Digital PDFs

EB-N0104-51

November 2004

31 pages

Original

0.5MB

Document:	Alpha AXP Server Family Performance Brief - OpenVMS
Order Number:	EB-N0104-51
Revision:
Pages:	31
Original Filename:	https://web-docs.gsi.de/~kraemer/COLLECTION/DEC/alpha-axp-server.pdf

OCR Text

Alpha AXP Server Family
Performance Brief - OpenVMS

DEC 10000 Model 610 AXP Server

DEC 7000 Model 610 AXP Server

DEC 4000 Model 610 AXP Server

DEC 3000 Model 500S AXP Server

DEC 3000 Model 400S AXP Server

INSIDE
Benchmark results:

Digital Equipment Corporation
November 20, 1992
Second Edition
EB-N0104-51

• SPEC
• Perfect
• LINPACK
• Dhrystone
• CERN
• ANSYS

• X-PLOR
• CHARMm
• DN&R Labs
CPU2
• SLALOM

Alpha AXP Server Family
Performance Brief - OpenVMS
Introducing Digital’s Alpha AXP Server Family
This document presents the results of industry-standard benchmarks on Digital’s Alpha
AXPTM servers in the DEC OpenVMS AXPTM operating system environment. Alpha AXP
servers are based on Digital’s breakthrough 64-bit RISC architecture.
DEC 3000 AXP Servers
The DEC 3000 Model 400S AXP desktop is an entry-level server or multiuser system that
provides affordable power for small workgroup computing and distributed applications. The
DEC 3000 Model 500S AXP server is a deskside system that provides a blend of
performance and expansion to meet the challenges of the most demanding workgroup
computing. The DEC 3000 Model 400S AXP server runs at a CPU clock speed of 133 MHz,
and the DEC 3000 Model 500S AXP server at 150 MHz.
DEC 4000 AXP Servers
DEC 4000 AXP servers deliver superior computing power in office-sized or rackmountable
packages. Designed for symmetric multiprocessing, multiuser and technical server
applications, DEC 4000 AXP servers meet the needs of technical and scientific markets.
Optimized for speed and availability, DEC 4000 AXP servers will satisfy the commercial
segment as well. The DEC 4000 AXP server runs at a CPU clock speed of 160 MHz and can
be configured with one or two CPU boards.
DEC 7000 AXP Servers
DEC 7000 AXP servers deliver unsurpassed data center performance and price/performance.
DEC 7000 AXP servers offer single or multiprocessing capability for such commercial and
technical applications as transaction processing, general ledger, securities trading, signal
processing, molecular modeling, and imaging. The DEC 7000 AXP server runs at a CPU
clock speed of 182 MHz and can be configured with up to six processors.
DEC 10000 AXP Servers
DEC 10000 AXP servers are Digital’s highest performance Alpha AXP systems. Designed
to meet rigorous demands of compute-intensive, large enterprise applications, DEC 10000
AXP servers can be configured with up to six processors. The DEC 10000 AXP server
provides the fastest enterprise server performance in the industry—at a fraction of the price
of a traditional mainframe or supercomputer. The DEC 10000 AXP server runs at a CPU
clock speed of 200 MHz.

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
Table 1 Digital’s Alpha AXP Server Family Benchmark Results

Benchmark

DEC 3000
Model
400S AXP

DEC 3000
Model
500S AXP

DEC 4000
Model
610 AXP

DEC 7000 DEC 10000
Model
Model
610 AXP
610 AXP

SPECmark89

108.1

121.5

136.2

167.4

184.1

SPECthruput89 - 1 cpu
- 2 cpu
- 3 cpu
- 4 cpu

na
na
na
na

131.2
248.8
na
na

161.4
308.7
454.4
604.4

tbd
tbd
tbd
654.6

SPECint92
SPECfp92

65.3
112.2

74.3
126.0

83.5
143.1

96.6
182.1

106.5
200.4

SPECrate_int92 - 1 cpu
- 2 cpu
- 3 cpu
- 4 cpu

1549.3
na
na
na

1762.1
na
na
na

1985.8
3816.1
na
na

2188.6
4291.4
6306.5
8366.8

tbd
tbd
tbd
9107.9

SPECrate_fp92 - 1 cpu
- 2 cpu
- 3 cpu
- 4 cpu

2631.6
na
na
na

2967.4
na
na
na

3317.1
6214.5
na
na

4126.0
8135.1
11859.8
15739.4

tbd
tbd
tbd
17187.2

Dhrystone
V1.1 (instructions/sec) 228,310.0
V2.1 (instructions/sec) 249,625.6

257,731.0
281,214.8

279,329.0
303,951.4

311,526.0
343,642.6

342,465.0
377,358.5

LINPACK 64-bit Double-precision
100x100 (MFLOPS)
26.4
1000x1000 (MFLOPS) 70.8

30.2
79.9

36.3
86.4

38.6
102.1

42.5
111.6

CERN Benchmark Suite

16.9

19.0

21.0

23.6

26.0

Perfect Benchmarks Suite (MFLOPS)
(geometric mean)
18.1

20.4

22.9

26.0

28.6

DN&R Labs CPU2 (MVUPs)

