBM eServer pSeries Turb o GHz

Compaq ESEV AlphaServer SC

Fujitsu VPPnsec

HP Sup erdome MHz

IBM SP no des MHz POWER Thin

hp server rp MHz Hyp erPlex

hp server rp MHz bT

Sun Fire K MHzMB E

Dell PowerEdge HPCDual Pentium I I I GHz

Fujitsu VPP nsec

Paragon XPS MP No des OSSUNMOS S

NEC SX nsec

NEC SXM nsec

SGI Origin MHz CPU

IBM Ss MHz SP switch

Sun HPC Cluster MHz MB L

HITACHI SRGMHz

Origin MHz Cluster

NEC SXM ns

Compaq AlphaServer SC EV MHz

Dell PowerEdge Cluster WKDual PI I IGHzGnet

Sun Fire K MHzMB E

Linux cluster PI I I GHz w Mbs enet

Fujitsu PRIMEPOWERMHz

CRAY TE MHz

HITACHI SRMHz

Fujitsu PRIMERGY CLJ Pentium GHz

Cray TE MHz

IBM Ss MHz SP switch

CRAY TE MHz

IBM SP MHz PSC

SGI Cluster Pentium I I I GHz

Fujitsu VPP nsec

Sun UltraSPARC I I MHz ER pro cno de

Fujitsu VPP nsec

SGI Origin MHz Clusterx

Dell PowerEdge HPC Dual Pentium I I I GHz

IBM SP no des MHz e

SGI Origin MHz

Sun Fire K MHzMB E

CRAY TEE MHz

Sun Fire K MHzMB L p erib

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

IBM Ss MHz SP switch

IBM SP MHz POWER

Fujitsu PRIMEPOWERMHz

NEC TXi ItaniumGHz

SGI Origin MHz

HITACHI SRGMHz

Compaq AlphaServer SC EV MHz

SGI Origin MHz CPU

Pentium GHz Giganet Fenet

Cray TD MHz

Sun Ultra HPC Cluster MHzMB L

IBM SP MHz POWER

SGI Origin MHz Cluster

xxxx

Kepler PI I I MHz PI I I MHz

IBM SP MHz POWER

IBM eServer pSeries Turb o GHz POW

Sun Fire K MHzMB E

Fujitsu VPP nsec

SGI Origin MHz wMyrinet

IBM Ss MHz SP switch

SGI Origin MHz Clusterx

NEC SXM ns

Sun Fire K MHzMB L p erib

Sun Ultra HPC Cluster MHzMB L

IBM eServer pSeries Turb o MHz

CRAY TEE

Fujitsu VPP nsec

IBM SPT MHz

Sun Ultra HPC Cluster MHzMB L

Compaq Alphaserver GS Mhz MB L

Presto I I I Athlon ClusterGHz Myrinet

HewlettPackard Sup erDome MHz

SGI Origin MHz CPU

CRAY TE MHz

SGI Origin MHz CPU

HP evectra Pentium I I I MHz

Origin MHz Clusterx

Sun Fire K MHzMB L p erib

IBM Ss MHz SP switch

IBM SP no des MHz POWER

Sun Fire K MHzMB E

HITACHI SRFMHz

Compaq AlphaserverSC mhz

CRAY TE MHz

Sun HPC Cluster MHzMB L

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

Hitachi S cluster x ns

IBM SP no des MHz POWER

FujitsuSiemens hp cLinePentium I I I MHz

Presto I I I Athlon ClusterGHz Fenet

Sun Ultra HPC Cluster MHzMB L

Fujitsu VPP nsec

Intel Paragon XPS MHz

SGI Origin MHz Cluster

IBM Ss MHz SP switch

IBM SP no des MHz POWER High

Sun Fire K MHzMB L p erib

IBM SP no des MHz POWER Thin

Fujitsu VPP nsec

Sun Fire K MHzMB E

Sun HPC Cluster MHz MB L

NEC SX ns

RWC SCore Cluster I IDual PI I I MHzMyrinet

Origin MHz CPU

Paragon XPS MP No des OSSUNMOS S

Compaq ESEV AlphaServer SC

Compaq Alphaserver GS Mhz MB L

NEC SX nsec

HP Kayak Intel Cluster NT MHz PI I I

SGI Origin Mhz

Cray T ns

NEC SX ns

NEC SXM ns

Alphleet Alpha clusterMyranet MHz

Thinking Machines CM

