HP 9000/500 vs. Vectra with 386 CPU and 387 co-processor?
Roger N. Clark
> Does anyone have any experience and/or comments on comparing
> a HP 9000 Series 500 system with a Intel 386-based machine (perhaps like
> the HP Vectra line)?
> ....
The assertions that the HP9000 series 500 is slower than the 300 and 800
series is not exactly true. Neither are the 386 machines faster in
certain cases. Of course you can always find a benchmark that will show
one machines strength. HP doesn't seem to know how good of a machine
they have in the 500. The 500 excells in a number crunching multiuser
(or multitasking) environment. The following benchmarks demonstrate
some points.
For example, the Byte sieve benchmark shows the 350 to be very fast:
SYSTEM sieve.c rsieve.c fsieve.f cbench.c fbench.f dbench.f
u+s u+s u+s u+s u+s u+s
===============================================================================
HP9000 s835 0.11 0 0.11 0 0.17 0 0.05 0 0.04 0 0.05 0
HP9000 s825 0.14 0 0.14 0 0.26 0 0.17 0 0.10 0 0.17 0
-------------------------------------------------------------------------------
HP9000 s350 FPA 0.74 0 0.46 0 0.64 0 0.33 0 0.15 0 0.17 0
HP9000 s350 68881 0.74 0 0.46 0 0.64 0 0.33 0 0.15 0 0.21 0
-------------------------------------------------------------------------------
HP9000 s500 FPA 2.7 0.0 2.7 0.0 3.1 0.0 0.79 0 0.65 0 0.90 0
-------------------------------------------------------------------------------
Compaq 386/20 w80387 0.39
Compaq Portable III
12 MHz MS C 5.0
w/ 80287 1.1
IBM-AT MS F77 80287 4.7 4.8 4.9
-------------------------------------------------------------------------------
Cray-1 Fortran 0.11
-------------------------------------------------------------------------------
The sieve.c is the Byte benchmark, rsieve.c uses registers, fsieve.f
is in fortran, and the cbench and fbench are my own invention that
just does a bunch of +, -, *, /, sin, cos, sqrt, log, etc. On the
face of it, it does look like the series 500 is slow. But the above
benchmarks are separated into to categories: 1) array indexing (the
sieve) and 2) compute power.
I put the Cray in there to illustrate how benchmarks can be used to give a
false picture: no one ine their right mind would say a Compaq 382/20 is less
than 4 times slower than a Cray 1! Maybe that's indicated by the sieve,
but just try doing some REAL problems!
When the 2 are combined (array indexing and floating point
computation) the picture changes:
A Multitasking, CPU intensive Benchmark ("weird box filter" in Fortran)
-----------------------------------------------------------------------
Number of Tasks
System 1 2 3 4 5 7 10 12
-----------------------------------------------------------------------
HP9000/835 HP-UX:2.0 0.5 1.0 1.5 2.0 2.4 3.4 4.9 5.9
HP9000/825 HP-UX:1.2 1.9 3.8 5.7 7.6 9.5 13.3 19.1 22.8
HP9000/500 3 CPUs 5.9 6.0 6.3 8.4 10.5 14.7 21.5 27.8
HP9000/840 HP-UX:1.2 2.1 3.9 5.8 7.8 10.9 16.2 26.0 32.3
HP9000/350 FPA UX6.0 2.7 5.2 7.9 10.5 13.2 18.4 26.4 31.6
HP9000/500 2 CPUs 5.9 6.1 9.1 14.1 16.9 25.4 35.8 43.0
HP9000/320 no FPA 11.4 22.7 33.8 45.2 56.4 80.2 119.4 143
Compaq Portable
III/40:
12 MHz True Basic 122.
" with 80287 119.
with MS C 5.0 7
IBM-PC AT +80287 7.2 - - - - - - -
IBM-PC no 8087 1111. - - - - - - -
-----------------------------------------------------------------------
The "weird box filter" does +, -, /, * on a 200 by 200 array and the result
goes into a 200 by 200 array.
In a compute intensive multitasking environment, the series 500 looks very
good (at least for this benchmark). If you have one task to run, and it
fits in the memory of a 386 machine, then it is probably the
cheapest way to go. If you need multitasking then the 500 is
impressive! Of course, if you are buying new machines, a 350 is
probably cheaper.