179.4

197.5

216.5

247.5

272.7

SLALOM (patches)

3750

3934

4072

4432

4608

Dhrystone scales linearly
na = not applicable
tbd = to be determined

November 20, 1992

Digital Equipment Corporation

Alpha AXP Server Family
Performance Brief - OpenVMS
SPEC Benchmark Suites
SPEC (Standard Performance Evaluation Corporation) was formed to identify and create
objective sets of applications-oriented tests, which can serve as common reference points and
be used to evaluate performance across multiple vendors’ platforms.
SPEC Release 1
In October 1989, SPEC introduced SPEC Release 1, a benchmark suite that measures
CPU-intensive, single stream performance of uniprocessor systems. SPEC Release 1
consists of ten portable programs similar to those found in technical environments. Four
programs are written in C and primarily test integer performance. The remaining six
programs are written in FORTRAN and measure floating-point performance.
SPECmark89 is the metric for this suite.
SPECmark89 represents the geometric mean of the ten benchmark SPECratios. The
SPECratio for a benchmark is the quotient derived from dividing the SPEC Reference Time
by a particular machine’s corresponding run time. The SPEC Reference Time is the time
that it takes a DEC VAX 11/780 to run each benchmark (in seconds).

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
Figure 1 SPECmark89 Benchmark Results for Alpha AXP Servers

SPECmark89
200.0

184.1
167.4

150.0

136.2
121.5
108.1

100.0

50.0

0.0
DEC 3000/400S AXP
DEC 4000/610 AXP
DEC 10000/610 AXP
DEC 3000/500S AXP
DEC 7000/610 AXP

Figure 2 SPECmark89 Benchmark Results for Competitive Systems

SPECmark89

200.0

146.8

150.0
136.1

126.2

100.0

87.6

126.3

92.3

86.6
73.7

69.0

50.0

0.0
HP 9000/730

HP 9000/887S
HP 9000/750

November 20, 1992

HP 9000/735/755

HP 9000/715/50

IBM RS 6000/560

IBM RS 6000/350

Digital Equipment Corporation

IBM RS 6000/980

IBM RS 6000/580

Alpha AXP Server Family
Performance Brief - OpenVMS
SPECthruput89
SPEC devised the SPECthruput89 benchmark to measure the performance of multiprocessor
systems. SPEC defines the SPECthruput89 benchmark as a measure of the amount of work,
given a particular workload, that a system can perform compared to a reference system. In
this case, the time posted by a VAX 11/780 concurrently running two copies of each SPEC
Release 1 benchmark is the reference time. SPECthruput89 is the metric for this benchmark.
SPECthruput89 is calculated as follows. The system under test concurrently runs two copies
of the same SPEC Release 1 benchmark per processor. Individual "thruput" ratios are
calculated by comparing these run times with the reference system’s run times for the same
benchmark. SPECthruput89 is the geometric mean of all the individual thruput ratios
multiplied by the number of processors in the system under test.

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
Figure 3 SPECthruput89 Benchmark Results for Alpha AXP Servers

SPECthruput89

700.0

654.6
604.4

600.0
500.0

454.4

400.0
308.7

300.0
248.8
200.0

161.4
131.2

100.0
0.0
DEC 4000/610 AXP

DEC 7000/610 AXP
DEC 4000/620 AXP

DEC 7000/630 AXP
DEC 7000/620 AXP

DEC 10000/640 AXP
DEC 7000/640 AXP

Figure 4 SPECthruput89 Benchmark Results for Competitive Systems

SPECthruput89

700.0

600.0

500.0

400.0

300.0
218.0

200.0
127.7

95.3

100.0

109.0

67.3
35.4

0.0
HP 9000 870/100

HP 9000 870/300
HP 9000 870/200

November 20, 1992

SUN SPARC 690 MP 52
HP 9000 870/400

Digital Equipment Corporation

SUN SPARC 690 MP 54

Alpha AXP Server Family
Performance Brief - OpenVMS
SPEC CINT92 and CFP92
In January 1992, SPEC announced the availability of two new benchmark suites, CINT92
and CFP92. Each suite provides performance indicators for different market segments
because each has different workload characteristics. SPEC CINT92 is a good base indicator
of CPU performance in a commercial environment. SPEC CFP92 may be used to compare
floating-point intensive environments, typically engineering and scientific applications.
SPEC CINT92
CINT92, the integer suite, contains six real-world application benchmarks written in C. The
geometric mean of the suite’s six SPECratios is the SPECint92 figure. CINT92 suite
includes the following application classes:
•