Hitachi S cluster x ns

IBM Ss MHz SP switch

Sun Fire K MHzMB E

HITACHI SRMHz

Sun Fire K MHzMB L p erib

HITACHI SRMHz

Fujitsu VPPE nsec

IBM SP MHz

Cray TE MHz

Sun Ultra HPC Cluster MHzMB L

CRAY TE MHz

Sun HPC ClusterMHzMB L cache

SGIOrigin EnetClusterx MHz

IBM SP no des MHz e

Hitachi S cluster x ns

SGI Origin MHz

CRAY TEE MHz

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

Presto I I PC clusterPI I I MHz wfast enet

IBM Ss MHz SP switch

Sun Fire K MHzMB E

Sun Fire K MHzMB L p erib

SGI Origin MHz

Cray T ns

Sun Ultra HPC Cluster MHzMB L

Cray TD MHz

HITACHI SRGMHz

Sun Ultra HPC Cluster MHzMB L

LANL Avalon ClusterAlpha MhzMbs sw

SGIOrigin Clusterx MHz

Cray SVex MHz

SGI Origin x MHz wfast enet

Intel Cluster PI I I MHz quad wGiganetNT

HewlettPackard V MHz

Compaq Alphaserver GS Mhz MB L

HewlettPackard Sup erDome MHz

Sun Fire K MHzMB L p erib

Sun Ultra HPC Cluster MHzMB L

Cray SVex MHz

Fujitsu VPP nsec

Sun Fire K MHzMB E

SGI Origin Mhz

Fujitsu VPP nsec

Sun HPC MHz MB L Cache

Cray SVex MHz

IBM SPT MHz

HITACHI SRMHz

Sun HPC MHz MB L Cache

SGI Origin MHz CPU

Cray SVex MHz

IBM SP no des MHz POWER

Sun E WildFire servers MHz

SGI Origin MHz CPU

CRAY TE MHz

FujitsuSiemens hp cLinePentium I I MHz

Sun HPC MHz MB L Cache

Sun Fire K MHzMB L p erib

Hitachi S cluster x ns

HITACHI SRFMHz

Hitachi S cluster x ns

Compaq ESEV AlphaServer SC

SGI Origin MHz

IBMSPnodesMHzPOWER

Sun Fire K MHzMB E

SGI Origin x MHz wfast enet

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

Cray SVex MHz

Sun HPC MHz MB L Cache

Sun Ultra HPC Cluster MHzMB L

Cray SVex MHz

IBM Ss MHz SP switch

Sun E WildFire servers MHz

Sun Ultra HPC Cluster MHzMB L

SGI Origin Ether Cluster MHzx

Fujitsu VPP nsec

Sun Ultra HPC Cluster MHzMB L

Compaq GS cluster

Cray T ns

Fujitsu VPPE nsec

IBM SP no des MHz POWER High

HewlettPackard V MHz

Sun HPC MHz MB L Cache

Sun Fire MHzMB L

HITACHI SREMHz

Sun Fire K MHzMB L p erib

IBM SP no des MHz POWER Thin

NEC SX ns

Parnass Cluster PI I MHz wMyricon

Sun HPC MHz MB L Cache

Fujitsu VPP nsec

Sun HPC MHz MB L Cache

Compaq GS cluster

SGI Origin x MHz wfast enet

Compaq Alphaserver GS Mhz MB L

Sun Fire K MHzMB E

Sun HPC MHz MB L Cache

NEC ExpressXa MHz

SGI Origin Mhz

Sun HPC MHz MB L Cache

Cray SVex MHz

SGI Origin x MHz wfast enet

Paragon XPS MP No des OSSUNMOS S

HP V pro c MHz

SGI Origin Ether Cluster MHzx

NEC SX ns

NEC SXM ns

Sun HPC Cluster MHz MB L

DEC AlphaServer MHz

NEC SXA ns

Cray SV MHz

SGI Origin Ether Cluster MHzx

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

Thinking Machines CM

Sun HPC MHz MB L Cache

Cray SV MHz

Sun Fire MHzMB L

Hitachi SRMHz

IBM SP MHz

Sun Fire K MHzMB L p erib

HITACHI SRMHz

IBM SPT MHz

Compaq GS cluster

Cray TE MHz

IBM eServer pSeries MGHz POWER

Hitachi S ns

Sun HPC MHz MB L Cache

CRAY TE MHz

