-
1 - 53
NAS Parallel Benchmark (Version 1.0) Results 11-96Subhash Saini1
and David H. Bailey2
Report NAS-96-18, November 1996
Numerical Aerospace Simulation ProgramNASA Ames Research
Center
Mail Stop T 27A-1Moffett Field, CA 94035-1000, USA
email: [email protected]
Abstract
The NAS Parallel Benchmarks have been developed at NASA Ames
Research Center to studythe performance of parallel supercomputers.
The eight benchmark problems are specified in a‘‘pencil and paper’’
f ashion. In other words, the complete details of the problem to be
solved aregiven in a technical document, and except for a few
restrictions, benchmarkers are free to selectthe language
constructs and implementation techniques best suited for a
particular system. Theseresults represent the best results that
have been reported to us by the vendors for the specificsystems
listed. In this report3, we present new NPB (Version 1.0)
performance results for thefollowing systems:
- DEC Alpha Server 8400 5/440,- Fujitsu VPP Series (VX, VPP300,
and VPP700),- HP/Convex Exemplar SPP2000,- IBM RS/6000 SP P2SC node
(120 MHz)- NEC SX-4/32,- SGI/CRAY T3E,- SGI Origin200- SGI
Origin2000
We also report High Performance Fortran (HPF) based NPB results
for IBM SP2 Wide Nodes,HP/Convex Exemplar SPP2000, and SGI/CRAY
T3D. These results have been submitted byApplied Parallel Research
(APR) and Portland Group Inc. (PGI). We also present
sustainedperformance per dollar for Class B LU, SP and BT
benchmarks.
1. Subhash Saini is an employe of MRJ Inc. This work is
supported by NASA Ames Research Center throughcontract NAS
2-14303.
2. David H. Bailey is an employee of NASA Ames Research
Center.3. URL: http://www.nas.nasa.gov/NAS/NPB/
-
2 - 53
1: Intr oduction
The Numerical Aerodynamic Simulation (NAS) Program, located at
NASA Ames ResearchCenter, is a pathfinder in high-performance
computing for NASA and is dedicated to advancingthe science of
computational aerodynamics. One key goal of the NAS organization is
todemonstrate by the year 2000 an operational computing system
capable of simulating an entireaerospace vehicle system in one to
several hours. It is currently projected that the solution of
thisGrand Challenge problem will require a system that can perform
scientific computations at asustained rate of approximately 1000
times faster than 1990 generation supercomputers. Such acomputer
system will most likely employ hundreds or even thousands of
powerful RISCprocessors operating in parallel.
In order to objectively measure the performance of various
highly parallel computer systemsand to compare them with
conventional supercomputers, NAS has developed the NAS
ParallelBenchmarks (NPB 1.0) [1, 2]. Note that the NPB 1.0 are
distinct from the NAS High SpeedProcessor (HSP) benchmarks and
procurements. The HSP benchmarks are used for evaluatingproduction
supercomputers for procurements in the NAS organization, whereas
the NPB 1.0 areused for studying highly parallel processor (HPP)
systems in general.
2: NAS Parallel Benchmarks
The NPB 1.0 consist of a set of eight benchmark problems, each
of which focuses on someimportant aspect of highly parallel
supercomputing for aerophysics applications. Some extensionof
Fortran or C is required for implementations, and reasonable limits
are placed on the use ofassembly code and the like. Otherwise,
programmers are free to utilize language constructs thatmaximize
performance on the particular system being studied. The choice of
data structures,processor allocation, and memory usage are
generally left open to the discretion of theimplementer.
The eight problems consist of five kernels and three simulated
computational fluid dynamics(CFD) applications. The five kernels
comprise relatively compact problems, each emphasizing aparticular
type of numerical computation. Compared with the simulated CFD
applications, theycan be implemented fairly readily and provide
insight as to the general levels of performance thatcan be expected
on these specific types of numerical computations.
The simulated CFD applications, on the other hand, usually
require more effort to implement,but they are more representative
of the types of actual data movement and computation required
instate-of-the-art CFD application codes. For example, in an
isolated kernel, a certain data structuremay be very efficient on a
certain system; and yet, this data structure may be inappropriate
ifincorporated into a larger application. By comparison, the
simulated CFD applications requiredata structures and
implementation techniques that are more typical of real CFD
applications.
(Space does not permit a complete description of these benchmark
problems. A more detaileddescription of these benchmarks, together
with the rules and restrictions associated with them, isgiven in
reference 2.)
There are now three standard sizes for the NAS Parallel
Benchmarks: Class A, Class B andClass C size problems. The nominal
benchmark sizes for Class A, Class B and Class C problems
-
3 - 53
are shown in Table 1. These tables also give the standard
floating-point operation (flop) counts forClass A and Class B. We
recommend that those wishing to compute performance rates in
millionsof floating point operations per second (Mflop/s) use these
standard flop counts. Table 1 containsMflop/s rates calculated in
this manner for the current fastest implementation on one processor
ofCRAY Y-MP for Class A and on one processor of CRAY C90 for Class
B. Note, however, that inthis report, performance rates arenot
cited in Mflop/s; instead we present, the wall clock times(and, the
equivalent performance ratios). We suggest that these, not Mflop/s,
be examined whencomparing different systems and
implementations.
With the exception of the IS benchmark, these standard flop
counts were determined by usingthe hardware performance monitor on
the CRAY Y-MP or CRAY C90, and we believe that theyare close to the
minimal counts required for these problems. In the case of the IS
benchmark,which does not involve floating-point operations, we
selected a value approximately equal to thenumber of integer
operations required, in order to permit the computation of
performance ratesanalogous to Mflop/s rates. We reserve the right
to change these standard flop counts in the future,if
necessary.
The NAS organization reserves the right to verify any NPB
results that are submitted to us. Wemay, for example, attempt to
run the submitter’s code on another system of the same
configurationas that used by the submitter. In those instances
where we are unable to reproduce the vendor’ssupplied results
(allowing a 5% tolerance), our policy is to alert the submitter of
the discrepancyand allow submitter to resolve the discrepancy in
the next release of this report. If the discrepancyis not resolved
to our satisfaction, then our own observed results and not the
submitter’s resultswill be reported.
3: NAS Parallel Benchmark Results
In the following section, each of the eight benchmarks will be
briefly described, and then thebest performance results we have
received to date for each computer system will be given inTables 2
through 10 and Tables 14-18. These tables include run times and
performance ratios. Theperformance ratios compare individual
timings with the current best time for that benchmarkachieved on
one processor of CRAY Y-MP for Class A and on one processor of CRAY
C90 forClass B. The run times in each case are elapsed time
measured in accordance with thespecifications of NPB rules. This
report includes a number of new results including
previouslyunpublished - DEC Alpha Server 8400 5/440, Fujitsu VPP
Series (VX, VPP300, VPP700),HP/Convex Exemplar SPP2000, IBM RS/6000
SP P2SC node (120 MHz), NEC SX-4/32,SGI/CRAY T3E, SGI Origin200 and
SGI Origin2000
3.1: Kernels
The results for five kernels (EP, MG, CG, FT, and IS) are given
below in the following section:
3.1.1: The Embarrassingly Parallel (EP) Benchmark
The first of the five kernel benchmarks is anembarrassingly
parallel problem. In thisbenchmark, two-dimensional statistics are
accumulated from a large number of Gaussian pseudo-random numbers,
which are generated according to a particular scheme that is
well-suited forparallel computation. This problem is typical of
many Monte Carlo applications. Since it requiresalmost no
communication, in some sense this benchmark provides an estimate of
the upperachievable limits for floating-point performance on a
particular system. Results for EP benchmark
-
4 - 53
are given in Table 2.
3.1.2: Multigrid (MG) Benchmark
The second kernel benchmark is a simplified multigrid kernel,
which solves a 3-D PoissonPDE. This problem is simplified in the
sense that it has constant rather than variable coefficientsas in a
more realistic application. This code is a good test of both short
and long distance highlystructured communication. The Class B
problem uses the same size grid as of Class A but agreater number
of inner loop iterations. Results for this benchmark are shown in
Table 3.
3.1.3: Conjugate Gradient (CG) Benchmark
In this benchmark, a conjugate gradient method is used to
compute an approximation to thesmallest eigenvalue of a large,
sparse, symmetric positive definite matrix. This kernel is typical
ofunstructured grid computations in that it tests irregular
long-distance communication and employssparse matrix-vector
multiplication. Results are shown in Table 4.
3.1.4: 3-D FFT PDE (FT) Benchmark
In this benchmark a 3-D partial differential equation is solved
using FFTs. This kernel performsthe essence of many spectral
methods. It is a good test of long-distance
communicationperformance. The rules of the NPB specify that
assembly-coded, library routines may be used toperform matrix
multiplication and one-dimensional, two-dimensional, or
three-dimensional FFTs.Thus this benchmark is somewhat unique in
that computational library routines may be legallyemployed. Results
are shown in Table 5.
3.1.5: Integer Sort (IS) Benchmark
This benchmark tests a sorting operation that is important
inparticle method codes. This typeof application is similar to
particle-in-cell applications of physics, wherein particles are
assignedto cells and may drift out. The sorting operation is used
to reassign particles to the appropriatecells. This benchmark tests
both integer computation speed and communication performance.This
problem is unique in that floating point arithmetic is not
involved. Significant datacommunication, however, is required.
Results are shown in Table 6.
3.2: Simulated CFD Application Benchmarks
The three simulated CFD application benchmarks are intended to
accurately represent theprincipal computational and data movement
requirements of modern CFD applications.
3.2.1: LU Simulated CFD Application (LU) Benchmark
The first of these is the so-called the lower-upper diagonal
(LU) benchmark. It does not performa LU factorization but instead
employs a symmetric successive over-relaxation (SSOR)
numericalscheme to solve a regular-sparse, block5x5 lower and upper
triangular system. This problemrepresents the computations
associated with a newer class of implicit CFD algorithms, typified
atNASA Ames by the codeINS3D-LU. This problem exhibits a somewhat
limited amount ofparallelism compared to the next two benchmarks. A
complete solution of the LU benchmarkrequires 250 iterations.