008.espresso–Circuit theory

•

022.li–LISP Interpreter

•

023.eqntott–Logic design

•

026.compress–Data compression

•

072.sc–UNIX spreadsheet

•

085.gcc–GNU C compiler

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS

Figure 5 SPEC CINT92 Benchmark Results for Alpha AXP Servers
SPECint92

120.0

106.5
100.0

96.6

83.5

80.0

74.3

65.3

60.0

40.0

20.0

0.0

DEC 3000/400S AXP

DEC 3000/500S AXP

DEC 4000/610 AXP

DEC 7000/610 AXP

DEC 10000/610 AXP

Figure 6 SPEC CINT92 Benchmark Results for Competitive Systems
SPECint92
120.0

100.0

80.0

78.2

59.1

60.0
52.0

53.3

51.1
45.6

40.0

36.5

43.2
35.4

34.8
31.4

20.0

0.0
HP 9000/730

HP 9000/715/50
HP 9000/750

November 20, 1992

HP 9000/827S

HP 9000/735/755

HP 9000/867S

HP 9000/847S

IBM RS 6000/350

HP 9000/887S

Digital Equipment Corporation

IBM RS 6000/580/980

IBM RS 6000/560

SUN SPARC 10-41

Alpha AXP Server Family
Performance Brief - OpenVMS
SPEC CFP92
CFP92 consists of fourteen real-world applications; two are written in C and twelve in
FORTRAN. Five of the fourteen programs are single precision, and the rest are double
precision. SPECfp92 equals the geometric mean of this suite’s fourteen SPECratios. This
suite contains the following application classes:
•

013.spice2g6–Circuit design

•

015.doduc–Monte Carlo simulation

•

034.mdljdp2−Quantum chemistry

•

039.wave5–Maxwell equations

•

047.tomcatv–Coordinate translation

•

048.ora–Optics ray tracing

•

052.alvinn–Robotics; neural nets

•

056.ear–Human ear modeling

•

077.mdljsp2–Single precision version of 034.mdljdp2

•

078.swm256–Shallow water model

•

089.su2cor–Quantum physics

•

090.hydro2d–Astro physics

•

093.nasa7–NASA math kernels

•

094.fppp–Quantum chemistry

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
Figure 7 SPEC CFP92 Benchmark Results for Alpha AXP Servers
SPECfp92
250.0

200.4

200.0
182.1

150.0

143.1
126.0
112.2

100.0

50.0

0.0
DEC 3000/400S AXP

DEC 4000/610 AXP
DEC 3000/500S AXP

DEC 10000/610 AXP
DEC 7000/610 AXP

Figure 8 SPEC CFP92 Benchmark Results for Competitive Systems
SPECfp92

250.0

200.0

150.6

150.0

141.6
124.7
109.0

100.0

97.6

86.7

84.8
76.6

74.2

72.1

65.1

50.0

0.0
HP 9000/730

HP 9000 715/50
HP 9000/750

November 20, 1992

HP 9000/887S

HP 9000/735/755

IBM RS 6000/560

IBM RS 6000/350

SUN SPARC 10-41

IBM RS 6000/580/980

Digital Equipment Corporation

SUN SPARC 2000-8

SUN SPARC 2000-2

Alpha AXP Server Family
Performance Brief - OpenVMS
SPEC Homogeneous Capacity Method based on SPEC CINT92 and CFP92
SPEC Homogeneous Capacity Method benchmarks test multiprocessor efficiency.
According to SPEC, "The SPEC Homogeneous Capacity Method provides a fair measure for
the processing capacity of a system — how much work can it perform in a given amount of
time. The "SPECrate" is the resulting new metric, the rate at which a system can complete
the defined tasks....The SPECrate is a capacity measure. It is not a measure of how fast a
system can perform any task; rather it is a measure of how many of those tasks that system
completes within an arbitrary time interval (SPEC Newsletter, June 1992)." The SPECrate is
intended to be a valid and fair comparative metric to use across systems of any number of
processors.
The following formula is used compute the SPECrate:
SPECrate = #CopiesRun * ReferenceFactor * UnitTime / ElapsedExecutionTime
SPECrate_int92 equals the geometric mean of the SPECrates for the six benchmarks in
CINT92. SPECrate_fp92 is the geometric mean of the SPECrates of the fourteen
benchmarks in CFP92.

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
Figure 9 SPECrate_int92 Results for Alpha AXP Servers

SPECrate_int92
10000.0
9107.9
8366.8
8000.0

6306.5
6000.0

4291.4
3816.1

4000.0

2000.0

2188.6

1985.8

1762.1

1549.3

0.0
DEC 3000/400S AXP

DEC 4000/610 AXP

DEC 3000/500S AXP

DEC 7000/610 AXP

DEC 4000/620 AXP

DEC 7000/630 AXP

DEC 7000/620 AXP

DEC 10000/640 AXP

DEC 7000/640 AXP

Figure 10 SPECrate_int92 Results for Competitive Systems
SPECrate_int92
10000.0

8047.0

8000.0

6000.0

4301.0
4000.0
3306.0
2326.0

2253.0
2000.0
816.0

890.0

1263.0

1215.0

1201.0

0.0
HP 9000/827S

HP 9000/867S
HP 9000/847S

November 20, 1992

HP 9000/890-2
HP 9000/890-1

HP 9000/890-3

HP 9000/890-4
SUN SPARC 2000-2
SUN SPARC 10-41
SUN SPARC 2000-8

Digital Equipment Corporation

Alpha AXP Server Family
Performance Brief - OpenVMS

Figure 11 SPECrate_fp92 Benchmark Results for Alpha AXP Servers
SPECrate_fp92

20000.0
17187.2
15739.4

15000.0
11859.8
10000.0
8135.1
6214.5

5000.0

4126.0
2631.6

2967.4

3317.1

0.0
DEC 3000/400S AXP
DEC 4000/610 AXP
DEC 7000/610 AXP
DEC 7000/630 AXP
DEC 10000/640 AXP
DEC 3000/500S AXP
DEC 4000/620 AXP
DEC 7000/620 AXP
DEC 7000/640 AXP