IBM SP no des MHz e

Cray SV MHz

Hitachi S cluster x ns

HP Exemplar XClass SPPUX

Sun HPC MHz MB L Cache

IBM Ss MHz SP switch

Hitachi S cluster x ns

CRAY SV MHz

SGI POWER CHALLENGE MHz

CRAY TEE MHz

Sun Fire K MHzMB E

Sun Ultra HPC MHz MB L Cache

Sun HPC MHz MB L Cache

SGI Origin MHz

HITACHI SRGMHz

Sun HPC Cluster MHz MB L

DEC AlphaServer MHz

Cray TD MHz

Compaq GS cluster

CRAY SV MHz

Sun Ultra HPC MHz MB L Cache

Sun HPC MHz MB L Cache

Sun Fire MHzMB L

DEC MHz

HP V pro c MHz

SGI Origin Ether Cluster MHzx

Sun HPC MHz MB L Cache

Cray SVex MHz

Sun Fire MHzMB L

Cray SV MHz

Fujitsu VPP nsec

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

SGI Origin MHz

Sun Fire K MHzMB L p erib

IBM SP thinno deSPswMBno de MHz

Sun Ultra HPC MHz MB L Cache

NEC SXR ns

IBM SPT MHz

Sun HPC MHz MB L Cache

IBM POWER Sup er Chip RS SP MHz

IBM eServer pSeries GHz POWER

Sun HPC MHz MB L Cache

DEC MHz

Sun Ultra HPC MHz MB L Cache

Sun HPC MHz MB L Cache

Hitachi S ns

Hitachi S cluster x ns

Sun Ultra HPC MHz MB L Cache

SGI Origin Ether Cluster MHzx

Sun Ultra HPC MHz MB L Cache

IBM SP no des MHz POWER

CRAY TE MHz

SGI Origin MHz

Cray C MHz

DEC AlphaServer MHz

HITACHI SRFMHz

HewlettPackard V MHz

Sun Ultra HPC MHz MB L Cache

Compaq GS cluster

Sun HPC MHz MB L Cache

Cray SV MHz

Sun HPC MHz MB L Cache

NEC SX ns

IBMSPnodesMHzPOWER

Sun Fire MHzMB L

Sun HPC Cluster MHz MB L

LANL Avalon ClusterAlpha MhzMbs sw

Cray SVex MHz

DEC MHz

Sun HPC MHz MB L Cache

Sun Ultra HPC MHz MB L Cache

Fujitsu VPP nsec

Sun HPC MHz MB L Cache

Sun Ultra HPC MHz MB L Cache

Fujitsu VPPE nsec

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

SGI POWER CHALLENGE MHz

IBM SP no de MHz POWER High

Sun Fire MHzMB L

DEC AlphaServer MHz

Sun HPC MHz MB L Cache

IBM SP no des MHz POWER Thin

HP V pro c MHz

SGI Origin Ether Cluster MHzx

NEC SXR ns

Sun HPC MHz MB L Cache

Sun Ultra HPC MHz MB L Cache

Fujitsu VPP nsec

Sun HPC MHz MB L Cache

Sun Ultra Sparc Cluster

DEC MHz

Sun Ultra HPC MHz MB L Cache

Sun HPC MHz MB L Cache

Cray SV MHz

Paragon XPS MP No des OSSUNMOS S

Cray C MHz

Sun HPC Cluster MHz MB L

Sun HPC MHz MB L Cache

SGI Origin Mhz

Sun HPC MHz MB L Cache

SGI POWER CHALLENGE MHz

Sun HPC MHz MB L Cache

NEC SXM ns

NEC SX ns

Sun Ultra HPC MHz MB L Cache

Compaq GS cluster

NEC SXA ns

Intel Paragon XPS MHz OSR

IBM S MHz

hp server rx MHz MB L Cache

Thinking Machines CM

HP Exemplar XClass SPPUX

Sun Fire MHzMB L

IBM SP MHz

Sun HPC MHz MB L Cache

IBM eServer pSeries MGHz POWER

HITACHI SRMHz

Hitachi S ns

HITACHI SRMHz

Sun HPC MHz MB L Cache

IBM SP MHz switch of

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

Hitachi S cluster x ns

Sun HPC MHz MB L Cache

Cray TE MHz

Sun Ultra HPC MHz MB L Cache

CRAY TE MHz

Intel Delta MHz

DEC AlphaServer MHz

