System Benchmarks
Once again, I can provide some CPU specific benchmark data, including
multiprocessor benchmarks. Since SUN finally came into the 20th century by
supporting
MOTIF (X11R5) I can run on those machines as well. IBM and HP do not rate
as I have no access to those systems. These benchmarks are all double precision
floating point computations in molecular modelling. The software should be
available
in the June 1994 timeframe and does reflect ongoing research and automation.
Other
multiprocessor machines will be posted as they become available to me for
testing.
If you want serious compute power, see below... :-)
Berkley Shands
Receptor-III (TM - Tripos Associates) Beta Release version V2.4 timings for the
ACE
series. Uniform distance space scanning at 0.0625A maximum increment with
Adaptive Scanning in modified mode applied. All bonds Adaptively scanned.
Minimum scan factor 1.27 degrees, maximum scan factor 3.52 degrees, Average of
2.73.
Uniform grid resolution of 0.25A. Fully automated setup and processing applied.
All computations in double precision!
This is the last time around for this dataset. Later benchmarks will use a
MOPAC/AM1 minimized subset with torsions variable.
*******************************************************************************
Benchmarks of the ACE series of 57 molecules IC50 > 6.0
*******************************************************************************
Elapsed Child_CPU System_CPU Chargable Forks Ratio
======= ========= ========== ========= ===== =====
* 1311 1264.400 6.140 1245.310 2 0.916 SUN 4m/690 SPARC2 GCC -O2
Solaris 2.3
* 1287 1301.180 10.270 1281.230 5 1.013 SUN 4m/690 SPARC2 SC -xO4
Solaris 2.3
* 1148 1108.950 4.980 1099.070 2 0.969 SGI Indigo 48Mb -O2
(R3K/33Mhz) IRIX 5.1.1.2
* 942 957.470 29.730 940.250 4 1.020 SGI 4d/380s 128Mb -O2
(R3K/33Mhz)
* 643 579.383 5.633 567.733 2 0.907 AXP 3K/300L Gem CC -O4 32mb
-feedback
* 603 529.440 5.270 524.770 2 0.882 SGI Indigo2 64Mb -O2
(R4K/50Mhz)
* 566 516.970 2.070 512.500 2 0.917 SGI Challenge M 96Mb -O2
(R4K/100Mhz)
* 519 504.690 4.880 496.800 3 0.978 SUN Sparc-100 (2X10/51's)
Solaris 2.3 SC -xO4
* 498 445.430 1.960 440.250 2 0.902 SGI Challenge M 96Mb -O3
-mips2 (R4K/100Mhz)
* 437 414.233 2.450 407.533 2 0.952 AXP 3K/300 Std CC -O2 32mb
* 388 337.350 1.717 331.200 2 0.876 AXP 3K/400 Gem CC -O4 32mb
-feedback
* 346 332.033 1.600 325.233 2 0.968 AXP 3K/300 Gem CC -O4 32mb
-feedback
*******************************************************************************
No current data - single precision mode only... Sorry :-(
239 212.367 1.503 200.117 7 0.904 DEC 3K/600 AXP 128mb GEM
"CC" -O4 -feedback
201 187.817 1.333 177.183 7 0.949 DEC 3K/800 AXP 256mb GEM
"CC" -O4 -feedback
*******************************************************************************
Notes:
DEC AXP 3000/300L is 100 Mhz, 3K/300 is 150 Mhz, 3K/600 is 175 Mhz, 3K/800 is
200 Mhz
Many thanks to DEC for providing the 3K/600 and 3K/800 Machines.
All SGI machines ran IRIX 4.0.5 except as indicated.
All SUN machines ran Solaris 2.3 with Motif optimized to -xO4.
All AXP machines ran some flavor of OSF1 :-)
The "GEM" compiler is in save set CMPDEVENH on the OSF/1 CD.
The "STD" compiler is the default "CC" for OSF/1.
The "GCC" compiler is the GNU sun_sparc compiler.
The "SC" compiler is the SUN Sparcworks optional C compiler.
"Child_CPU" is the recorded total CPU for all forked processes,
including
overhead and I/O.
"Elapsed" is the recorded elapsed time from command line parsing to
the exit of the last child processes.
"System_CPU" is the recorded UNIX system overhead from the times()
function
"Chargable" is the algorithmic chargable time from initial rotations
to termination (child process CPU)
*****************************************************************************
CPU benchmarking for current architectures February 1994.
Software is an analytic ring closure method on an eight
membered ring (cylooctane) from the Sybyl library.
"Elapsed" is the wall clock time.
"Child_CPU" is the total CPU over all forks.
"System_CPU" is measured system overhead.
"Conformers" are stericly allowed conformers.
"Forks" is the number of fork() system calls.
"Ratio" is Child_CPU / Elapsed
"Ovhd/Fork" is the System_CPU / Forks.
"Confs/Second" is the number of conformations per elapsed
second.
* CycloOctane Absolute Orientation 02/04/94 Full log files.
wubs2 DEC Alpha AXP 3000/300 Ringsearch
ringsearch -NOX11R4 -nrings -S0.050 -l -Bmodified cyclooctane -Tabsolute
<idle system OSF1 2.0>
Elapsed Child_CPU System_CPU Conformers Forks Ratio Ovhd/Fork 336.5
confs/second
264 258.550 0.283 88843 1 xxx xxxxx STD CC -O2
Elapsed Child_CPU System_CPU Conformers Forks Ratio Ovhd/Fork 403.8
confs/second
220 219.200 0.033 88843 1 xxx xxxxx GEM CC -O4
-feedback
************ Multisearch (multi-threaded ringsearch) *****************
multisearch -NOX11R4 -m4 -nrings -S0.050 -l -Bmodified cyclooctane -Tabsolute
(Absolute torsion orientation, 4 cpus concurent, Scan non-analytic ring sections
in 0.050A increments, Adaptive search in Modified mode with log files for I/O
Ring closure epsilon at 0.10A (a medium tolerance)).
Sun 4m (Sparc2 class machine) 690 4 processor 128Mb SUN Sparcworks Compiler -xO4
Elapsed Child_CPU System_CPU Conformers Forks Ratio Ovhd/Fork 245.4
confs/second
362 1046.980 24.770 88843 179 2.892 0.138380
Sun Sparc-1000 (dual sparc10/51 cpus 128Mb Solaris 5.3 + motif)
Elapsed Child_CPU System_CPU Conformers Forks Ratio Ovhd/Fork 396.6
confs/second
224 347.020 8.070 88843 179 1.549 0.045084
sgi 4d/380 4/8 cpus
<lightly loaded system irix 4.0.1>
Elapsed Child_CPU System_CPU Conformers Forks Ratio Ovhd/Fork 475.0
confs/second
187 728.570 7.040 88843 179 3.896 0.039330
sgi 4d/380 8/8 cpus
Elapsed Child_CPU System_CPU Conformers Forks Ratio Ovhd/Fork 897.4
confs/second
99 728.970 7.070 88843 179 7.363 0.039497