Results are given in Table 7.
-
5 - 53
3.2.2: SP Simulated CFD Application (SP) Benchmark
The second simulated CFD application is called the scalar
pentadiagonal (SP) benchmark. Inthis benchmark, multiple
independent systems of nondiagonally dominant, scalar
pentadiagonalequations are solved. A complete solution of the SP
benchmark requires 400 iteration. Results aregiven in Table 8.
3.2.3: BT Simulated CFD Application (BT) Benchmark
The third simulated CFD application is called the block
tridiagonal (BT) benchmark. In thisbenchmark, multiple independent
systems of non-diagonally dominant, block tridiagonalequations with
a5x5 block size are solved.
SP and BT are representative of computations associated with the
implicit operators of CFDcodes such asARC3D at NASA Ames. SP and BT
are similar in many respects, but there is afundamental difference
with respect to the communication to computation ratio. Timings are
citedas complete run times, in seconds, as with the other
benchmarks. For the BT benchmark, 200iterations are required.
Results of BT benchmark are given in Table 9.
3.3 NPB 1.0 Class C Results
In Table 10 are given the Class C NPB 1.0 results for EP, MG,
CG, FT, LU, SP and BTbenchmarks on CRAY T3E.
4: Sustained Performance Per Dollar
One aspect of the relative performance of these systems has not
been addressed so far, namelythe differences in price between these
systems. One should not be too surprised that the CRAYC90 system,
for example, exhibits superior performance rates on these
benchmarks, since itscurrent list price is much greater than that
of the other systems tested.
One way to compensate for these price differences is to compute
sustained performance permillion dollars,i.e. the performance ratio
figures shown in Tables 2 through 9 divided by the listprice in
millions. Some figures of this type are shown in Tables 11-13 for
Class B LU, SP, and BTbenchmarks, respectively. The table includes
the list price of the minimal system (in terms ofmemory per node
and number of processors) required to run the full Class B size NPB
asimplemented by the vendor. These prices were provided by the
vendors and include anyassociated software costs, i.e. operating
system, compilers, scientific libraries as required,etc. butdo not
include maintenance. Note that some vendors’ standard
configurations may includesubstantially more hardware than required
for the benchmark.
5 High Performance Fortran Based NPB
High Performance Fortran (HPF), the high-level language for
parallel Fortran programming, isbased on Fortran 90. HPF was
defined by an informal standards committee known as the
HighPerformance Fortran Forum (HPFF) in 1993, and modeled on TMC’s
CM Fortran language.Several HPF features have since been
incorporated into the draft ANSI/ISO Fortran 95, the nextformal
revision of the Fortran standard.
HPF allows users to write a single parallel program that can
execute on a serial machine, ashared-memory parallel machine, or a
distributed-memory parallel machine. HPF eliminates the
-
6 - 53
complex, error-prone task of explicitly specifying how, where,
and when to pass messagesbetween processors on distributed-memory
machines, or when to synchronize processors onshared-memory
machines. HPF is designed in a way that allows the programmer to
code anapplication at a high level, and then selectively optimize
portions of the code by dropping intomessage-passing or calling
tuned library routines as “extrinsics”.
Compilers supporting High Performance Fortran features first
appeared in late 1994 and early1995 from Applied Parallel Research
(APR) Digital Equipment Corporation, and The PortlandGroup (PGI).
IBM introduced an HPF compiler for the IBM RS/6000 SP in April of
1996. Overthe past year, these implementations have shown steady
improvement in terms of both featuresand performance. NAS is
working toward a standard set of HPF NPBs which can be used as
ameasure of both HPF compilers and hardware systems. In advance of
that release we have chosento publish here results achieved using
versions of the HPF NPBs developed outside of NAS andsubmitted by
APR and PGI. These results can be compared to the MPI results in
NPB 2.2 [4] andthe other NPB 1.0 results supplied here to provide
perspective on performance currentlyobtainable using HPF versus MPI
or versus hand-tuned implementations such as those suppliedby the
hardware vendors.
For more details on HPF, see reference 3 and references therein.
HPF NPB results appear inTables 14-18.
References
[1] D. H. Bailey, E. Barszcz, J. T. Barton, D. S. Browning, R.
L. Carter, L. Dagum, R. A. Fatoohi, P. O. Frederickson, T. A.
Lasinski, R. S. Schreiber, H. D. Simon, V. Venkatakrishnan, and S.
K. Weeratunga, ‘‘The NAS Parallel Benchmarks,’’ International
Journal of Supercomputer Applications, Vol 5, No.3 (Fall1991), pp.
63-73.
[2] D. H. Bailey, J. Barton, T. Lasinski, and H. D. Simon, eds.,
“The NAS Parallel Benchmarks,’’NASA Technical Memorandum 103863,
NASA Ames Research Center, Moffett Field, CA 94035-1000, July
1993.
[3] S. Saini, “NAS Experiences of Porting CM Fortran Codes to
HPF on IBM SP2 and SGIPower Challenge”, pp. 873-880, inThe
Proceedings of IEEE 10th International ParallelProcessing Symposium
(IPPS), Honolulu, Hawaii, April 15-19, 1996.
[4] David Bailey, Tim Harris, William Saphir, Rob van der
Wijngaart, Alex Woo, and MauriceYarrow, “The NAS Parallel
Benchmarks 2.0”, Report NAS-95-020, December, 1995.
-
7 - 53
Table 1: NAS Parallel Benchmarks Problem Size.
Benchmark Name Abb.
Class A Class B Class C
NominalSize
Operation Count (x 109)
Mflop/sCRAY
Y-MP/1
Nominal Size
OperationCount(x109)
Mflop/sCRAYC90/1
NominalSize
EmbarrassinglyParallel
EP 228 26.68 211 230 100.9 689 232
Multigrid MG 2563 3.905 176 2563 18.81 557 5123
Conjugate Gradient CG 14x103 1.508 127 75x103 54.89 447 1.5 x
105
3-D FFT PDE FT 2562x128 5.631 196 512x2562 71.37 645 5123
Integer Sort IS 223x219 0.7812 68 225x221 3.150 244 227
LU Simulated CFDAppl.
LU 643 64.57 194 1023 319.6 711 1623
SP Simulated CFDAppl.
SP 643 102.0 216 1023 447.1 648 1623
BT Simulated CFDAppl.
BT 643 181.3 229 1023 721.5 705 1623
-
8 - 53
Table 2: Results of the Embarrassingly Parallel (EP)
benchmark.
Computer System
DateReceived
NumberProcessor
Class A Class B
Time in seconds
Ratio toCRAY
Y-MP/1
Time in seconds
Ratio toCRAYC90/1
BBN TC2000 Dec 91 64 284.0 0.44 NA NA
Convex ExemplarSPP1000
Mar 95 18
163264
376.848.124.311.86.1
0.332.625.19
10.6920.68
NA191.096.048.024.5
NA0.771.533.055.98
HP/Convex ExemplarSPP2000
Nov 96 148
16
160.941.220.710.5
0.783.066.10
12.02
NANANANA
NANANANA
CRAY C90 Feb 95 1248
16
36.6218.429.154.612.36
3.456.85
13.7927.3753.46
146.4173.6636.7818.379.35
1.01.993.987.97
15.66
CRAY J916 Feb 95 1248
16
169.4486.7043.0921.5410.78
0.741.462.935.86
11.70
675.71340.13170.1585.4943.16
0.220.430.861.713.39
CRAY T3D Feb 95 163264
128256512
1024
22.7411.375.682.871.440.720.55
5.5511.1022.2143.9687.62
175.24229.40
91.8345.9222.9511.475.742.872.19
1.593.196.38
12.7625.5151.0166.85
CRAY T3E Nov 96 248
163264
128256
56.228.114.17.03.51.80.90.4
2.254.498.95
18.0236.0570.09
140.19315.43
224.5112.256.128.114.07.03.51.8
0.651.312.615.21
10.4620.9241.8381.34
-
9 - 53
CRAY T916 July 95 1248
18.569.544.772.42
6.8013.2326.4552.14
76.1338.1119.129.65
1.923.847.66
15.17
CRAY Y-MP Aug 92 18
126.1715.87
1.07.95
NANA
NANA
DEC Alpha Server8400 5/300(300 MHz)
Oct 95 148
16
155.677.9739.1
19.71
0.811.613.236.40
622.22311.9
156.6978.43
0.240.470.931.87
DEC Alpha Server8400 5/440(437 MHz)
Nov 96 1248
12
94.632847.375023.812512.10168.4609
1.332.665.30
10.4314.91
378.3750189.234494.859447.617233.1172
0.390.771.543.074.42
Fujitsu VPP500 Aug 94 1248
163264
44.2522.3411
11.245.672.871.460.75
2.855.65
11.2322.2643.9686.42
168.23
176.6488.7866
44.5222.3611.265.682.88
0.831.653.296.5
13.0025.7850.84
Fujitsu VX Nov 96 124
30.331215.31057.3720
4.168.24
17.11
120.948557.690730.4367
1.212.544.81
Fujitsu VPP300 Nov 96 1248
16
30.331215.31057.37203.92171.9962
4.168.24
17.1132.1763.20
120.948557.690730.436714.60487.7563
1.212.544.81
10.0218.88
Fujitsu VPP700 Nov 96 1246
1632
30.331215.31057.37203.92171.99621.0265
4.168.24
17.1132.1763.20
122.91
120.948557.690730.436714.60487.75633.7444
1.212.544.81
10.0218.8839.10
IBM RS/6000 SPWide-node1 (67 MHz)
Mar 95 8163264
128
19.919.954.982.491.25
6.3412.6925.3450.67
100.94
79.7539.8919.99.954.99
1.843.677.36
14.7129.34
Table 2: Results of the Embarrassingly Parallel (EP)
benchmark.