Sun Ultra HPC MHz MB L Cache

Sun HPC MHz MB L Cache

Cray YMP C MHz ns

DEC MHz

hp server rx MHz MB L Cache

IBM eServer pSeries GHz POWER

CRAY TEE MHz

Sun Ultra HPC MHz MB L Cache

Sun HPC MHz MB L Cache

SGI Origin MHz

Sun HPC MHz MB L Cache

HITACHI SRGMHz

HITACHI SRMHz

Cray SVex MHz

Sun HPC Cluster MHz MB L

Cray TD MHz

Sun Ultra HPC MHz MB L Cache

IBM SP thinno deSPswMBno de MHz

Intel Paragon XPS MHz

Sun Ultra HPC MHz MB L Cache

DEC MHz

HewlettPackard N MHz

Sun HPC MHz MB L Cache

Cray SV MHz

Sun HPC MHz MB L Cache

Sun Ultra HPC MHz MB L Cache

Fujitsu VPP nsec

Sun Fire MHzMB L

Sun Ultra HPC MHz MB L Cache

NEC SXR ns

Sun HPC MHz MB L Cache

IBM SPT MHz

IBM POWER Sup er Chip RS SP MHz

Sun Ultra HPC MHz MB L Cache

Sun HPC MHz MB L Cache

Sun Ultra HPC MHz MB L Cache

Sun HPC MHz MB L Cache

IBM SP no des MHz POWER

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

Compaq GS cluster

Sun HPC MHz MB L Cache

Cray C MHz

SGI Origin MHz MB L Cache

DEC MHz

HITACHI SRFMHz

Sun Ultra HPC MHz MB L Cache

HP Exemplar VClass MHz

HewlettPackard V MHz

CRAY TE MHz

Sun Ultra HPC MHz MB L Cache

IBM SP no des MHz e

HP N MHz

Intel Delta MHz

Sun HPC MHz MB L Cache

Berkeley NOWUltraSPARCMhzMyricom

Sun HPC MHz MB L Cache

IBM SP no des MHz POWER

NEC SX ns

Sun Ultra HPC MHz MB L Cache

Sun Fire MHzMB L

Thinking Machines CM MHz

Sun HPC MHz MB L Cache

DEC AlphaServer MHz

Sun Ultra HPC MHz MB L Cache

DEC MHz

Fujitsu VPP nsec

Cray T ns

SGI POWER CHALLENGE MHz

Fujitsu VPPE nsec

Fujitsu VXE nsec

HP V pro c MHz

HP Exemplar VClass MHz

Sun HPC MHz MB L Cache

IBM SP no des MHz POWER Thin

HITACHI SREMHz

Sun Ultra HPC MHz MB L Cache

SGI Origin Mhz

Sun HPC MHz MB L Cache

NEC SXR ns

Sun Ultra HPC MHz MB L Cache

Fujitsu VPP nsec

Fujitsu VX nsec

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

Sun HPC MHz MB L Cache

Cray C MHz

SGI POWER CHALLENGE MHz MB cache

HP Exemplar VClass MHz

Cray SV MHz

Sun Ultra HPC MHz MB L Cache

Parsytec GCPower Plus MHz

SGI Origin MHz MB cache

Sun Ultra HPC MHz MB L Cache

Paderb orn SCI ClusterSNIScaliMHz PI I

SGI POWER CHALLENGE MHz

HP Exemplar SClass SPPUX

DEC MHz

Thinking Machines CM

Cray J ns

SGI POWER CHALLENGE MHz MB cache

Intel Paragon XPS MHz OSR

IBM SP MHz

HITACHI SRMHz

SGI POWER CHALLENGE MHz

Cray J ns

Hitachi S ns

IBM SP MHz switch of

IBM eServer pSeries MGHz POWER

Sun HPC MHz MB L Cache

Sun Ultra HPC MHz MB L Cache

Cray TE MHz

CRAY TE MHz

DEC MHz

Cray T ns

HP Exemplar VClass MHz

Sun Ultra HPC MHz MB L Cache

Intel Delta MHz

SGI POWER CHALLENGE MHz MB cache

DEC MHz

Sun Ultra HPC MHz MB L Cache

DEC Alphaserver MHz MB cache

CRAY TEE MHz

Cray J ns

IBM SP thinno deSPswMBno de MHz

HewlettPackard N MHz

Cray SVex MHz

Compaq Alphaserver ES Mhz MB L

Cray TD MHz

Sun Ultra HPC MHz MB L Cache

Convex SPP pro cs MHz