Figure 12 SPECrate_fp92 Benchmark Results for Competitive Systems

SPECrate_fp92
20000.0

15000.0

10600.0
10000.0

4685.0

5000.0
3529.0
1335.0

1483.0

2953.0

2360.0

1937.0

1544.0

1180.0

0.0
HP 9000/827S

HP 9000/867S
HP 9000/847S

HP 9000/890-2
HP 9000/890-1

HP 9000/890-4
HP 9000/890-3

Digital Equipment Corporation

SUN SPARC 2000-2

SUN SPARC 10-41

SUN SPARC 2000-8

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
Perfect Benchmarks Suite
The Perfect (Performance Evaluation for Cost-effective Transformations) Benchmarks
Suite represents an ongoing effort among several universities, research centers, and
supercomputing firms to produce a benchmark package oriented to supercomputers and
parallel processing. Currently, 13 FORTRAN programs totaling over 50,000 lines of code
make up the Perfect Benchmark Suite. These programs include scientific and engineering
applications representing four types of real applications areas: fluid flow, chemical and
physical, engineering design, and signal processing.
Perfect Benchmark Suite’s results are measured in millions of floating-point operations per
second (MFLOPS). Figure 13 shows the geometric mean of the suite’s results in MFLOPS.
Note: Comparable results from other vendors were not available.
Figure 13 PERFECT Benchmark Results for Alpha AXP Servers

MFLOPS
35.0
30.0

28.6
26.0

25.0

22.9
20.4

20.0

18.1

15.0
10.0
5.0
0.0
DEC 3000/400S AXP
DEC 4000/610 AXP
DEC 10000/610 AXP
DEC 3000/500S AXP
DEC 7000/610 AXP

November 20, 1992

Digital Equipment Corporation

Alpha AXP Server Family
Performance Brief - OpenVMS
LINPACK 100x100 and 1000x1000
LINPACK is a linear equation solver written in FORTRAN. LINPACK programs consist of
floating-point additions and multiplications of matrices. The LINPACK benchmark suite
consists of two benchmarks.
1.

100x100 LINPACK solves a 100x100 matrix of simultaneous linear equations.
Source code changes are not allowed so that the results may be used to evaluate the
compiler’s ability to optimize for the target system.

1000x1000 LINPACK solves a 1000x1000 matrix of simultaneous linear equations.
Vendor optimized algorithms are allowed.

The LINPACK benchmarks measure the execution rate in MFLOPS (millions of
floating-point operations per second). When running, the benchmark depends on
memory-bandwidth and gives little weight to I/O. Therefore, when LINPACK data fit into
system cache, performance may be higher.

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
Figure 14 LINPACK Double-Precision Benchmark Results for Alpha AXP
Servers
MFLOPS
180.0
100x100
160.0
140.0

1000x1000

120.0

111.6
102.1

100.0
86.4
79.9

80.0

70.8

60.0
40.0

42.5

38.6

36.3

30.2

26.4
20.0
0.0
DEC 3000/400S AXP

DEC 4000/610 AXP
DEC 3000/500S AXP

DEC 10000/610 AXP
DEC 7000/610 AXP

Figure 15 LINPACK Double-Precision Benchmark Results for Competitive
Systems
MFLOPS

180.0
100x100

160.0
140.0

159.6

1000x1000

120.0
100.0

84.0
80.0
61.0

60.0

46.7

40.8

40.0

38.1
31.3

23.7

20.0

13.2
N/A

13.7
N/A

N/A

N/A N/A

17.1

22.4

18.8

N/A N/A

N/A

0.0
HP 9000/730/750
HP 9000/735/755
HP 9000/867S
IBM RS 6000/560
SUN SPARC 10-41
SUN SPARC 2000-8
HP 9000/715/50
HP 9000/847S
IBM RS 6000/350
IBM RS 6000/580/980
SUN SPARC 2000-2

November 20, 1992

Digital Equipment Corporation

Alpha AXP Server Family
Performance Brief - OpenVMS
Dhrystone
Developed as an Ada program in 1984, the Dhrystone benchmark was rewritten in C in
1986. Dhrystone measures processor and compiler efficiency and is representative of
systems programming environments. Dhrystones are most commonly expressed in
Dhrystone instructions per second or in integer MIPS (millions of instructions per second).
For V1.1, one Dhrystone MIP equals the number of Dhrystone instructions per second
performed by a DEC VAX 11/780 (1757 Dhrystone instructions/second).
Dhrystone V1 and V2 vary considerably. Version 1.1 contains sequences of code segments
that calculate results never used later in the program. These code segments are known as
"dead code." Compilers able to identify the dead code can eliminate these instruction
sequences from the program. These compilers allow a system to complete the program in
less time and result in a higher Dhrystones rating. Dhrystones V2 was modified to execute
all instructions.
Note: The Dhrystone benchmark is small and fits completely in most system caches. Level
of compiler optimization, as well as the particular hardware architecture, can affect results.
Additionally, Dhrystone scales linearly. For example, use 311,526x2 for DEC 7000 AXP
server with 2 CPUs, use 311,526x3 for DEC 7000 AXP server with 3 CPUs, etc.