-
10 - 53
IBM RS/6000 SPP2SC node (120 MHz)
Nov 96 8163264
NANANANA
NANANANA
39.8420.0410.025.02
3.687.31
14.6129.17
IBM RS/6000 SPThin-node2 (67 MHz)
Mar 95 8163264
128
20.8210.425.232.621.31
6.0612.1124.1248.1696.31
82.9441.4720.7510.375.19
1.773.537.06
14.1228.21
Intel iPSC/860 May 92 3264
128
102.751.425.7
1.232.464.91
NANANA
NANANA
Intel ParagonOSF R 1.2
Jan 95 64128256512
1024
15.29 7.673.872.231.15
8.2516.4532.6056.58
109.71
61.0430.5715.378.934.45
2.404.799.53
16.4032.90
Intel Paragon(SunMos turbo)
Jan 95 64128256512
1024
7.803.93 2.001.12 .59
16.1832.1063.09
112.65213.85
31.1515.607.823.982.05
4.709.39
18.6236.7971.42
Intel Paragon MPOSF R1.3
Jan 95 64128256512
8.024.172.261.28
15.7330.2655.8395.57
31.4215.888.114.23
4.669.22
18.0534.61
Intel Paragon MP(SunMos turbo)
Nov 95 64128256512
5.752.941.540.87
21.9442.9281.93
145.02
22.7011.405.782.98
6.4512.8425.3449.13
Kendall Square KSR1 Oct 93 163264
128
101.951.426.012.8
1.232.454.859.86
NANANANA
NANANANA
Kendall Square KSR2 Feb 94May 94
3264
24.813.0
5.099.71
NA46.6
NA3.14
Kyoto/MatsushitaADENART
Feb 94 256 32.9 3.83 NA NA
Table 2: Results of the Embarrassingly Parallel (EP)
benchmark.
-
11 - 53
MasPar MP-1 Aug 92 4K 16K
248.0 69.3
0.511.82
NANA
NANA
MasPar MP-2 Nov 92 16K 22.4 5.63 NA NA
Meiko CS-1 Aug 92 16 116.8 1.08 NA NA
Meiko CS-2 Oct 94 16326496
128
39.3920.4511.007.846.29
3.206.16
11.4616.0720.06
152.8177.2039.4826.8421.16
0.961.903.715.456.92
nCUBE-2S Mar 94 64128256512
1024
83.841.9320.9710.505.25
1.513.016.02
12.0224.03
336.3168.284.142.121.0
0.440.871.743.486.97
NEC SX-3 Oct 94 1 21.27 5.93 NA NA
NEC SX-4/32 Nov 96 1248
1632
NANANANANANA
NANANANANANA
89.5644.79
22.41711.2245.6582.944
1.633.276.53
13.0425.8749.73
Silicon GraphicsPower Challenge XL(75 MHz)
Oct 94 148
16
242.9561.4430.7715.48
0.522.054.108.15
973.62245.74122.9861.79
0.150.601.192.37
SGIPower Challenge XL(90 MHz)
May 95 1248
16
169.1087.4643.8721.9811.05
0.751.442.885.74
11.42
676.78352.31176.5287.8044.22
0.220.420.831.673.31
SGI Origin200(180 MHz)
Nov 96 124
110.4756.6829.43
1.142.234.29
442.77224.93114.87
0.330.651.27
Table 2: Results of the Embarrassingly Parallel (EP)
benchmark.
-
12 - 53
SGI Origin2000(195 MHz)
Nov 96 1248
163132
101.9351.7025.8313.006.68NA
3.55
1.242.444.889.71
18.89NA
35.54
408.25207.22103.4452.0626.71
NA14.17
0.360.711.422.815.48NA
10.33
Thinking MachinesCM-2
Oct 91 8K16K32K64K
126.663.9 33.718.8
1.001.973.746.71
NANANANA
NANANANA
Thinking MachinesCM-200
Oct 91 8K16K32K64K
76.939.220.710.9
1.643.226.10
11.58
NANANANA
NANANANA
Thinking MachinesCM-5
Nov 92 163264
128256512
42.421.510.95.42.71.4
2.985.87
11.5823.3646.7390.12
NANANANANANA
NANANANANANA
Thinking MachinesCM-5E
Sep 95 1632
27.213.6
4.649.28
108.554.3
1.352.70
Thinking MachinesCM-500
Sept 95 64128256
6.93.62.0
18.2935.0463.09
27.113.7 7.0
5.4010.6920.92
Table 2: Results of the Embarrassingly Parallel (EP)
benchmark.
-
13 - 53
Table 3: Results of the Multigrid (MG) benchmark.
Computer System
DateReceived
NumberProcessor
Class A Class B
Time in seconds
Ratio toCRAY
Y-MP/1
Time in seconds
Ratio toCRAYC90/1
Convex ExemplarSPP1000
Mar 95 18
163264
208.029.917.311.0NA
0.110.741.282.02NA
NA150.485.152.739.6
NA0.220.400.640.85
HP/Convex ExemplarSPP2000
Nov. 96 148
16
28.18.05.34.3
0.792.784.195.17
NANANANA
NANANANA
CRAY C90 Feb 95 1248
16
7.273.711.921.100.71
3.065.99
11.5820.2031.30
33.7817.248.894.593.43
1.001.963.807.369.85
CRAY EL Aug 92 148
89.1927.9422.30
0.25 0.800.95
NANANA
NANANA
CRAY J916 Feb 95 1248
16
39.0820.5210.755.883.82
0.571.092.073.782.06
184.8894.7148.6926.6016.12
0.180.360.691.272.10
CRAY T3D Feb 95 163264
128256512
1024
13.786.402.611.360.740.390.25
1.613.478.51
16.3430.0356.9788.88
66.5830.1012.566.573.371.741.15
0.511.112.695.14
10.0219.4129.38
CRAY T3E Nov 96 248
163264
128256
NA11.25.62.81.50.80.40.2
NA1.983.977.94
14.8127.7855.55111.1
NA52.726.413.47.03.82.01.1
NA0.641.282.524.838.89
16.8930.71
-
14 - 53
CRAY T916 July 95 1248
4.432.281.270.99
5.029.75
17.5022.44
20.3010.505.544.06
1.663.226.108.32
CRAY Y-MP Aug 92 18
22.222.96
1.007.51
NANA
NANA
Fujitsu VPP500 Mar 95 48
1632
1.440.750.420.26
15.4329.6352.9085.46
6.813.592.011.26
4.969.41
16.8126.81
Fujitsu VX Nov 96 4 1.38 16.10 6.58 5.13
Fujitsu VPP300 Nov 96 48
16
1.380.750.41
6.1029.6354.20
6.583.591.95
5.139.41
17.32
Fujitsu VPP700 Nov 96 48
1632
1.380.750.410.25
6.1029.6354.2088.88
6.583.591.951.30
5.139.41
17.3225.98
IBM SP-1 Mar 94 8163264
17.509.495.102.89
1.272.344.367.69
82.0344.57 24.3713.86
0.410.761.392.44
IBM RS/6000 SPWide-node1 (67 MHz)
Oct 94 8163264
128
6.043.171.690.950.53
3.687.01
13.1523.3941.92
27.9214.587.724.362.46
1.212.324.387.75
13.73
IBM RS/6000 SPP2SC node (120 MHz)
Nov 96 8163264
NANANANA
NANANANA
17.009.024.912.91
1.993.756.88
11.61
IBM RS/6000 SPThin-node2 (67 MHz)
Feb 95 8163264
128
7.183.741.991.120.63
3.095.94
11.1719.8435.27
32.7317.139.145.202.95
1.041.973.966.50
11.45
Intel iPSC/860 Aug 92 128 8.6 2.58 NA NA
Table 3: Results of the Multigrid (MG) benchmark.
-
15 - 53
Intel Paragon MP(OSF1.3)
Jan 95 64128256
6.13.532.48
3.646.298.96
28.715.910.4
1.182.123.24
Intel Paragon( SunMos )
Jan 95 64128256512
7.724.052.351.75
2.885.499.46
12.70
36.919.511.4 8.51
0.921.732.963.97
Intel Paragon(SunMos Turbo)
Nov 95 64128256512
5.853.171.871.46
3.87.01
11.8815.22
27.915.29.057.01
1.212.223.734.82
Intel Paragon MP SunMos Turbo
Jan 95 64128256512
4.832.781.671.37
4.607.99
13.3116.22
23.313.48.126.72
1.452.524.165.03
Kendall Square KSR1 Feb 94 3264
128
19.710.35.6
1.132.163.97
NANANA
NANANA
Kendall Square KSR2 Feb 94May 94
32 64
10.35.7
2.20 3.90
NA26.1
NA1.29
MasPar MP-1 Aug 92 16K 12.0 1.86 NA NA
MasPar MP-2 Nov 92 16K 4.36 5.09 NA NA
Meiko CS-1 Aug 92 16 42.8 0.52 NA NA
Meiko CS-2 Oct 94 16 64128
7.602.351.43
2.93 9.8315.54
35.4610.766.55
0.953.145.16
NEC SX-3 Oct 94 1 2.80 7.94 13.16 2.57
NEC SX-4/32 Nov 96 1248
1632
NANANANANANA
NANANANANANA
19.289.765.002.631.460.95
1.753.466.76
12.8423.1435.56
Table 3: Results of the Multigrid (MG) benchmark.