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

SGI POWER CHALLENGE MHz MB cache

Fujitsu VPP nsec

Sun Fire MHzMB L

HP Exemplar SClass SPPUX

Cray J ns

DEC pro c MHz

SGI POWER CHALLENGE MHz MB cache

Sun Ultra HPC MHz MB L Cache

DEC Alphaserver MHz MB cache

SGI POWER CHALLENGE MHz MB cache

Sun HPC MHz MB L Cache

IBM SPT MHz

NEC SXR ns

Sun HPC MHz MB L Cache

IBM POWER Sup er Chip RS SP MHz

Cray C MHz

HP Exemplar VClass MHz

HewlettPackard V MHz

IBM SP no de MHz POWER

Convex SPP pro cs MHz

SGI POWER CHALLENGE MHz MB cache

HP N MHz

IBM SP no des MHz e

SGI Origin MHz MB cache

CRAY TE MHz

Intel Delta MHz

Parsytec GCPower Plus MHz

Thinking Machines CM MHz

IBM SP no des MHz POWER

Compaq ESEV AlphaServer SC MHz

DEC AlphaServer

MeikoCS

NEC SX ns

Thinking Machines CM MHz

SGI POWER CHALLENGE MHz MB cache

Sun Fire MHzMB L

Cray J ns

SGI POWER CHALLENGE MHz

Cray T ns

Sun HPC MHz MB L Cache

SGI POWER CHALLENGE MHz MB cache

HP Exemplar VClass MHz

Alliant CAMPUS MHz

IBM SP

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

DEC Alphaserver MHz MB cache

HP V pro c MHz

IBM RS P pro c MHz MB L

IBM RS P MHz MB L

IBM eServer pSeries MHz MB L

SGI POWER CHALLENGE MHz

IBM SP no de MHz POWER Thin

Compaq Digital AlphaServer MHz

IBM RS P MHz MB L

IBM eServer pSeries MHz MB L

IBM RS P pro c MHz MB L

IBM RS B pro c MHz MB L

SGI POWER CHALLENGE MHz MB cache

Compaq Digital AlphaServer MHz

NEC SXR ns

Sun HPC MHz MB L Cache

SGI POWER CHALLENGE MHz

Cray C MHz

Sun Ultra HPC MHz MB L Cache

SGI POWER CHALLENGE MHz

Compaq ESEV AlphaServer SC MHz

Cray SV MHz

HP Exemplar SClass SPPUX

Alliant CAMPUS MHz

SGI POWER CHALLENGE MHz MB cache

SGI Origin MHz MB cache

DEC MHz

Intel Paragon XPS MHz OSR

Convex SPP pro cs MHz

DEC AlphaServer

Parsytec GCPower Plus MHz

IBM SP MHz

Thinking Machines CM

Cray J ns

SGI POWER CHALLENGE MHz

HITACHI SRMHz

IBM SP MHz switch of

SGI POWER CHALLENGE MHz

IBM eServer pSeries MGHz POWER

Sun HPC MHz MB L Cache

SGI POWER CHALLENGE MHz MB cache

Sun Ultra HPC MHz MB L Cache

DEC MHz

CRAY TE MHz

Intel Delta MHz

Alliant CAMPUS MHz

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

SGI POWER CHALLENGE MHz MB cache

IBM eServer pSeries GHz POWER

IBM SP thinno deSPswMBno de MHz

SGI POWER CHALLENGE MHz

HewlettPackard N MHz

CRAY TEE MHz

Sun Ultra HPC MHz MB L Cache

Convex SPP pro cs MHz

Cray SVex MHz

Intel Pentium GHz

Cray TD MHz

DEC AlphaServer

SGI POWER CHALLENGE MHz

Sun Ultra MHzMB L

IBM SPT MHz

IBM eServer pSeries GHz POWER

Cray C MHz

HP Exemplar VClass MHz

Alliant CAMPUS MHz

NEC SXR ns

Sun HPC MHz MB L Cache

Sun HPC MHz

IBM POWER Sup er Chip RS SP MHz

SGI POWER CHALLENGE MHz

Convex SPP pro cs MHz

SGI POWER CHALLENGE MHz

MeikoCS

Parsytec GCPower Plus MHz

Convex SPP pro cs MHz

HP N MHz

Sun Ultra HPC MHz MB L Cache

