SPECmarks for 1991 - Sun and IBM
David Hough
uunet!Eng.Sun.COM!dgh
Mon Nov 5 13:40:53 PST 1990
The following results from various non-official sources may not agree
with what you get from official SPEC or marketing channels, so they shouldn't
be propagated. My understanding is that in both cases the compilers in ques-
tion are in development and are expected to be available in 1991. Thus the
final compiler products may differ from what's being tested now, so numbers
that users could verify may change. The Base Prices should be the cost of the
cheapest system that can duplicate these results, but the Sun prices are just
the cheapest systems in the July price book, except for SS2 which comes from
today's announcement.
I would be curious (so would others no doubt) about what can be expected
next year from MIPS, DEC, and HP, as well as 88K, Intel 486, 860, 960,
Motorola 68040, AMD, Clipper, etc. Actual measurements from actual hardware
using actual compilers would be appropriate comparisons, even if the hardware
and compilers were not expected to be shipped until 1991.
Of course, by 1991 SPEC-2 numbers should be defined and available. I
hope the integer and floating-point computational subsets will be sufficiently
diverse that I would feel good about just quoting the geometric mean with some
measure of dispersion. That would be a good time to get rid of a couple of
the programs that fall a little short of being "realistic applications" -
nasa7 and matrix300 - and get some substitutes from the PERFECT club, for
instance. Some of the larger PERFECT programs are good for the larger systems
that don't look cost-effective on SPEC-1-marks, but they won't run at all on
small systems. Maybe SPEC needs another category of "supercomputer applica-
tions" defined as applications that run on fewer than half the SPEC platforms.
The march of technology has had the result that tenths of a SPECmark are
gratuitous noise, so I rounded them off.
Sun-4/
Benchmark 20 40 65 70 3x0 4x0
SLC IPC SS1+ SS2
001.gcc 9 13 13 20 14 21
008.espresso 9 11 11 18 12 17
013.spice2g6 8 10 10 17 12 17
015.doduc 6 8 8 18 12 17
020.nasa7 11 16 16 27 17 24
022.li 11 13 13 23 14 22
023.eqntott 11 13 13 21 14 20
030.matrix300 13 16 16 28 17 25
042.fpppp 7 11 11 24 14 19
047.tomcatv 7 9 9 18 11 16
SPECmark 9 12 12 21 13 19
int 10 12 12 20 13 20
fp 8 11 11 22 14 19
Base Price 5000 9000 9000 15000 25000 57000
Price/SPECmark 600 800 800 700 2000 3000
IBM RiscSystem/6000
Benchmark 320 520 530 540 550
001.gcc 15 13 17 22 30
008.espresso 16 16 21 25 35
013.spice2g6 21 20 29 35 48
015.doduc 21 21 27 33 46
020.nasa7 33 33 46 55 78
022.li 16 16 20 24 34
023.eqntott 19 19 24 29 40
030.matrix300 31 34 42 49 78
042.fpppp 44 43 55 67 90
047.tomcatv 59 59 80 96 133
SPECmark 25 25 32 39 54
int ? ? ? ? ?
fp ? ? ? ? ?
Base Price ? ? ? ? ?
Price/SPECmark ? ? ? ? ?
More information about the Numeric-interest
mailing list