-
16 - 53
nCUBE-2S Mar 94 64128512
1024
37.619.2 5.3 2.8
0.591.164.197.94
NANANANA
NANANANA
SGIPower Challenge90 MHz
Oct 95 1248
16
37.9720.0310.63 6.55 5.71
0.591.112.093.393.89
176.22 93.3049.4530.4326.30
0.190.360.681.111.28
SGI Origin200(180 MHz)
Nov 96 124
34.6921.7712.89
0.641.021.72
245.09151.0477.05
0.140.220.44
SGI Origin2000(195 MHz)
Nov 96 1248
163132
30.1417.7510.676.003.34NA
2.12
0.741.252.083.706.65NA
10.48
194.88113.7956.1928.4115.03
NA8.61
0.170.300.601.192.25NA
3.92
Thinking MachinesCM-2
Dec 91 16K32K64K
45.826.014.1
0.490.851.58
NANANA
NANANA
Thinking MachinesCM-200
Dec 91 16K32K
30.217.2
0.741.29
NANA
NANA
Thinking MachinesCM-5
Aug 93 3264
128
19.510.96.1
1.142.033.64
NANANA
NANANA
Thinking MachinesCM-5E
Sep 95 1632
7.73.8
2.895.85
36.118.7
0.941.81
Thinking MachinesCM-500
Sep 95 64128256
2.21.410.91
10.115.7624.42
10.66.23.9
3.195.498.67
Table 3: Results of the Multigrid (MG) benchmark.
-
17 - 53
Table 4: Results of the Conjugate Gradient (CG) benchmark.
Computer System
DateReceived
.NumberProcessor
Class A Class B
Time in seconds
Ratio toCRAYY-MP/1
Time in seconds
Ratio toCRAYC90/1
BBN TC2000 Dec 91 40 51.4 0.23 NA NA
Convex ExemplarSPP1000
Mar 95 18
163264
202.922.28.944.30NA
0.060.541.332.77NA
NANA
837.0485.4292.1
NANA
0.150.250.42
HP/Convex ExemplarSPP2000
Nov 96 148
16
37.79.85.02.7
0.321.222.384.41
NANANANA
NANANANA
CRAY C90 Feb 95 1248
16
3.431.790.950.530.34
3.486.66
12.5522.4935.06
122.9063.1133.2518.1110.61
1.001.953.706.79
11.58
CRAY EL Sep 93 148
45.2414.2910.14
0.260.831.18
NANANA
NANANA
CRAY J916 July 95 1248
16
15.938.424.422.611.68
0.751.422.704.577.10
532.03293.24150.9280.6742.86
0.230.420.811.522.87
CRAY T3D Feb 95 163264
128256512
1024
14.377.443.932.111.210.720.58
0.831.603.035.659.85
16.5620.55
570.11291.30158.8182.0747.1527.3416.58
0.220.420.771.502.614.507.41
-
18 - 53
CRAY T3E Nov 96 128
163264
128256
NANA6.52.91.60.90.60.4
NANA
1.834.117.45
13.2419.8729.80
802.3404.5205.6107.159.333.822.922.2
0.150.300.601.152.073.645.375.54
CRAY T916 July 95 148
16
1.951.100.580.38
6.1110.8420.5531.37
73.9837.7919.6511.43
1.663.256.25
10.75
CRAY Y-MP Aug 92 18
11.922.38
1.005.01
NANA
NANA
Fujitsu VPP500 Aug 94 1248
151630
5.683.061.721.04NA
0.80NA
2.103.906.93
11.46NA
14.90NA
NA104.5155.4031.8020.85
NA15.21
NA1.182.223.865.89NA
8.08
Fujitsu VX Nov 96 124
5.87033.18021.8185
2.033.756.55
NA104.140857.8277
NA1.182.13
Fujitsu VPP300 Nov 96 1248
1516
5.87033.18021.81851.0656
NA0.7514
2.033.756.55
11.19NA
15.86
NA104.140857.827732.060519.7696
NA
NA1.182.133.836.22NA
Fujitsu VPP700 Nov 96 1248
15163035
5.87033.18021.81851.0656
NA0.7514
NA0.6726
2.033.756.55
11.19NA
15.8NA
17.72
NA104.140857.827732.060519.7696
NA13.4585
NA
NA1.182.133.836.22NA
9.13NA
Table 4: Results of the Conjugate Gradient (CG) benchmark.
-
19 - 53
IBM SP-1 Feb 94 8163264
21.3712.827.984.72
0.560.931.492.53
NA 638.2362.9193.4
NA 0.190.340.64
IBM RS/6000 SPWide-node1 (67 MHz)
Mar 94 8163264
128
4.913.092.091.6
1.38
2.433.865.707.458.64
156.2188.4
52.5333.7925.44
0.791.392.343.644.83
IBM RS/6000 SPP2SC node (120 MHz)
Nov 96 8163264
NANANANA
NANANANA
115.8862.2136.2222.71
1.061.983.395.41
IBM RS/6000 SPThin-node2 (67 MHz)
Mar 95 8163264
128
5.603.482.341.721.48
2.133.435.096.938.05
234.46120.2367.1638.5228.50
0.521.021.833.194.31
Intel iPSC/860 Sep 93 128 7.0 1.70 NA NA
Intel Paragon(OSF1.2)
Mar 94 64128256512
4.103.302.83NA
2.913.614.21NA
NA132.570.0 47.6
NA0.931.762.58
Intel Paragon(SunMos)
Nov 95 64128256512
3.592.76 2.44
NA
3.324.314.89NA
NA125.4 63.640.5
NA0.981.933.03
Kendall Square KSR1 Feb 94 3264
19.013.4
0.630.89
NANA
NANA
Kendall Square KSR2 Feb 94 3264
9.8 6.1
1.221.95
NA182.0
NA0.67
MasPar MP-1 Aug 92 4K16K
64.514.6
0.180.82
NANA
NANA
MasPar MP-2 Nov 92 16K 11.0 1.08 NA NA
Meiko CS-1 Aug 92 16 67.5 0.18 NA NA
Meiko CS-2 Oct 94 1632
7.185.60
1.662.13
248.30156.50
0.490.79
Table 4: Results of the Conjugate Gradient (CG) benchmark.
-
20 - 53
nCUBE-2S Mar 94 64128256512
1024
29.616.9 9.66.24.1
0.400.7 11.241.922.91
NANANANANA
NANANANANA
NEC SX-4/32 Nov 96 124
NANANA
NANANA
81.7745.1226.31
1.502.724.67
Silicon GraphicsPower Challenge XL(75 MHz)
Oct 94 1248
16
39.016.97.24.53.5
0.310.711.662.653.41
NANANANANA
NANANANANA
SGIPower Challenge(90 MHz)
May 95
Oct 95
1248
16
35.1419.588.794.032.54
0.340.611.352.964.69
NANANANANA
NANANANANA
SGI Origin200(180 MHz)
Nov 96 124
31.3818.679.76
0.380.641,22
1731.181002.32624.62
0.070.120.20
SGI Origin2000(195 MHz)
Nov 96 1248
1631
26.8914.376.233.022.04NA
0.440.831.913.955.84NA
1538.71851.82490.76283.19156.5779.65
0.080.140.250.430.781.54
Thinking MachinesCM-2
Mar 92 8K16K32K
25.614.1 8.8
0.470.851.35
NANANA
NANANA
Thinking MachinesCM-5
Aug 93 32 64128
20.710.66.2
0.581.121.92
NANANA
NANANA
Thinking MachinesCM-5E
Sep 95 1632
13.58.0
0.881.49
454251
0.270.49
Thinking MachinesCM-500
Sep 95 64128256
5.43.93.4
2.213.063.51
14991 62
0.821.351.98
Table 4: Results of the Conjugate Gradient (CG) benchmark.
-
21 - 53
Table 5: Results of the 3-D FFT PDE (FT) benchmark.
Computer System
DateReceived
NumberProcessor
Class A Class B
Time in seconds
Ratio toCRAY
Y-MP/1
Time in seconds
Ratio toCRAYC90/1
Convex ExemplarSPP1000
Mar 95 18
1632
178.625.520.513.9
0.161.131.402.07
NA375.4
NANA
NA0.29NANA
CRAY C90 Feb 95 1248
16
8.954.532.291.290.80
3.216.35
12.5622.3035.96
110.6055.7527.9514.127.65
1.001.983.967.83
14.46
CRAY EL May 93 148
105.127.918.5
0.271.031.56
NANANA
NANANA
CRAY J916 July 95 1248
16
42.8422.0811.215.803.41
0.671.302.574.978.44
530.06267.92134.9270.5138.06
0.210.410.821.572.91
CRAY T3D Feb 95 163264
128256512
1024
11.805.902.991.520.770.510.32
2.444.879.62
18.9337.3656.4189.91
NANA
40.5720.6810.776.443.76
NANA
2.735.35
10.2717.1729.41
CRAY T3E Nov 96 248
163264
128256
28.914.47.33.71.91.00.50.2
1.002.003.947.78
15.1428.7757.54
143.85
NANA
86.243.121.811.05.52.8
NANA
1.282.575.07
10.0520.1139.5
-
22 - 53
CRAY T916 July 95 1248
5.232.671.400.98
5.5010.7820.5529.36
64.8132.3916.659.48
1.713.416.64
11.67
CRAY Y-MP Aug 92 18
28.774.19
1.06.87
NANA
NANA
Fujitsu VPP500 Sept 95 1248
163264
11.255.672.881.440.750.400.24
2.565.079.99
19.9838.3671.93
119.88
NANA
31.2115.927.944.072.18
NANA
3.546.95
13.9127.1750.73
Fujitsu VX Nov 96 124
6.32724.37882.2557
4.556.57
12.75
NA49.688524.5128
NA2.234.51
Fujitsu VPP300 Nov 96 1248
16
6.32724.37882.25571.14910.6005
4.556.57
12.7525.0447.91
NA49.688524.512812.37586.3013
NA2.234.518.94
17.55
Fujitsu VPP700 Nov 96 1248
1632
6.32724.37882.25571.14910.60050.3345
4.556.57
12.7525.0447.9186.01
NA49.688524.512812.37586.30133.2577
NA2.234.518.94
17.5533.95
IBM SP-1 Feb 94 8163264
43.6822.8612.086.46
0.661.262.384.45
NA286.5143.274.5
NA0.390.771.49
IBM RS/6000 SPWide-node1 (67 MHz)
Oct 94 8163264
128
13.317.173.962.191.23
2.164.017.2713.4
23.39
NA91.8
47.2326.0514.52
NA1.202.344.257.62
Table 5: Results of the 3-D FFT PDE (FT) benchmark.