SGI Origin MHz MB cache

CRAY TE MHz

Intel iPSC MHz

Cray J ns

SGI POWER CHALLENGE MHz MB cache

NEC SX ns

HP Exemplar VClass MHz

Sun Fire MHzMB L

Sun Ultra HPC MHz MB L Cache

Cray T ns

Sun HPC MHz MB L Cache

DEC AlphaServer

IBM SP

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

Thinking Machines CM MHz

SGI POWER CHALLENGE MHz

IBM RS P pro c MHz MB L

IBM RS P MHz MB L

IBM eServer pSeries MHz MB L

SGI POWER CHALLENGE MHz

Alliant CAMPUS MHz

Intel iPSC MHz

Fujitsu AP

IBM RS P pro c MHz MB L

IBM RS B pro c MHz MB L

IBM RS P MHz MB L

IBM eServer pSeries MHz MB L

Sun Ultra HPC MHz MB L Cache

HP Exemplar SClass SPPUX

Sun Ultra HPC MHz

Cray SV MHz

Convex SPP pro cs MHz

DEC MHz

Intel Paragon XPS MHz OSR

SGI POWER CHALLENGE MHz

Thinking Machines CM

Intel iPSC MHz

nCUBE MHz

CRAY TE MHz

IBM SP MHz switch of

Sun Ultra HPC MHz MB L Cache

IBM SP thinno deSPswMBno de MHz

Intel Delta MHz

SGI POWER CHALLENGE MHz

CRAY TEE MHz

Cray SVex MHz

Alliant CAMPUS MHz

MasPar MP ns

Sun Ultra HPC MHz MB L Cache

Sun Ultra MHzMB L

DEC MHz

Sun Fire MHzMB L

IBM SPT MHz

MeikoCS

NEC SXR ns

Parsytec GCPower Plus MHz

Convex SPP pro cs MHz

Sun HPC MHz

SGI POWER CHALLENGE MHz

Intel iPSC MHz

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

SGI Origin MHz MB cache

CRAY TE MHz

SGI POWER CHALLENGE MHz MB cache

Meiko Computing Surface MHz

NEC SX ns

SGI CHALLENGE ns

Sun Fire MHzMB L

SGI CHALLENGE ns

DEC AlphaServer

Fujitsu AP

IBM SP

Thinking Machines CM MHz

SGI CHALLENGE ns

SGI POWER CHALLENGE MHz

Alliant CAMPUS MHz

Sun Ultra HPC MHz

SGI POWER CHALLENGE MHz

Cray SV MHz

Convex SPP pro cs MHz

SGI CHALLENGEOnyx ns

Sun HPC MHz

Convex SPP pro cs MHz

SGI POWER CHALLENGE MHz

Intel iPSC MHz

Thinking Machines CM

nCUBE MHz

HITACHI SRMHz

IBM PVS MHz

Intel Delta MHz

CRAY TE MHz

SGI CHALLENGEOnyx ns

Meiko Computing Surface MHz

Meiko CS

SGI CHALLENGEOnyx ns

Sun Ultra MHzMB L

NEC SXLR ns

Sun HPC MHz

SGI CHALLENGEOnyx ns

SGI Origin MHz MB cache

IBM RS Cluster PARC MHz

Parsytec GCPower Plus MHz

NEC SXL ns

Intel iPSC MHz

SGI POWER CHALLENGE MHz MB cache

SGI CHALLENGEOnyx ns

SGI POWER CHALLENGE MHz

Fujitsu AP

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

SGI CHALLENGEOnyx ns

IBM RS Cluster PARC MHz

Sun Ultra HPC MHz

Alliant CAMPUS MHz

SGI POWER CHALLENGE MHz

Sun HPC MHz

Intel iPSC MHz

nCUBE MHz

MasPar MP ns

SGI CHALLENGEOnyx ns

Intel Delta MHz

Meiko Computing Surface MHz

MasPar MP ns

IBM RS Cluster PARC MHz

ALR Revolution Quad Pentium MHz

MasPar MP ns

IBM RS Cluster PARC MHz

Intel iPSC MHz

SGI Origin MHz MB cache

SGI POWER CHALLENGE MHz MB cache

SGI POWER CHALLENGE MHz

IBM RS Cluster PARC MHz

Fujitsu AP

SGI POWER CHALLENGE MHz

nCUBE MHz

HITACHI SRMHz

Meiko Computing Surface MHz

Parsytec FT MHz

Intel Delta MHz

Intel iPSC MHz