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS

Figure 16 Dhrystone V1.1 Benchmark Results for Alpha AXP Servers
Dhrystones/second
(in thousands)
400.0
377.4

V1.1
343.6

350.0
V2.1

311.5

304.0

300.0

281.2

279.3

257.7

249.6
228.3

250.0

342.5

200.0
150.0
100.0
50.0

0.0
DEC 3000/400S AXP

DEC 3000/500S AXP

DEC 4000/610 AXP

DEC 7000/610 AXP

DEC 10000/610 AXP

Figure 17 Dhrystone V1.1 Benchmark Results for Competitive Systems
Dhrystones/second
(in thousands)
400.0

V1.1
350.0

300.0
250.0

215.5
194.2

200.0
150.0

133.7
133.7
122.3
121.6

100.0

109.4
97.0

192.4

123.8
92.6

131.6
106.5

50.0

N/A

0.0

N/A

HP 9000/730
HP 9000/715/50
HP 9000/827S/847S IBM RS 6000/350
SUN SPARC 10-41
HP 9000/750
HP 9000/735/755
HP 9000/867S
IBM RS 6000/560

November 20, 1992

Digital Equipment Corporation

Alpha AXP Server Family
Performance Brief - OpenVMS
CERN Benchmark Suite
In the late 1970’s the User Support Group at CERN, the European Laboratory for Particle
Physics, collected from different experimental groups a set of typical programs for event
simulation and reconstruction and created the CERN Benchmark Suite. In 1985, Eric
McIntosh, system analyst, redefined the tests in order to make them more portable and more
representative of the then current workload and FORTRAN 77.
Presently, the CERN Benchmark Suite contains four production tests: two event generators
(CRN3 and CRN4) and two event processors (CRN5 and CRN12). These applications are
basically scalar and are not significantly vectorizable nor numerically intensive.
Additionally, several "kernel" type applications were added to supplement the production
tests to get a feel for compilation times (CRN4C), vectorization (CRN7 and CRN11), and
character manipulation (CRN6).
The CERN Benchmark Suite metric is CPU time. Results are normalized to a DEC VAX
8600, and the geometric mean of the four production tests’ ratios yields the number of CERN
units. CERN units increase with increasing performance.
Note: Alpha AXP numbers shown in Figure 18 are not official CERN statistics.

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
Figure 18 CERN Benchmark Results for Alpha AXP Servers
CERN Units

30.00
26.00

25.00

23.60
21.00

20.00

19.00
16.90

15.00
10.00

5.00
0.00
DEC 3000/400S AXP
DEC 4000/610 AXP
DEC 10000/610 AXP
DEC 3000/500S AXP
DEC 7000/610 AXP

Figure 19 CERN Benchmark Results for Competitive Systems
CERN Units

30.00

25.00

20.00

15.30

15.00

10.00

13.60

13.50

9.11

10.70

9.96

8.48

4.49

5.00

3.49

0.00
HP 9000/710
HP 9000/720

HP 9000/750
IBM RS 6000/970
SUN SPARC 10
SUN SPARC IPX
IBM RS 6000/550
SGI IRIX (R4000)
SUN MP 4/670

Source: "CERN Report" (11/92).

November 20, 1992

Digital Equipment Corporation

Alpha AXP Server Family
Performance Brief - OpenVMS
ANSYS BENCHMARKS
Principal applications used for mechanical engineering range from drafting and detailing for
the design of products, to NC machining for producing parts, to finite element analysis to
understand the behavior of the product as it is intended to be used. These applications
require a wide range of processing capabilities, including high speed computation and
graphics. Digital chose the ANSYS software to demonstrate the superiority of the Alpha
AXP servers because ANSYS is representative of the requirements of applications in the
mechanical engineering environment.
SASI’s (Swanson Analysis Systems Inc.) ANSYS software is a leadership finite element
pre/post processing and analysis application used extensively by the aerospace, automotive,
and other industries. SASI provides a set of benchmarks to system vendors, and
measurements are typically made by the system vendors and validated by SASI.
Digital ran the ANSYS benchmark tests described below:
•

SP1 - Modal analysis of a flat plate.

•

SP2 - Static analysis of a weight dropped from 40 feet.

•

SP3 - Static analysis of a pressure vessel.

•

SP4 - Thermal and stress analysis of a curved plate.

ANSYS benchmark results are reported in CPU time and elapsed time required to complete
each analysis. The timings are reported from an internal mechanism in the ANSYS software
which keeps track of this data. Smaller numbers represent better performance.

