mash@mips.COM (John Mashey) (05/17/89)
I finally laid my hands on the April & May issues.
I have no editorial comments, other than
descriptions of some charts, but I do suggest you read the quotes
very carefully......
-------
1) April 1989 Issue, p.38-44. "Maverick: Data General's 88000 workstation
is first and fast", by Phil Lemmons & Mark Mallett.
"The performance of this machine is truly outstanding. We promised not
to release the exact results of our CPU benchmarks because they were
run as stand-alone programs using a pre-release cross-development
compiler-indeed, the version of C we used for our first tests was not
the Green Hills C compiler that will be sold with the machine."
"But, we can say that this is the fastest workstation MIPS has tested.
Although Data General expects to see a 30 percent improvement with the
newer, native-mode version of the C compiler, our performance numbers
for this machine were faster than those for the DECstation 3100, DEC's
very fast new RISC-based workstation described elsewhere in this issue.
We will publish benchmark results on the Maverick as soon as Data General's
technical people are satisfied that the languages and other systems software
are mature enough to produce results representative of the production machine."
"We can also cite the 88000 performance data that we previously gathered
on the Motorola Platform-88 development system. We would expect the
production version of the Data General workstation to show similar
performance. The Platform-88 was running at 20 MHz and did 32,0000
Dhrystones per second on version 2.0 of the benchmark. Figure 1 shows the
results of running the multitasking AIM Workstation Benchmark Suite V on
the Platform-88".
"DG chose to retarget its own DG/UX kernel for the 88000 rather than use
someone else's port. In fact, DG says it will license DG/UX for other
88000-based systems and has already negotiated one such license."
MASH: Figure 1 is Platform-88 versus SGI Personal IRIS & Everex Step 386/33;
Other than the Aim benchmarks that appear in a figure, I've omitted no
88K numbers cited in the article.
----------
2) April 1989, "PMAX, PVAX, and a DEC 386", MIPS staff, 46-51.
"After experimenting with a prototype, we were fortunate enough to be
able to move onto what DEC calls a "revenue machine", a production model that
could be shipped to a customer."
"We compiled Dhrystone 2.0 with version 1.31 of the MIPS Computer Systems
C compiler. The DECstation 3100 executed 21,925 Dhrystones per second -
a phenomenal number for a 16.67 MHz machine."
"On single-precision Livermore Loops, the geometric mean was 2.945 MFLOPS,
with a minimum of 1.068, a maximum of 5.383, and a standard deviation of
1.233. On double-precision Livermore Loops, the geometric mean was 2.041,
the minimum was 0.808, the maximum 4.153, and the standard deviation 1.013."
"On Whetstone, the DECstation 3100 performed 13,330 Kwhetstones for single
precision and 10,000 for double-precision.... On LINPACK, the DECstation
3100 did 3.157 MFLOPS single-precision and 1.36 MFLOPS double-precision..."
----------
3) May 1989, "Sun's SPARCstation1 & 3/80", p.36-44, by Mark Mallett & Bud Smith.
Table 1 shows Dhrystone 2 w/o and with REG, for:
SPARCstation1, Sun-3/80, HP9000 (340, 360, 370), Everex 386 (25&33),
SGI Personal IRIS, DECstation 3100, and
Data General 16.67 MHz-88000. The DG numbers say "32000 est."
Table 2 shows Whetstones (SP & DP), and Linpack (Single & Double; FORTRAN,
I assume, although I couldn't find where it said.) Machines included
here were: SPARCstation1, Sun-3/80, HP 9000 (340, 360, 370, Whetstone,
but not LINPACK), Everex 386/25, AMI Mark II, DECstation 3100.
Figure 2 has AIM benchmark results, one of which is the Software Development
load for Everex 386/25, DG 88000, DEC PMAX (3100), SG Personal IRIS,
and SPARCstation1. The DG 88000 is not labeled as to its clock rate,
but it seems to have almost identical numbers as the Platform-88 mentioned
in the earlier article [20Mhz], so:
a) The numbers are mislabeled, and are really from the Platform-88 OR
b) It's a 16.7MHz DG 88000 and DG/UX is faster than Motorola's
SVR3 (they're fairly different internally, I think).
c) This is a 20MHz DG 88000, and the performance is about the same
as Moto's SVR3.
I have no idea which of these is true; if somebody can elucidate....?
In the AIM software Development Chart:
0 The curves are grossly similar, running from 1-20 users.
A curve starts at (for example) 110 seconds/user, then goes up
to 140 (around 4-8) and stays there, with some jiggle.