BUT WAIT...
What if you have to do complex floating point? Here is a complex
floating point benchmark:
Complex BENCHMARKS
SYSTEM Complex.f
u s real %cpu
===============================================================================
Cray XMP/48 3.9
-------------------------------------------------------------------------------
HP 9000/825 94.4 0.3 96 99%
-------------------------------------------------------------------------------
HP 9040 Unix FPA 271.1 0.3 279 97% <--- 1 cpu only
-------------------------------------------------------------------------------
HP 9000/350 (A) 423.1 0.2 424 100%
HP 9000/350 (B) 442.8 0.1 443 100%
-------------------------------------------------------------------------------
Now we see a single cpu 500 much faster than the 350, and in a multitasking
compute heavy environment with 3 cpus, the series 500 would be similar to
an 825 (e.g. to run 3 jobs, the 825 would take about 283 seconds,
the 500 about 300 seconds)!
Note that the 350 (A) is with a 68881 and (B) is with the FPA; the
FPA is slower than using the 68881 alone!!!!! Both ran HPUX6.0.
My conclusion (as I've posted to the net before):
The HP9000 series 500 is a DAMN GOOD MACHINE. Its fast, and never
crashes. This is borne out by many users doing varied compute, edit, compile,
debug, nroff, etc. jobs. The 500 holds up very well under load. Try that
on a compaq 386 running OS/2 or Unix: it will roll over an die long before
the 500 does.
If you have an application that many people need to run, do a careful
cost-benefit analysis before you buy everyone PC's; I have often seen that
it is cheaper to buy multiuser workstations/minis and still get the same
response, plus you have MUCH easier time sharing data bases and peripherals.
SUPPORT THE 500!
(If you are interested in the actual benchmarks, I can post them or email
them; the box filter benchmark was posted a few months ago).
Finally: use your exact application when benchmarking a machine
before deciding what to buy if speed is that important!
If you have a 500, it might be worth keeping a while unless you need things
like NFS (would somebody please put NFS on the 500!). I'm still pissed HP
pulled the rug out from those who bought the 500. I can understand why, but it
doesn't make it any easier these days when budgets are so tight. I just wish
HP gave more for the 500 in the upgrade program to the 800 series.
Roger N. Clark (stuck with a 500; not unhappy but I want NFS)
bgphp1!rclark
Paul Raveling
[This was an attempted email reply, but it didn't make it
through the mailers]
Interesting results... I'll cast a vote for posting the
benchmarks (including the box filter benchmark -- I must
have missed it the first time around). If you don't post,
I'd appreciate copies to try on assorted machines here.
BTW, a multi-CPU approach seems to me to be the appropriate
architecture for purportedly high performance workstations,
but anything using a Unix kernel will be fighting OS architecture
to utilize multiple processes (and processors) efficiently.
---------------------
Paul Raveling
Rave...@vaxb.isi.edu
Steve Fullerton
with 20 Mhz 80387, 4 Mb RAM running SCO Xenix 386.
While I haven't done much in the area of `real' benchmarks comparing
the two machines, we are a software development company and the only
meaningful benchmark for us is how fast an application runs in our
package. Using this as a ruler, the 80386 is about 4 times the speed
of the HP9000/550 (1 task vs. 1 task). We are not using an ESDI disk
so the I/O throughput on the HP9000/550 is much better than the 80386.
We have had our HP9000/550 for over two years and it has been a real
workhorse but as I write this message, HP is here installing our new
Series 825. The trade-in program ends October 1st and while the hardware
credits aren't the greatest, the software credits will save you a lot
of money. In order to minimize the cost of the upgrade we bought an older
HP9000/540 from HP with 15 1/4 Mb RAM cards for around $1,000 and then
turned around and traded it in on the Series 825 for around $5,000 in
hardware credit (strangely enough, an older Model 540 is worth more as
a trade-in than the Model 550). This still leaves us with the Model 550
which we will sell through to one of the HP equipment resellers.
HP's bottom line is that they want to sell Series 8xx computers and the
salesmen will get as creative as possible in order to make a sale. We
liked the Model 550 but its reign is over.
--
Steve Fullerton Statware, Inc.
scf%statwa...@cs.orst.edu 260 SW Madison Ave, Suite 109
orstcs!statware!scf Corvallis, OR 97333
503/753-5382