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
Table 2 ANSYS Benchmark Results for Alpha AXP Servers (in seconds)
SP1

SP2

SP3

SP4

CPU elapsed

DEC 10000 Model 610 AXP

DEC 7000 Model 610 AXP

DEC 4000 Model 610 AXP

DEC 3000 Model 500S AXP

Benchmarks ran on Alpha AXP servers running DEC OpenVMS AXP version FT4. Results obtained using a preliminary copy
of the ANSYS V4.4A software product built with DEC FORTRAN for Alpha AXP version BL24. All results for Alpha AXP
servers obtained using DECRAM, a layered product which uses the Alpha AXP memory subsystem for file I/O.

Table 3 ANSYS Benchmark Results for Competitive Systems (in seconds)
SP1

SP2

SP3

SP4

CPU elapsed

HP 9000 Model 750

IBM RS 6000 Model 580

SUN SPARC 10 Model 30

111

114

IBM and HP benchmark results obtained from "ANSYS Benchmark Timing Results" report (7/15/92). SUN SPARC 10 Model
30 results obtained in Digital Equipment Corporation’s labs, due to no currently reported data from SASI.

November 20, 1992

Digital Equipment Corporation

Alpha AXP Server Family
Performance Brief - OpenVMS
X-PLOR AND CHARMm BENCHMARKS
X-PLOR
X-PLOR is a widely used program that helps automate a critical portion of the task of
structural refinement of molecules. This task arises in structure determination work,
including both X-ray crystallography and Nuclear Magnetic Resonance solution structure
studies. The goal is to make an approximate or trial structure fit the experimental data as
well as possible.
X-PLOR builds a trial structure by taking a model of the energy of the structure and adds to
it a pseudo energy representing the difference between the real data and what the data would
be for the current structure. X-PLOR pours in a lot of energy, heating up the molecule to a
high temperature (like 1000 centigrade degrees) which is physically impossible, but quite
useful for a simulated study. X-PLOR runs molecular dynamics for a short time at this high
temperature, which lets the molecule try out and explore many alternate structures. Sample
structures are saved and later "cooled" back to normal temperature. The set of final
structures are analyzed to see whether there is a consensus structure, distinct families of
structures, or something else. This structure refinement step has been greatly improved and
automated by X-PLOR.
Task 1 of the benchmark is the molecular dynamics study, which is the fundamental
operation to "explore" possible structural alternatives. It might be done several times for one
problem and, of course, will be repeated for each new problem. Tasks 2, 3, and 4 need to be
completed to look at and analyze the final results.
The four tasks used in the benchmark are:
•

Task 1: 0.5 picosecond Molecular Dynamics simulation on a DNA-undodecamer.

•

Task 2: Electron density calculation at 2.8 A for 3086 atoms.

•

Task 3: 3-D real Fast Fourier Transformation on a 180x96x96 grid.

•

Task 4: Application of four symmetry operators in reciprocal space to 8783 reflections.

Test 1 consists of Task 1, and Test 2 is the cumulative time spent in Tasks 2, 3, and 4.

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
Table 4 X-PLOR V3.0 Benchmark Results for Alpha AXP Servers (in seconds)

CPU

Test 1
elapsed

CPU

Test 2
elapsed

DEC 10000 Model 610 AXP

89.6

22.0

DEC 7000 Model 610 AXP

94.3

24.1

DEC 4000 Model 610 AXP

109.8

111

31.3

DEC 3000 Model 500S AXP

119.4

121

32.6

November 20, 1992

Digital Equipment Corporation

Alpha AXP Server Family
Performance Brief - OpenVMS
CHARMm
CHARMm is one of a group of programs that scientists use to study matter at an atomic
level of description. These programs calculate the energy of a molecule or a group of
molecules from the positions of all the atoms. CHARMm benchmarks represent one
common use of CHARMm called molecular dynamics. This is the calculation of the time
evolution of the system to see how it vibrates and jiggles about due to its thermal energy.
The CHARMm benchmarks consist of two tests. In "dtest" a fragment of an immune-system
protein is simulated, while in "ptest" it’s a piece of a polystyrene chain. Each molecule
contains roughly 1,000 atoms. This simulation runs for 1,000 steps, which amounts to one
picosecond—that’s one millionth of a microsecond or one thousandth of a nanosecond. In
an actual calculation, the simulation would typically be run from 30 picoseconds to a
nanosecond or more. Therefore, these benchmarks represent short versions of real jobs on
real, typical molecules.
Both benchmarks are straightforward molecular dynamics. For each one, there’s a "fast" and
a "standard" version; this refers to internal approximations that may take advantage of
hardware on some machines. Times are CPU usage measured in minutes.
Table 5 CHARMm 21.3 Benchmark Results for Alpha AXP Servers (in
minutes)
ptest

dtest

Test1-STD

Test1-Fast

Test1-STD

Test1-Fast

CPU elapsed

DEC 4000 Model 610 AXP

2.5

1.5

2.8

2.9

1.8

1.9

DEC 3000 Model 500S AXP

2.7

2.8

1.6

3.1

3.2

2.0

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
Table 6 lists other manufacturers’ CHARMm results normalized to the DEC 3000 Model
500S AXP and DEC 4000 Model 610 AXP servers. The range indicates the performance
variation between the slowest and fastest performance numbers obtained from all tests
executed on each platform. A range gives an accurate representation of the performance
since a user is likely to alternate between the two modes while working on different problems
or phases of problems.
Table 6 CHARMm Benchmark Results Normalized to DEC 3000 Model 500S
AXP and DEC 4000 Model 610 AXP Servers