1) The 88K (whichever of the above) is fastest (usually by about
what looks like 20 seconds over a PMAX, except when you
get to 19-20 users, where the 88K jumps up even to the PMAX.
2) The SPARCstation1 is uusually around 200 seconds, except after
16, it drops down around the 160ish seconds that both PMAX and
88K appear to be (which really surprises me; maybe this is just
measurement jiggle, or maybe it's some weird cache/scheduling
effect; it is interesting.)
This article also says:
"The SPARCstation 330 is a fast 25-MHz SPARC-based workstation/server-
the first generally available system based on any RISC processor to
run at 25 MHz - and includes the very fast TI 8847 floating-point
processor." .....
"Although we did not test the SPARCstation 330 or 370, both machines
should be faster than the DEC and Silicon Graphics R2000-based machines
and near the top of the heap in floating-point processing."
--
-john mashey DISCLAIMER: <generic disclaimer, I speak for me only, etc>
UUCP: {ames,decwrl,prls,pyramid}!mips!mash OR mash@mips.com
DDD: 408-991-0253 or 408-720-1700, x253
USPS: MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086chris@mimsy.UUCP (Chris Torek) (05/17/89)
In article <19847@winchester.mips.COM> mash@mips.COM (John Mashey) writes [quoting MIPS magazine on D/G's 88000-based w/s] >"But, we can say that this is the fastest workstation MIPS has tested. >Although Data General expects to see a 30 percent improvement with the >newer, native-mode version of the C compiler, our performance numbers >for this machine were faster than those for the DECstation 3100, DEC's >very fast new RISC-based workstation described elsewhere in this issue. >We will publish benchmark results on the Maverick as soon as Data General's >technical people are satisfied that the languages and other systems software >are mature enough to produce results representative of the production machine." It should be interesting to see. It is worth noting that the DS3100 is about as low-end as you can get with a MIPS [the company, not the magazine] R2000 system. I believe the list price is around $12k (which gets you an 8 MB workstation with a small 1k x 1k display); I expect it to drop rapidly. If DEC have any sense they will put out some R3000 based systems soon too. Right now there is a big market race between DEC and Sun, with D/G about to enter into the fray. (Three cheers for the pending price war :-) ) -- In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 7163) Domain: chris@mimsy.umd.edu Path: uunet!mimsy!chris
baum@Apple.COM (Allen J. Baum) (05/18/89)
[] >In article <19847@winchester.mips.COM> mash@mips.COM (John Mashey) writes: >I finally laid my hands on the April & May issues. >I have no editorial comments, other than >descriptions of some charts, but I do suggest you read the quotes >very carefully...... Is there an address we can write to to get a subscription? -- baum@apple.com (408)974-3385 {decwrl,hplabs}!amdahl!apple!baum
mpogue@dg.dg.com (Mike Pogue) (05/18/89)
In article <17557@mimsy.UUCP> chris@mimsy.UUCP (Chris Torek) writes: >In article <19847@winchester.mips.COM> mash@mips.COM (John Mashey) writes >[quoting MIPS magazine on D/G's 88000-based w/s] >>"But, we can say that this is the fastest workstation MIPS has tested. >>Although Data General expects to see a 30 percent improvement with the >>newer, native-mode version of the C compiler, our performance numbers >>for this machine were faster than those for the DECstation 3100, DEC's >>very fast new RISC-based workstation described elsewhere in this issue. >>We will publish benchmark results on the Maverick as soon as Data General's >>technical people are satisfied that the languages and other systems software >>are mature enough to produce results representative of the production machine." > In fact, I believe that the MIPS people have completed their tests, and will be publishing the numbers shortly. Although there does seem to be a price war about to happen, I suspect that "price/performance war" is a better term :) With all these rapid architectural changes happening right now, are customers really able to keep up? (a rhetorical question....) Mike Pogue Data General Corp. My opinions are my own....
rnovak@mips.COM (Robert E. Novak) (05/18/89)
In article <30942@apple.Apple.COM> baum@apple.UUCP (Allen Baum) writes: >Is there an address we can write to to get a subscription? > baum@apple.com (408)974-3385 Call 1-800-456-1211 for subscription info. Try uunet!mipsmag!editors if your phone is broken. It's not a bad magazine, but I look forward to a name change. Almost everyone I talk to assumes that the magazine is a house organ of MIPS Computer Systems Corporation. THIS IS DEFINITELY NOT THE CASE! -- Robert E. Novak MIPS Computer Systems, Inc. {ames,decwrl,pyramid}!mips!rnovak 928 E. Arques Ave. Sunnyvale, CA 94086 rnovak@abbott.mips.COM (rnovak%mips.COM@ames.arc.nasa.gov) +1 408 991-0402