-
23 - 53
IBM RS/6000 SPP2SC node (120 MHz)
Nov 96 8163264
NANANANA
NANANANA
83.2440.5221.0715.07
1.332.735.257.34
IBM RS/6000 SPThin-node2 (67 MHz)
Mar 95 8163264
128
14.788.094.312.391.30
1.953.566.68
12.0422.13
NA101.0351.3828.0215.68
NA1.092.153.957.05
Intel iPSC/860 Dec 91Apr 92
64128
20.99.7
1.38 2.97
NANA
NANA
Intel ParagonOSF R 1.2
Nov 95 163264
128256
24.712.996.833.853.46
1.162.214.217.478.32
NANA
83.642.0725.93
NANA
1.322.634.27
Intel ParagonSunMos
Nov 95 163264
128256512
20.4610.545.473.132.82NA
1.412.735.269.19
10.20NA
NANA
60.932.1518.6616.17
NANA
1.823.445.936.84
Intel Paragon MPR1.3
Nov 95 163264
128256
17.69.525.162.932.73
1.633.025.589.82
10.54
NANA
60.931.0
19.87
NANA
1.823.575.57
Intel Paragon MPSunMos Turbo
Nov 95 64 128256512
3.872.382.201.92
7.4312.0913.0814.98
41.822.7614.212.4
2.654.867.798.92
Kendall Square KSR1 Feb 94 3264
16.29.2
1.783.13
NANA
NANA
Kendall Square KSR2 Feb 94May 94
3264
9.06.5
3.204.43
NA124.0
NA0.89
Kyoto/MatsushitaADENART
Feb 94 256 72.7 0.4 NA NA
Table 5: Results of the 3-D FFT PDE (FT) benchmark.
-
24 - 53
MasPar MP-1 Aug 92 16K 18.3 1.57 NA NA
MasPar MP-2 Nov 92 16K 8.0 3.60 NA NA
Meiko CS-1 Aug 92 16 170.0 0.17 NA NA
Meiko CS-2 Oct 94 16 3264
12.677.174.53
2.274.016.35
NA82.7148.04
NA1.342.30
nCUBE-2S Mar 94 64128256512
1024
62.832.916.08.44.1
0.460.871.8
3.437.02
NANANANANA
NANANANANA
NEC SX-3 Oct 94 1 2.79 10.31 37.52 2.95
NEC SX-4/32 Nov 96 1248
1632
NANANANANANA
NANANANANANA
56.5728.1414.798.034.743.14
1.963.937.48
13.7723.3335.22
Silicon GraphicsPower Challenge XL(75 MHz)
Oct 94 1248
16
61.1735.5319.9812.5711.18
0.470.811.442.292.57
761.67414.52223.97130.15110.37
0.150.270.490.851.00
SGIPower Challenge(90 MHz)
May 95 1248
16
51.8927.4916.6410.479.23
0.551.051.732.753.12
642.3331.5181.4110.693.9
0.170.330.611.001.18
SGI Origin200(180 MHz)
Nov 96 124
66.5635.5118.61
0.430.811.55
NANANA
NANANA
Table 5: Results of the 3-D FFT PDE (FT) benchmark.
-
25 - 53
SGI Origin2000(195 MHz)
Nov 96 1248
1632
55.4329.5315.568.054.553.16
0.520.971.853.576.329.10
810.41419.72215.95108.4254.8828.66
0.140.260.511.022.023.85
Thinking MachinesCM-2
Dec 91 16K 32K64K
37.018.211.4
0.781.582.52
NANANA
NANANA
Thinking MachinesCM-200
Dec 91 8K 45.6 0.63 NA NA
Thinking MachinesCM-5
Aug 93 32 64
128
14.97.96.6
1.933.644.36
NANANA
NANANA
Thinking MachinesCM-5E
Sep 95
Feb 94
1632 64
128
12.86.73.92.9
2.254.297.389.92
16082.046.034.0
0.691.352.403.25
Thinking MachinesCM-500
Sep 95 64 128256
3.51.961.33
8.2214.6721.63
41.922.513.2
2.644.928.38
Table 5: Results of the 3-D FFT PDE (FT) benchmark.
-
26 - 53
Table 6: Results of the Integer Sort (IS) benchmark.
Computer System Date
ReceivedNumber
Processor
Class A Class B
Time in seconds
Ratio toCray
Y-MP/1
Time in seconds
Ratio toCrayC90/1
Convex ExemplarSPP1000
Mar 95 18
83.210.1
0.141.13
NA43.5
NA0.30
HP/Convex ExemplarSPP2000
Nov 96 148
16
30.67.043.492.21
0.371.633.285.19
NANANANA
NANANANA
CRAY C90 Feb 95 1248
16
3.331.640.850.460.27
3.446.99
13.4824.9142.44
12.926.503.301.730.98
1.01.993.927.47
13.18
CRAY EL Sep 93 148
43.7612.99 8.45
0.260.881.35
NANANA
NANANA
CRAY J916 July 95 1248
16
13.757.023.812.211.63
0.831.633.005.197.03
54.4127.9613.937.604.91
0.240.460.931.702.63
CRAY T3D Feb 95 163264
128256512
1024
7.073.892.091.050.550.310.44
1.622.955.48
10.9120.8436.9726.05
NA16.578.744.562.361.331.22
NA0.781.482.835.479.71
10.59
-
27 - 53
CRAY T3E Nov 96 248
163264
128256
16.98.14.02.21.20.60.30.2
0.681.422.875.219.55
19.1038.2057.30
NA38.918.69.85.12.91.60.8
NA0.330.691.322.534.468.08
16.15
CRAY T916 July 95 1248
2.021.020.520.38
5.6711.2422.0430.16
7.443.741.921.41
1.743.456.739.16
CRAY Y-MP Aug 92 18
11.461.85
1.006.19
NANA
NANA
Fujitsu VPP500 Apr 94 1248
2.1891.5741.0980.917
5.247.28
10.4412.50
NANA
3.703.03
NANA
3.494.26
Fujitsu VX Nov 96 124
2.39681.80381.2519
4.786.359.15
9.19646.08754.1363
1.412.123.12
Fujitsu VPP300 Nov 96 1248
16
2.39681.80381.25191.12491.0204
4.786.359.15
10.1911.23
9.19646.08754.13633.19592.7231
1.412.123.124.044.74
Fujitsu VPP700 Nov 96 1248
1632
2.39681.80381.25191.12491.02040.9839
4.786.359.15
10.1911.2311.64
9.19646.08754.13633.19592.72312.7211
1.412.123.124.044.744.75
IBM SP-1 Feb 94 8163264
16.818.855.043.06
0.681.29 2.273.75
NA 37.320.111.2
NA0.350.641.15
Table 6: Results of the Integer Sort (IS) benchmark.
-
28 - 53
IBM RS/6000 SPWide-node1 (67 MHz)
Mar 95 8163264
128
4.932.651.540.890.59
2.324.327.44
12.8819.42
19.7510.605.923.411.98
0.651.222.183.796.53
IBM RS/6000 SPP2SC node (120 MHz)
Nov 96 8163264
NANANANA
NANANANA
11.165.853.201.75
1.162.214.047.38
IBM RS/6000 SPThin-node2 (67 MHz)
Feb 95 8163264
128
5.162.891.660.910.61
2.223.976.90
12.5918.79
20.7911.466.373.582.05
0.621.132.033.616.30
Intel iPSC/860 May 92 3264
128
25.717.313.6
0.450.660.84
NANANA
NANANA
Intel Paragon(OSF1.2)
Mar 94 3264
128256512
7.814.342.41NANA
1.472.644.76NANA
NA17.339.525.94 4.69
NA0.751.362.182.75
Intel Paragon(SunMos)
Mar 94 3264
128
5.483.772.29
2.093.045.00
NA11.987.22
NA1.081.79
Kendall Square KSR1 Feb 94 3264
10.86.6
1.06 1.74
NANA
NANA
Kendall Square KSR2 Feb 94May 94
3264
7.03.9
1.642.94
NA20.3
NA0.64
Kyoto/MatsushitaADENART
Feb 94 256 46.6 0.25 NA NA
MasPar MP-1 Jan 93 16K 11.5 1.00 NA NA
MasPar MP-2 Jan 93 16K 7.7 1.49 NA NA
Meiko CS-1 Aug 92 16 62.7 0.18 NA NA
Table 6: Results of the Integer Sort (IS) benchmark.
-
29 - 53
nCUBE-2S Mar 94 64128256512
1024
23.2 12.0 6.13.21.7
0.490.961.883.58 6.74
NA47.5NA
12.5 6.5
NA0.27NA
1.031.99
NEC SX-4/32 Nov 96 1248
16
NANANANANA
NANANANANA
6.773.50
1.8771.1390.896
1.913.696.88
11.3414.42
SGIPower Challenge(90 MHz)
Oct 95 1248
20.011.97.15.0
0.570.961.612.29
NANANANA
NANANANA
SGI Origin200(180 MHz)
Nov 96 124
17.269.184.89
0.661.252.34
235.05154.2879.18
0.0550.0840.16
SGI Origin2000(195 MHz)
Nov 96 1248
1632
13.667.073.802.141.341.21
0.841.623.025.368.559.47
104.4854.3627.2515.098.895.76
0.120.240.470.861.452.24
Thinking MachinesCM-200
Dec 91 64K 5.7 2.01 NA NA
Thinking MachinesCM-5
Aug 93 3264
128
43.124.2 12.0
0.270.470.96
NANA
NANANA
Thinking MachinesCM-5E
Sep 95
Feb 94
163264
128
11.96.13.1
1.66
0.961.88 3.706.90
NA31.416.4 8.4
NA0.41 0.791.54
Thinking MachinesCM-500
Sep 95 64128256
3.161.67 1.16
3.636.869.88
16.18.2 4.3
0.801.583.0
Table 6: Results of the Integer Sort (IS) benchmark.