Manufacturer
and Model

DEC 3000 Model 500S AXP
DEC 4000 Model 610 AXP

_________________Normalized to
DEC 3000 Model 500S AXP
DEC 4000 Model 610 AXP
Standard
Fast
Standard
Fast

1.00

1.00
1.00

1.00

HP 9000 Model 720

.60 to .61

.47 to .56

.55 to .56

.43 to .50

IBM RS 6000 Model 320
IBM RS 6000 Model 530
IBM ES/9000 (AIX/370)
IBM ES/9000 (VM/CMS)
IBM ES/9000 (MVS/ESA)

.30 to .31
.37 to .39
.38 to .39
.38 to .39
.38 to .39

.23 to .26
.29 to .33
.37 to .40
.36 to .40
.35 to .40

.27 to .28
.34 to .35
.35 to .36
.35 to .36
.35 to .35

.21 to .23
.27 to .30
.33 to .36
.33 to .36
.33 to .36

SGI Indigo (33 MHz R3000)
SGI Crimson (50 MHz, R4000)
SGI Indigo (50 MHz, R4000)

.18 to .21
.42 to .44
.42 to .43

.14 to .15
.32 to .36
.32 to .36

.17 to .19
.39 to .40
.38 to .39

.13 to .14
.30 to .33
.29 to .32

Stardent Titan

.12 to .13

.15 to .15

.11 to .12

.14 to .14

Audited by Bruce Gelin of Custom Research and Consulting, Cambridge, MA, USA (11/4/92)

November 20, 1992

Digital Equipment Corporation

Alpha AXP Server Family
Performance Brief - OpenVMS
DN&R Labs CPU2
DN&R Labs CPU2, a benchmark from Digital News & Review magazine, is a floating-point
intensive series of FORTRAN programs and consists of thirty-four separate tests. The
benchmark is most relevant in predicting the performance of engineering and scientific
applications. Performance is expressed as a multiple of MicroVAX II Units of Performance
(MVUPs).
Figure 20 DN&R Labs CPU2 Benchmark Results for Alpha AXP Servers
MVUPs
300.0
272.7
247.5

250.0
216.5
200.0

197.5
179.4

150.0
100.0
50.0
0.0
DEC 3000/400S AXP
DEC 4000/610
DEC 3000/500S AXP

DEC 7000/610

DEC 10000/610 AXP

Figure 21 DN&R Labs CPU2 Benchmark Results for Competitive Systems
MVUPs
300.0
250.0
200.0
150.0
100.0

88.2

86.8
71.4

50.0
0.0
HP 9000/730

IBM RS 6000/350

IBM RS 6000/560

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
SLALOM
Developed at Ames Laboratory, U.S. Department of Energy, the SLALOM (Scalable
Language-independent Ames Laboratory One-minute Measurement) benchmark solves a
complete, real problem (optical radiosity on the interior of a box). SLALOM is based on
fixed time rather than fixed problem comparison. It measures input, problem setup, solution,
and output, not just the time to calculate the solution.
SLALOM is very scalable and can be used to compare computers as slow as 104 floatingpoint operations per second to computers running a trillion times faster. You can use the
scalability to compare single processors to massively parallel collections of processors, and
you can study the space of problem size versus ensemble size in fine detail.
The SLALOM benchmark is CPU-intensive and measures, in units called patches, the size of
a complex problem solved by the computer in one minute.
Table 7 SLALOM Benchmark Results
System

Patches

DEC 10000 Model 610 AXP Server

4608

DEC 7000 Model 710 AXP Server

4432

DEC 4000 Model 610 AXP Server

4072

DEC 3000 Model 500S AXP Server

3934

DEC 3000 Model 400S AXP Server

3750

November 20, 1992

Digital Equipment Corporation

Alpha AXP Server Family
Performance Brief - OpenVMS
Information about Performance
The performance of the Alpha AXP server family was evaluated using industry-standard
benchmarks. These benchmarks allow comparisons across vendors’ systems.
Performance characterization is one "data point" to be used in conjunction with other
purchase criteria such as features, service, and price. For the Alpha AXP server family,
features may include resource sharing with VMSclusters, database systems, multi-vendor
integration with Network Application Support (NAS), and network management with
DECmcc.
We chose the competitive systems (shown in the preceding charts and tables) based on
comparable or close CPU performance and coupled with comparable expandability capacity,
mostly memory and disk. Although we do not present price comparisons in this report,
system price was a secondary factor in our competitive choices.
Notes: The performance information in this report is for guidance only. System
performance is highly dependent upon application characteristics. Individual work
environments must be carefully evaluated and understood before making estimates of
expected performance.
This report simply presents the data, based on specified benchmarks. Competitive
information is based on the most current published data for those particular systems and has
not been independently verified (except as noted). The Alpha AXP performance information
presented in this brief is the latest measured results as of the date published. Digital has an
ongoing program of performance engineering across all products. As system tuning and
software optimizations continue, Digital expects the performance of its servers to increase.
As more benchmark results become available, Digital will publish reports containing the new
and updated benchmark data.
For more information on Digital’s Alpha AXP server family, please contact your local
Digital sales representative.

