From tw45639@vub.ac.be Thu 22 Jun 95 12:18:12
From: eric@telebit.com (Eric Smith)
Newsgroups: comp.sys.hp48
Subject: Re: HP48 vs. TI92
Date: 20 Jun 1995 07:54:10 GMT
Distribution: world
Message-ID: <ERIC.95Jun20005410@marble.telebit.com>
References: <3rof35$s3a@zipper.zip.com.au> <kevin.jessup.695.0052F4E1@mail.mei.com>
To: vhadley@ee.utah.edu (Vince Hadley)

In article <3rqktj$a4s_001@cc.utah.edu> vhadley@ee.utah.edu (Vince Hadley)
writes (comparing the TI-92 to the HP 48):
> I think in order for HP to compete, they are going to have to rethink 
> their basic architecture in order to succeed.  Everyone wants power, power, 
> power- Power to handle more complex and diverse forms of symbolic math,
> power to do it and numerical calculations faster, and a more powerful user 
> interface.

IMHO, the "basic architecture" you are asking HP to rethink really boils down
to the RPL code embedded in the HP 48, and RPL is sufficiently powerful to
handle more complex and diverse forms of symbolic math, and to provide a more
powerful user interface.  This leaves the issue of "faster".

While RPL is fine, the Saturn processor architecture that has fallen behind
the times.  The Saturn architecture was designed as a more general replacement
for the Harvard architecture processors used in earlier HP calculators from
the HP-35 through the HP-41 and -10 series.  These earlier processors were
amazingly well suited to simple BCD floating point calculation.  For example,
the HP-35 implemented a full featured non-programmable scientific calculator
with only 768 words of ROM!  Unfortunately they were not as capable for
general purpose computing, and even the HP-41 stretched them to their limits.
The Harvard architecture required many HP-41 microcode routines to handle two
special cases depending on whether the "user code" program was stored in ROM
or RAM.

Some of the Saturn design goals must have been (in no particular order):

1.  Very low power consumption.

2.  Unified address space for easier and more powerful programming (both for
    HP and the customer).

3.  Automatic configuration of memory and I/O devices on the bus, similar to
    today's cry for Plug-and-Play on PCs.  This avoids the type of problems
    seen on the HP-41, where memory modules (including extended memory) had to
    be located in the appropriate expansion ports by the user in order to
    avoid memory gaps or overlap.

4.  Minimum bus pin count to allow expansion modules with low pin count.
    For example, the HP-41 expansion module connector only requires 12 pins.
    More connector pins increase size and cost.

These goals were likely based on assumptions that the Saturn would be used
by a followon to the HP-41, in which all of the silicon would be manufactured
by HP, and the expansion ports would have a proprietary design.  Saturn is
very well suited for this design paradigm.

Goal 1 (low power consumption) will always be critical for handheld devices.
A major part of the traditional way HP (and most vendors) handled this goal
was by keeping the clock rate low.  This is fine when you press the cosine key
manually and get an answer in a half a second, but more sophisticated
calculators will require more horsepower.  It can in fact be shown that
reduced operating frequency is actually less energy efficient (Burd and
Brodersen '95, Proceedings of the 28th Annual HICSS Conference, Jan. 1995;
Vol. I, section 5.2, available on the web from
"http://infopad.eecs.berkeley.edu/CIC/gpp.html").

Goal 2 (unified address space) remains quite worthwhile today, and any Saturn
successor will likely have this.

Goal 3 (automatic configuration) is still important but can be provided by
different means than Saturn used.

Goal 4 (low bus pin count) has been rendered obsolete and goal 3 _must_ be
achieved in a different fashion because of cost issues that have been
recognized more recently.  With the advent of the HP-28S, HP realized that the
only practical way to offer larger memory capacity was to use industry
standard RAM chips rather than the custom RAM chips with the Saturn bus
structure which were used in the HP-71B, HP-18C, and HP-28C.  Then when the HP
48SX was desgined they extended this reasoning to memory cards.

Advances in technology have allowed each successive generation of Saturn
processors to be faster, from the original 650 KHz (if memory serves) of the
HP-71B to the 4.0 MHz of the HP 48G and HP 48GX.  Unfortunately one of the
drawbacks of using a proprietary processor is that you typically can't take
advantage of the cutting-edge process technology that is available to
mass-marketed processors.  The Saturn processor in the HP 48G is probably
faster at BCD floating point math than the 68000-derivative processor in the
TI-92, but for non-numeric problems (including symbolic algebra and calculus),
the 68000 is probably the winner.  Over time we are calling on our calculators
to do more and more non-numeric operations.

Part of the reason the Saturn can't keep up with the 68000 for non-numeric
operation is that the 68000 was designed to take advantage of a wide data
bus, whereas the deliberately narrow bus of the Saturn keeps the memory
bandwidth low.  Since the narrow bus requirement no longer exists, it is
reasonable to expect that HP will switch to a new processor architecture for
future calculators.

Fortunately, RPL is not strongly tied to the Saturn architecture, although the
current implementation is.  It is very plausible to expect to see RPL running
on a RISC processor.  There have recently been many silly claims posted to
this newsgroup that the next generation HP would have a PowerPC, Sparc, or
similar high-end RISC processor.  While these claims can certainly be
dismissed as wishful thinking, there are a number of power-efficient RISC and
hybrid RISC/CISC chips which would be quite suitable, such as the ARM, the
Hitachi SH series, the NEC V8xx series, etc.  It isn't completely inconceivable
that they could use a low end HP Precision Architecture processor, although I
don't know of any existing HP PA implementation that would be suitable.

> The thing that will always win out for me though is "STAY with 
> RPN" while doing the rest.

To increase market share it may be necessary to put more "traditional"
(i.e., non-RPN) functionality into future units.  Yes, I know about algebraic
objects in RPL, but to fully take advantage of the power of the HP 48 it is
necessary for the user to learn RPN.  Many people resist this.  HP has flirted
in recent years with non-RPN calculators such as the 17B and 19B, but I must
surmise that they realized that they had alienated loyal customers since they
added RPN capabilites back in with the 17BII and 19BII.

I am personally not the least bit opposed to HP making efforts to increase
the accessibility of future high-end HP calculators to the masses by making
more capabilities usable with algebraic entry, as long as they don't forget
their loyal customer base who want RPN.  If RPN is removed, or becomes the
less-supported "other" mode, I will be much less inclined to buy HP.

Cheers,
Eric