-
30 - 53
Table 7: Results of the LU CFD Application (LU)benchmark..
ComputerSystem
DateReceived
No.Proc.
Class A Class B
Time in seconds
Ratio toCray
YMP/1
Time in seconds
Ratio toCrayC90/1
BBN TC2000 Dec 91 62 3032.0 0.11 NA NA
Convex ExemplarSPP1000
Mar 95 18
1632
2668331 196 126
0.131.001.702.65
NA1492827
465.9
NA0.300.540.9
CRAY C90 Feb 95 1248
16
119.7862.2932.2017.1510.17
2.785.35
10.3619.4532.79
449.54231.98121.2663.0337.93
1.001.943.717.13
11.85
CRAY EL Aug 92 148
1449.0522.3351.6
0.23 0.64 0.95
NANANA
NANANA
CRAY J916 July 95 1248
16
492.83254.94135.0772.7347.59
0.681.312.474.517.01
1994.051024.74526.29286.17170.26
0.230.440.851.572.64
CRAY T3D Feb 95 163264
128256512
1024
205.69106.8955.3228.7115.949.027.09
1.623.126.03
11.6220.9236.9747.04
844.53451.18233.45120.5365.0636.3920.77
0.531.001.933.736.9
12.3521.64
-
31 - 53
CRAY T3E Nov 96 248
163264
128256
356.4179.290.246.924.012.97.14.2
0.941.863.707.11
13.9025.8546.9779.41
1497.4728.9365.8185.195.951.529.516.8
0.300.621.232.434.698.73
15.2426.76
CRAY T916 July 95 1248
16
82.6746.9123.6815.74
NA
4.037.11
14.0821.19
NA
293.38149.3377.9543.1328.26
1.533.015.77
10.4215.91
CRAY Y-MP Aug 92 18
333.549.5
1.006.74
NANA
NANA
DEC Alpha Server8400 5/300(300 MHz)
Oct 95 1248
12
528.6301.77158.7891.7279.13
0.631.112.103.644.21
2304.91350.0691.8
376.23296.19
0.200.330.651.191.51
DEC Alpha Server8400 5/440(437 MHz)
Nov 96 1248
378.61219.41115.8367.965
0.881.522.884.91
1684.47987.54500.12271.32
0.270.460.901.66
Fujitsu VPP500 Oct 95 123468
161734
101.96061.409
NA35.789
NA21.04113.944
NANA
3.275.43NA
9.31NA
15.8523.92
NANA
414.82242.23175.20
NA97.851
NANA
43.64428.095
1.081.862.57NA
4.59NANA
10.3016.00
Fujitsu VX Nov 96 1234
77.73250.880
NA30.500
4.296.55NA
10.93
297.94190.00138.34
NA
1.512.373.25NA
-
32 - 53
Fujitsu VPP300 Nov 96 123468
16
77.73250.880
NA30.500
NA18.44412.553
4.296.55NA
10.93NA
18.0826.57
297.94190.00138.34
NA78.026
NANA
1.512.373.25NA
5.76NANA
Fujitsu VPP700 Nov 96 123468
161732
77.73250.880
NA30.500
NA18.44412.553
NA10.063
4.296.55NA
10.93NA
18.0826.57
NA33.14
297.94190.00138.34
NA78.026
NANA
36.736NA
1.512.373.25NA
5.76NANA
12.24NA
IBM SP-1 Feb 94 8163264
291.4172.9101.863.2
1.141.933.285.28
NA604.8348.1207.5
NA0.741.292.17
IBM RS/6000 SPWide-node1 (67 MHz)
Mar 95 8163264
128
112.564.636.522.715.2
2.965.169.14
14.6921.94
429.8234.4129.776.847.8
1.051.923.475.859.4
IBM RS/6000 SPWide-node2 (77 MHz)
Oct 95 18
163264
501.580.8148.3629.4119.20
0.674.136.90
11.3417.37
2066.6311.53171.7998.0459.45
0.221.442.624.59 7.56
IBM RS/6000 SPP2SC node (120 MHz)
Nov 96 8163264
NANANANA
NANANANA
236.64126.7773.1642.89
1.903.556.14
10.48
IBM RS/6000 SPThin-node1 (67 MHz)
Mar 95 8163264
128
120.870.940.124.515.9
2.764.708.32
13.6120.97
477.3255.4141.382.951.2
0.941.763.185.428.78
Intel iPSC/860 Mar 91 64128
690.8 442.5
0.480.75
NANA
NANA
-
33 - 53
Intel Paragon (OSF1.2) Jul 94 64128256512
190.0118.075.0NA
1.762.834.45NA
675.0406.0254.0175.0
0.671.111.772.57
Kendall Square KSR1 Feb 94 3264
128
341.0199.0155.0
0.981.682.15
NANANA
NANANA
Kendall Square KSR2 Feb 94May 94
32 64
172.0102.0
1.933.27
NA424.0
NA1.06
MasPar MP-1 Aug 92 4K 1580.0 0.21 NA NA
MasPar MP-2 Nov 92 4K 463.5 0.72 NA NA
Meiko CS-1 Aug 92 16 2937.0 0.11 NA NA
nCUBE-2S Mar 94 64128256512
1024
1322.0712.5389.1226.1134.1
0.250.47 0.861.482.49
NANANANANA
NANANANANA
NEC SX-4/32 Nov 96 1248
161732
NANANANANANANA
NANANANANANANA
291.06148.2977.24643.08727.15125.29420.909
1.543.035.82
10.4316.5617.7721.50
Silicon GraphicsPower Challenge XL(75 MHz)
Oct 94 148
16
604.0231.8111.765.3
0.551.442.995.11
2617.91010.5550.2308.1
0.170.440.821.46
SGIPower Challenge - XL(90 MHz)
Oct 94 1248
16
549.02356.73188.33101.0065.90
0.610.931.773.305.06
2439.901500.30774.93419.90292.02
0.180.300.581.071.54
SGI Origin200(180 MHz)
Nov 96 124
397.30206.52114.39
0.841.612.92
1873.60958.01519.32
0.240.470.87
-
34 - 53
SGI Origin2000(195 MHz)
Nov 96 1248
162531
351.35181.3997.5250.5629.91
NA18.93
0.951.843.426.60
11.15NA
17.62
1522.30777.43401.80221.32130.4189.75
NA
0.300.581.122.033.455.01NA
Thinking MachinesCM-2
Mar 91 8K16K32K
1307.0850.0546.0
0.260.39 0.61
NANANA
NANANA
Thinking MachinesCM-5
Aug 93 3264
128
418.0272.0171.0
0.801.231.95
NANANA
NANANA
Thinking MachinesCM-5E
Sep 95
Feb 94
163264
128
25614397.065.0
1.302.333.445.13
957533
367.0318.0
0.470.841.221.41
Thinking MachinesCM-500
Sep 95 64128256
90 6143
3.715.477.76
336247149
1.341.823.02
-
35 - 53
Table 8: Results of the SP simulated CFD application (SP)
benchmark.
Computer System
DateReceived
NumberProcessor
Class A Class B
Time in seconds
Ratio toCray
YMP/1
Time in seconds
Ratio toCrayC90/1
BBN TC2000 Dec 91 112 880.0 0.54 NA NA
Convex Exemplar SPP1000
Mar 95 18
163264
2533345228144102
0.191.372.073.274.62
NA15841068697.4449.5
NA0.440.650.991.5
HP/Convex Exemplar SPP2000
Nov 96 148
16
770.9198.6101.656.6
0.612.374.648.33
NANANANA
NANANANA
CRAY C90 Feb 95 1248
16
174.5087.3244.7522.7412.82
2.705.40
10.5420.7336.78
689.60345.57175.8590.8052.22
1.002.003.927.59
13.21
CRAY EL Aug 92 148
2025.7601.9488.4
0.23 0.78 0.97
NANANA
NANANA
CRAY J916 July 95 1248
16
870.47436.91226.66118.3077.54
0.541.082.083.996.08
3728.691927.50941.08521.73316.58
0.180.360.731.322.18
CRAY T3D Feb 95 163264
128256512
1024
202.11104.1053.2627.5414.718.915.41
2.334.538.85
17.1232.0552.9287.15
818.07463.62233.52130.4574.8942.6325.23
0.841.492.955.299.21
16.1827.33
-
36 - 53
CRAY T3E Nov 96 248
163264
128256
621.9313.0157.579.839.220.411.26.0
0.761.513.005.91
12.0323.1142.1078.58
2427.71230.5611.7312.9162.785.548.126.9
0.280.561.132.204.248.07
14.3425.64
CRAY T916 Jul 95 1248
16
112.3756.4630.4218.69
NA
4.208.35
15.5025.23
NA
427.56214.63113.0260.6140.59
1.613.216.10
11.3816.99
CRAY Y-MP Aug 92 18
471.564.6
1.017.30
NANA
NANA
DEC Alpha Server8400 5/300(300 MHz)
Mar 95 148
12
749.61199.17118.04102.75
0.632.373.994.59
3448.10904.45452.13364.54
0.200.761.531.89
DEC Alpha Server8400 5/440(437 MHz)
Nov 96 1248
575.80290.83153.8991.430
0.821.623.065.16
2569.51294.9661.45363.35
0.270.531.041.90
Fujitsu VPP500 Mar 95 12468
161732345164
99.30961.58832.114
NA16.3998.5761
NA4.5355
NANA
2.5483
4.757.66
14.68NA
28.7554.98
NA103.96
NANA
185.03
404.08241.23127.4883.71064.930
NA30.474
NA15.67410.654
NA
1.712.865.418.24
10.62NA
22.63NA
44.064.73
NA
Fujitsu VX Nov 96 124
80.50460.43631.324
5.867.80
15.05
331.36219.11114.41
2.083.156.03
Table 8: Results of the SP simulated CFD application (SP)
benchmark.