Digital Equipment Corporation

November 20, 1992

Alpha AXP Server Family
Performance Brief - OpenVMS
References
System and Vendor

Sources

DEC 3000 Model 400S and 500S AXP Servers
DEC 4000 Models 610 and 620 AXP Servers
DEC 7000 Models 610, 620, 630, and
640 AXP Servers
DEC 10000 Models 610 and 640 AXP Server
HP 9000 Models 715/50, 735, and 755

All benchmarking performed by Digital Equipment Corporation.
All benchmarking performed by Digital Equipment Corporation.

HP 9000 Model 730

HP 9000 Model 750

HP 9000 Models 827S, 847S, and 867S

HP 9000 Models 870/100, /200, /300, and /400
HP 9000 Model 887S
IBM RS 6000 Models 350 and 560

IBM RS 6000 Models 580 and 980

SUN SPARC 10 Model 41
SUN SPARC 10 Model 52
SUN SPARC 690 MP52 and MP54
SUN SPARC 2000 (2-cpu and 8-cpu)

All benchmarking performed by Digital Equipment Corporation.
All benchmarking performed by Digital Equipment Corporation.
SPEC, LINPACK, and Dhrystone benchmark results reported in "HP Apollo
9000 Series 700 Workstation Systems" (11/92).
SPEC benchmark results reported by HP (11/10/92).
Dhrystone and LINPACK 100x100 benchmark results reported in "HP Apollo
9000 Series 700 Performance Brief" (1/92).
LINPACK 1000x1000 benchmark results reported in Dongarra, J.,
"Performance of Various Computers Using Standard Linear Equations
Software" (9/28/92).
DN&R Labs CPU2 results reported in Workstation Laboratories, Inc.,
Volume 16, Chapter 12 (3/1/92).
SPEC benchmark results reported by HP (11/10/92).
Dhrystone and LINPACK benchmark results reported in "HP Apollo 9000
Series 700 Performance Brief" (1/92).
SPEC benchmark results reported in SPEC Newsletter (3/92).
Dhrystone and LINPACK benchmark results reported in"HP
9000 Series 800 Business Server System Performance" (1992).
SPEC benchmark results reported in "HP Apollo 9000 Series 700
Performance Brief" (1/92).
SPEC benchmark results reported in SPEC Newsletter (9/92).
SPEC benchmark results reported by IBM (9/92).
Dhrystone and DN&R Labs CPU2 benchmark results reported in Workstation
Laboratories, Inc., Volume 16, Chapter 20 (3/1/92).
LINPACK benchmark results reported by IBM (9/92) and Dongarra, J.,
"Performance of Various Computers Using Standard Linear Equations
Software" (9/28/92).
SPEC benchmark results reported in SPEC Newsletter (9/92) and by IBM
(9/92).
LINPACK benchmark results reported by IBM (9/92) .
SPEC benchmark results reported in SPEC Newsletter (9/92).
LINPACK and Dhrystone benchmark results reported by SUN (11/10/92).
SPEC benchmark results reported in SUN’s Product Data sheet (5/92).
SPEC benchmark results reported in SUN’s Product Data sheet (5/92).
SPEC and LINPACK benchmark results reported by SUN (11/10/92).

Digital believes the information in this publication is accurate as of its publication date; such information is subject to change without notice.
The following are trademarks of Digital Equipment Corporation: AXP, Alpha AXP, the AXP logo, the AXP signature, DEC, and OpenVMS.
HP is a registered trademark of Hewlett-Packard Company. AIX, RS 6000, and IBM are trademarks of International Business Machines
Corporation. Perfect Benchmarks is a trademark of the Center for Supercomputing Research & Development, University of Illinois at
Urbana-Champaign. Prestoserve is a registered trademark of Legato Systems, Inc. SPEC, SPECratio, SPECmark89, SPECint92, SPECfp92,
SPECrate_int92, SPECrate_fp92 are trademarks of Standard Performance Evaluation Corporation. ANSYS is a registered trademark of
Swanson Analysis Systems Inc. SUN and SPARC are trademarks of Sun Microsystems, Inc. UNIX is a registered trademark of UNIX
System Laboratories, Inc. CHARMm and X-PLOR are trademarks of Molecular Simulations Inc.
Printed in U.S.A. EB-N0104-51. Copyright © 1992 Digital Equipment Corporation. All Rights Reserved.

November 20, 1992

Digital Equipment Corporation