Meiko Computing Surface MHz

nCUBE MHz

Intel Delta MHz

MasPar MP ns

Intel iPSC MHz

IBM RS MHz

MasPar MP ns

Meiko Computing Surface MHz

Thinking Machines CM

nCUBE MHz

Intel Delta MHz

Intel iPSC MHz

nCUBE MHz

Meiko Computing Surface MHz

MasPar MP ns

Intel iPSC MHz

nCUBE MHz

November

Number of R N N R Computer

max max peak

Full Precision Pro cessors Gops order order Gops

nCUBE MHz

Thinking Machines CM half precision

IBM GF half precision ns

Fujitsu AP half precision

The Numerical Wind Tunnel is not a commercial pro duct it is a computer of the National Aerospace

Lab oratory in Japan and is based on the Fujitsu vector pro cessor b oard

The CPPACS Computational Physics byParallel Array Computer System is not a commercial pro duct

it is a computer of the University of Tsukuba Japan Hitachi mo died several points in their SR

computer The pro cessor manufactured by Hitachi is a custom sup erscalar pro cessor It is based on the

PARISC Architecture enhanced with a PVPSW pseudo vector pro cessor based on slide window registers

scheme

The IBM GF is an exp erimental research computer and not a commercial pro duct

Indicates Strassen Algorithm was used in computing the solution Note the achieved rate is large

than the p eak rate for the computer The rate of execution for this problem is based on the the number

of oating p oint op erations divided bythetime to solve the problem The oating p oint op eration count

n O n is based on a conventional Gaussian Elimination implementation Strassens Algorithm

reduced the numb er of op erations actually p erformed The results obtained using Strassen Algorithm are as

accurate as that from Gaussian Elimination

The Earth Simulator is not a commercial pro duct it is a computer of the Earth Simulator Center

the arm of the Japan Marine Science and Technology Center It is based on vector pro cessors that are

manufactured byNEC

The columns in Table are dened as follows

R the p erformance in Gops for the largest problem run on a machine

max

N the size of the largest problem run on a machine

max

N the size where half the R execution rate is achieved

max

R the theoretical p eak p erformance in Gops for the machine

peak

In addition the number of pro cessors and the cycle time is listed Full or half precision reects the

computation was computed using or bit oating p oint arithmetic resp ectively

Acknowledgments

I am indebted to the many p eople who have help ed put together this collection

References

J Dongarra J Bunch C Moler and G W Stewart LINPACK Users Guide SIAM Philadelphia

J J Dongarra I S Du D C Sorensen and H A Van der Vorst Solving Linear Systems on Vector

and Shared Memory Computers SIAM Publications Philadelphia PA

C Lawson R Hanson D Kincaid and F Krogh Basic linear algebra subprograms for Fortran usage

ACM Trans Math Softw