-
37 - 53
Fujitsu VPP300 Nov 96 12468
16
80.50460.43631.324
NA16.0548.2465
5.867.80
15.05NA
29.3757.18
331.36219.11114.4178.38158.781
NA
2.083.156.038.80
11.73NA
Fujitsu VPP700 Nov 96 12468
16173234
80.50460.43631.324
NA16.0548.2465
NA4.5311
NA
5.867.80
15.05NA
29.3757.18
NA104.06
NA
331.36219.11114.4178.38158.781
NA27.860
NA14.621
2.083.156.038.80
11.73NA
24.75NA
47.17
IBM SP-1 Feb 94 8163264
441.6268.7165.0100.4
1.071.752.864.69
NA941.2522.4 302.3
NA0.731.322.28
IBM RS/6000 SPWide-node1 (67 MHz)
Mar 95 8163264
128
143.883.248.730.118.7
3.285.679.68
15.6625.21
589.3300.6163.891.754.8
1.172.294.217.52
12.58
IBM RS/6000 SPWide-node2 (77 MHz)
Oct 95 18
163264
711.8114.1569.3243.2026.46
0.664.136.80
10.9117.82
3087.0453.66248.51142.8880.17
0.221.522.774.838.60
IBM RS/6000 SPP2SC node (120 MHz)
Nov 96 8163264
NANANANA
NANANANA
359.81182.30107.1658.35
1.923.786.44
11.82
IBM RS/6000 SPThin-node2 (67 MHz)
Mar 95 8163264
128
161.193.353.632.720.6
2.935.058.80
14.4222.89
640.9342.3184.4101.659.9
1.082.013.746.79
11.51
Table 8: Results of the SP simulated CFD application (SP)
benchmark.
-
38 - 53
Intel iPSC/860 Jul 94Aug 92
64128
640.0 449.5
0.741.05
NANA
NANA
Intel Paragon(OSF1.2)
Jul 94 64102128204256324400484
226.0NA
143.0NA
97.089.0NANA
2.09NA
3.30NA
4.865.30NANA
960.0610.0
NA387.0301.0262.0246.0209.0
0.721.13NA
1.782.292.632.803.30
Kendall Square KSR1 Feb 94 3264
128
418.0257.0160.0
1.131.83 2.95
NANANA
NANANA
Kendall Square KSR2 Feb 94 May 94
3264
221.0131.0
2.133.59
NA495.0
NA1.39
Kyoto/MatsushitaADENART
Feb 94 256 209.9 2.24 NA NA
MasPar MP-1 Aug 92 4K 1772 0.27 NA NA
MasPar MP-2 Nov 92 4K 615 0.77 NA NA
Meiko CS-1 Aug 92 16 2975 0.16 NA NA
nCUBE-2S Mar 94 64128256512
1024
1243.2717.4387.3208.6120.9
0.380.661.222.263.90
NANANANANA
NANANANANA
NEC SX-3 Oct 94 1 75.72 6.23 495.0 1.39
NEC SX-4/32 Nov 96 1248
16172632
NANANANANANANANA
NANANANANANANANA
398.77199.0899.98452.35328.82324.69717.41017.130
1.733.466.90
13.1723.9327.9239.6140.26
Table 8: Results of the SP simulated CFD application (SP)
benchmark.
-
39 - 53
Silicon GraphicsPower Challenge XL(75 MHz)
Oct 94 148
16
858.3225.8119.567.2
0.552.093.957.02
3719.5947.6491.4313.1
0.190.731.402.20
SGIPower Challenge - XL(90 MHz)
May 95 1248
16
757.00387.90200.44106.4363.18
0.621.222.354.437.46
3343.901685.30853.35445.38294.22
0.210.410.811.552.34
SGI Origin200(180 MHz)
Nov 96 124
768.25408.32222.04
0.611.152.12
4036.902259.901229.30
0.170.310.56
SGI Origin2000(195 MHz)
Nov 96 1248
162531
675.28346.02187.8697.9752.01
NA30.47
0.701.362.514.819.07NA
15.47
3068.701628.00803.64423.61234.99140.48
NA
0.220.420.861.632.944.91NA
Thinking MachinesCM-2
Dec 91 16K 32K64K
1444.0917.0640.0
0.330.510.74
NANANA
NANANA
Thinking MachinesCM-5
3264
128
289.0170.0119.0
1.632.773.96
NANANA
NANANA
Thinking MachinesCM-5E
Sep 95
Feb 94
163264
128
268144
104.061.0
1.763.274.537.73
1580771
595.0320.0
0.440.891.162.16
Thinking MachinesCM-500
Sep 95 64128256
85 51 34
5.559.25
13.87
405 236140
1.702.924.93
Table 8: Results of the SP simulated CFD application (SP)
benchmark.
-
40 - 53
Table 9: Results of the BT simulated CFD application (BT)
benchmark.
Computer System Date
ReceivedNumber
Processor
Class A Class B
Time inseconds
Ratio toCRAY
Y-MP/1
Time in seconds
RatioCRAYC90/1
BBN TC2000 Dec 91 112 1378.0 0.58 NA NA
Convex ExemplarSPP1000
Mar 95 18
163264
282536621112578
0.282.173.766.34
10.16
NA1675984
559.8338.2
NA0.611.041.823.03
HP/Convex ExemplarSPP2000
Nov 96 148
16
829.4212.9107.255.3
0.963.727.39
14.33
NANANANA
NANANANA
CRAY C90 Feb 95 1248
16
276.80139.4472.1136.9920.30
2.865.68
10.9921.4239.03
1023.4519.46265.20138.1678.80
1.001.973.867.41
12.99
CRAY EL May 93 148
3832.81090.2764.1
0.210.731.04
NANANA
NANANA
CRAY J916 July 95 1248
16
1202.53608.83317.62168.4698.80
0.661.302.494.708.02
5356.592789.741414.43765.60426.91
0.190.370.721.342.40
CRAY T3D Feb 95 163264
128256512
1024
230.41115.5359.0129.9615.898.394.56
3.446.85
13.4326.4449.8794.45
173.77
918.04476.97252.86128.2168.3838.0120.45
1.112.154.047.9815.0
26.9250.04
-
41 - 53
CRAY T3E Nov 96 248
163264
128256
677.9339.8172.588.644.622.711.96.2
1.172.334.598.94
17.7734.9166.59
127.81
2988.71480.7735.6370.3192.299.954.529.3
0.340.691.392.765.32
10.2418.7834.93
CRAY T916 July 95 1248
16
193.19100.1053.2330.66
4.107.92
14.8925.84
649.10332.38169.2792.4364.06
1.583.086.05
11.0715.98
CRAY Y-MP Aug 92 18
792.4114.0
1.006.95
NANA
NANA
DEC Alpha Server8400 5/300(300 MHz)
Oct 95 1248
12
1048.7527.04271.13146.91103.47
0.761.502.925.397.66
4076.502525.001278.60649.53458.21
0.250.410.801.582.23
DEC Alpha Server8400 5/440(437 MHz)
Nov 96 1248
747.75376.03192.67103.54
1.062.114.117.65
3390.41727.7881.90454.39
0.300.591.162.25
Fujitsu VPP500 Mar 95 1248
161732345164
142.4275.1739.1419.829.99NA
5.09NANA
2.66
5.5610.5420.2539.9879.32
NA155.68
NANA
297.90
NANANANANA
37.26NA
18.8212.61
NA
NANANANANA
27.47NA
54.3881.16
NA
Fujitsu VX Nov 96 124
129.4068.51035.518
6.1211.5722.31
487.29258.20175.09
2.103.965.84
Table 9: Results of the BT simulated CFD application (BT)
benchmark.
-
42 - 53
Fujitsu VPP300 Nov 96 123468
16
129.4068.510
NA35.518
NA18.0889.2841
6.1211.57
NA22.31
NA43.8185.35
487.29258.20175.09
NA88.541
NANA
2.103.965.84NA
11.56NANA
Fujitsu VPP700 Nov 96 123468
16173234
129.4068.510
NA35.518
NA18.0889.2841
NA4.9286
NA
6.1211.57
NA22.31
NA43.8185.35
NA160.78
NA
487.29258.20175.09
NA88.541
NANA
31.634NA
16.824
2.103.965.84NA
11.56NANA
32.35NA
60.83
IBM SP-1 Aug 94 8163264
443.9249.2143.083.1
1.793.18 5.54 9.54
NA987.4511.2274.6
NA1.042.003.73
IBM RS/6000 SPWide-node1 (67 MHz)
Mar 95 8163264
128
206.7112.961.834.720.1
3.837.02
12.8222.8439.42
862.8440.6226.8119.167.0
1.192.324.518.59
15.27
IBM RS/6000 SPWide-node2 (77 MHz)
Oct 95 18
163264
1130.7170.7495.4851.3429.01
0.704.648.30
15.4427.31
4775.7708.47375.76197.91105.19
0.211.442.725.179.73
IBM RS/6000 SPP2SC node (120 MHz)
Nov 96 8163264
NANANANA
NANANANA
504.58263.77143.7776.49
2.033.887.12
13.38
IBM RS/6000 SPThin-node2 (67 MHz)
Feb 95 8163264
128
216.6118.064.936.320.8
3.666.72
12.2121.8338.10
889.8459.2237.2124.869.6
1.152.234.318.20
14.70
Table 9: Results of the BT simulated CFD application (BT)
benchmark.
-
43 - 53
Intel iPSC/860 Aug 92 64128
714.7414.3
1.11 1.91
NANA
NANA
Intel Paragon(OSF1.2)
Mar 94 64102128204256306408510512
235.0NA
129.0NA
83.0NANANA
63.0
3.37NA
6.14NA
9.55NANANA
12.58
NA633.0
NA359.0
NA257.0226.0196.0
NA
NA1.62NA
2.86NA
3.984.535.22NA
Intel Paragon(SunMos)
Nov 93Mar 94
64102128204306
224.0NA
113.0NANA
3.54NA
7.01NANA
NA598.0
NA324.0215.0
NA1.71NA
3.164.76
Kendall Square KSR1 Feb 94 3264
128
457256 145
1.743.1
5.46
NANANA
NANANA
Kendall Square KSR2 Feb 94May 94
3264
225130
3.526.10
NA542.0
NA1.89
Kyoto/MatsushitaADENART
Feb 94 256 314.1 2.52 NA NA
MasPar MP-1 Aug 92 4K 2396.0 0.33 NA NA
MasPar MP-2 Nov 92 4K 789.0 1.00 NA NA
Meiko CS-1 Aug 92 16 2984.0 0.27 NA NA
Meiko CS-2 Oct 94 81632
570.4 286.6 149.3
1.392.775.31
NANANA
NANANA
nCUBE-2S Mar 94
Jul 94
64128256512
1024
1243.2644.7336.7179.1100.9
0.641.222.354.427.85
NANANANANA
NANANANANA
NEC SX-3 Oct 94 1 100.31 7.90 NA NA
Table 9: Results of the BT simulated CFD application (BT)
benchmark.
-
44 - 53
NEC SX-4/32 Nov 96 1248
16172632
NANANANANANANANA
NANANANANANANANA
577.16288.72146.5975.39340.93235.57423.99624.042
1.773.546.98
13.5725.0028.7742.6542.57
Silicon GraphicsPower Challenge XL(75 MHz)
Oct 94 148
16
1330.3355.9177.091.8
0.602.234.488.63
5698.71450.0775.0426.0
0.180.711.322.40
SGIPower Challneg XL(90 MHz)
May 95 1248
16
1145.20574.37298.74152.6580.20
0.691.382.655.199.88
5089.602537.701278.60672.56391.88
0.200.400.801.522.61
SGI Origin200(180 MHz)
Nov 96 124
826.53420.08222.57
0.961.893.56
3953.702091.501069.20
0.260.490.95
SGI Origin2000(195 MHz)
Nov 96 1248
162531
768.42388.43196.13103.4155.44
NA30.05
1.032.044.047.66
14.29NA
26.37
3437.701767.00904.23468.77253.07151.87
NA
0.300.581.132.184.046.73NA
Thinking MachinesCM-2
Dec 91 16K 32K64K
1118.0634.0370.0
0.71 1.25 2.14
NANANA
NANANA
Thinking MachinesCM-200
Dec 91 16K 32K
832.0601.0
0.951.32
NANA
NANA
Thinking MachinesCM-5
May 93 3264
128
284.0175.0119.0
2.794.506.66
NANANA
NANANA
Table 9: Results of the BT simulated CFD application (BT)
benchmark.
-
45 - 53
Thinking MachinesCM-5E
Sep 95
Feb 94
163264
128
25913584.048.0
3.065.879.43
16.50
1480 712
464.0253.0
0.691.442.214.05
Thinking MachinesCM-500
Sep 95 64128256
7543 27
10.5618.4329.34
370209114
2.774.908.98
Table 9: Results of the BT simulated CFD application (BT)
benchmark.
-
46 - 53
Table 10: Results for NPB 1.0 Class C on CRAY T3E.
No. ofCPUs
Wall Clock Time in Seconds
EP MG CG FT IS LU SP BT
1 NA NA NA NA NA NA NA NA
4 448.8 NA NA NA NA NA NA NA
8 224.4 NA NA NA NA NA NA NA
16 112.2 NA 272.9 NA 48.9 773.4 1352.7 1806.5
32 56.1 53.0 146.2 93.4 23.8 387.8 661.5 904.3
64 28.1 26.7 78.5 47.2 14.5 195.3 327.0 454.4
128 14.0 13.6 51.7 23.7 7.5 101.5 174.3 230.6
256 7.0 7.1 46.9 12.1 4.0 52.8 93.9 115.4
-
47 - 53
Table 11: Approximate sustained performance per dollar for Class
B LU benchmark..
Computer System
#
Proc.
Memory Ratioto
C90/1
List PriceMillionDollars
Performanceper Million
Dollars
Date
CRAY T3E 128 64 MB/PE 15.24 5.0 3.05 Nov 96
DEC Alpha Server8400 5/440 (437 MHz)
8 2 GB 1.66 0.58 2.86 Nov 96
Fujitsu VX 3 2 GB /PE 3.25 1.11 2.93 Nov 96
Fujitsu VPP300 6 512 MB /PE 5.76 1.54 3.74 Nov 96
Fujitsu VPP700 17 512 MB/PE 12.24 5.17 2.37 Nov 96
IBM RS/6000 SPP2SC node (120 MHz)
64 128 MB/PE 10.48 3.52 2.98 Nov 96
NEC SX-4/32 32 4 GB 21.50 10.7 2.0 Nov 96
SGI Origin2000(195 MHz)
26 2 GB 5.01 0.96 5.21 Nov 96
-
48 - 53
Table 12: Approximate sustained performance per dollar for Class
B SP benchmark..
Computer System
# Proc Memory Ratioto
C90/1
List PriceMillionDollars
Performanceper Million
Dollars
Date
CRAY T3E 128 64 MB/PE 14.34 5.0 2.87 Nov 96
DEC Alpha Server8400 5/440 (437 MHz)
8 2 GB 1.90 0.58 3.28 Nov 96
Fujitsu VX 4 512 MB/PE 6.03 1.0 6.03 Nov 96
Fujitsu VPP300 8 512 MB/PE 11.73 1.99 5.89 Nov 96
Fujitsu VPP700 34 512 MB/PE 47.17 9.98 4.73 Nov 96
IBM RS/6000 SPP2SC node (120 MHz)
64 128 MB/PE 11.82 3.52 3.36 Nov 96
NEC SX-4/32 32 4 GB 40.26 10.7 3.76 Nov 96
SGI Origin2000(195 MHz)
26 2 GB 4.91 0.96 5.11 Nov 96
-
49 - 53
Table 13: Approximate sustained performance per dollar for Class
B BT benchmark..
Computer System
#Proc
Memory Ratioto
C90/1
List PriceMillionDollars
Performance per Million
Dollars
Date
CRAY T3E 128 64 MB/PE 18.78 5.0 3.76 Nov 96
DEC Alpha Server8400 5/440 (437 MHz)
8 2 GB 2.25 0.58 3.88 Nov 96
Fujitsu VX 3 512 MB/PE 5.84 1.11 5.26 Nov 96
Fujitsu VPP300 6 512 MB/PE 11.56 1.54 7.51 Nov 96
Fujitsu VPP700 34 512 MB/PE 60.83 9.98 6.1 Nov 96
IBM RS/6000 SPP2SC node (120 MHz)
64 128 MB/PE 13.38 3.52 3.80 Nov 96
NEC SX-4/32 32 4 GB 42.57 10.7 3.98 Nov 96
SGI Origin2000(195 MHz)
26 2 GB 6.73 0.96 7.01 Nov 96
-
50 - 53
Table 14: Results of the HPF based EP benchmark.
Computer System DateReceived
NumberProcessor
Time in seconds
Class A Class B
APR PGI APR PGI
HP/ConvexExemplar SPP200
Oct 96 1248
16
NANANANANA
6172891457539
NANANANANA
NANANANANA
IBM SP2Wide Nodes
Sept 96 1248
163264
NANANANA794223
947473237119593015
NANANANANANANA
NA189594647423711959
SGI/CRAY T3D Sept 96 1248
163264
NANANANA864322
1190596302151743719
NANANANANANANA
478223981190595298149NA
-
51 - 53
Table 15: Results of the HPF based MG benchmark.
Computer System DateReceived
NumberProcessor
Time in seconds
Class A Class B
APR PGI APR PGI
IBM SP2Wide Nodes
Sept 96 163264
1297
NANANA
NANANA
NANANA
SGI/CRAY T3D Sept 96 163264
643725
NANANA
NANANA
NANANA
Table 16: Results of the HPF based FT benchmark.
Computer System DateReceived
NumberProcessor
Time in seconds
Class A Class B
APR PGI APR PGI
IBM SP2Wide Nodes(67 mHz)
Sept 96 163264
NANANA
6.23.12.6
NANANA
NANANA
SGI/CRAY T3D Sept 96 163264
128
NANANANA
27.215.69.55.6
NANANANA
NANANA
19.6
-
52 - 53
Table 17: Results of the HPF based SP benchmark.
Computer System DateReceived
NumberProcessor
Time in seconds
Class A Class B
APR PGI APR PGI
HP/ConvexExemplar SPP200
Oct 96 1248
16
NANANANANA
38261908966504274
NANANANANA
NANANANANA
IBM SP2Wide Nodes(67 MHz)
Sept 96 163264
315207130
23714299
NANANA
NANANA
SGI/CRAY T3D Sept 96 8163264
NA803420226
1618834425228
NANANANA
NA332719501012
-
53 - 53
Table 18: Results of the HPF based BT benchmark.
Computer System DateReceived
NumberProcessor
Time in seconds
Class A Class B
APR PGI APR PGI
HP/ConvexExemplar SPP200
Oct 96 1248
16
NANANANANA
23721161699335175
NANANANANA
NANANANANA
IBM SP2Wide Nodes(67 MHz)
Sept 96 1248
163264
NANANANA332184109
3273169790746924913581
NANANANANANANA
NANA
378819631111652350
SGI/CRAY T3D Sept 96 8163264
NA1044536275
1669855438226
NANANANA
6795363421191096