[Sussex] SPARCBooks, Powerbooks and bog standard laptops

[Steve Dobson]

Geoff

On 16 January 2003 at 11:29 Geoff Teale write:
> Here I am a day later and a lot more informed:

Arhhh, a little older and a little wiser :-)
 
> Steve wrote:
> ------------
> > I've been following this SPARCbook thread.
> 
> Glad to hear it :)  As the SPARCSpert on the list I was kinda 
> hoping for your input.

I'm not as up to date as I use to be,  this company doesn't have
me track Sun in the way Computing Devices (my first company) did.

> > I would be wary of the SPARCbook.... <snip>
> 
> Yup.  This is consitant with my info.  I don't have Tadpoles 
> prices yet but the Naturtech beasts start at £3,995!!!!
> That's for an UltraSPARC IIi @ 400mhz, 256MB of RAM and a 10GB HD.
> Maximum battery life is 2.5 hours.

That price looks a lot better than want I remember.  My memory is 
a little off here because:
  1). First I didn't spec or get a quote on the box,
  2). It was five+ years ago, and mainly because
  3). A lot of alcohol has flowed since then.
But the spec on our box was more like: SparcV6 50mhz, 128MB RAM &
2GB disk.  The price was somewhere around £10,000 - £12,000.
 
> For that price I could by a low end Sun Workstation (still 
> slightly higher spec than the laptop) _and_ a top end x86 laptop
> or powerbook. 

Production volume is everything when it comes to pricing.  We wanted
ours as a demo sales machine for a system we had (more or less)
finished.  As it had custom imagery processing code tied into the
graphics card the most important factor was the graphics card that
came with the laptop.  £12,000 may sound like a lot to some of you,
but it took < 2 weeks to get the system up and running.  You don't
get much development resource in a company for £12,000, a good
approximate is £1000 per day.  So the whole project cost less
than £50,000, that's value for money in the Military market places.
More important than the money saving was the risk.  This was also
the lowest risk solution - no code to re-compile (yes recompile,
we just installed the production binaries and just hacked the
file systems under the covers).

> I was kind of hoping they'd come in at slightly above £2000 

...and in the real world...

> (I knew they weren't going to be cheap) - but really four grand is a tad 
> on the steep side.

As I said above, the production runs just aren't big enough to bring
the prices down.  At that price the buyers are going to be people 
like me back then:  hardware spec is the driving factor, price is a
minor consideration.

> > Okay, I know that on a single CPU system Linux flies compared 
> > with Linux, but that same would be true on the iBook too wouldn't
> > it?  I  can see you spending a lot of money and not getting that
> > much for it.

<snip>

> The latest thing I've seen is that Hyperthreading P4's are where the big
> performance gains are for LINUX - hopefully this will turn up in the
laptop
> chipsets soonish.

That's not surprising.  I read an article a while back saying that Intel
was the only real chip maker left.  I wouldn't go that far, but in the
slow down they do have the more income so can still afford to develop 
more.  Or maybe that's what the market expect so Intel have to keep 
pushing.  If they didn't come up with something new the market would still
expect the top of the range chips to become cheaper.

> > So dump the shipped OS and install Linux.  I'll run fast I'm sure.
> 
> See above - thats fine if you already have the hardware, but it doesn't
make
> economic sense to do this from scratch. plus Apple hardware support under
> LINUX is _much_ worse than PC support.

I agree, with volumes the way they are, and the difference between
processors
unimportant (most of the time) Intel based systems give the best value for
money.  But I'd still like a SPARC for home.
  
> Hmm.. www.novatech.co.uk or www.dnuk.com will do that for you !

I wish I'd know about these before.
 
> > If you're going into that then you have even more problems.  I can
remember
> > Sun's presentation to me years ago when they first announced 64-bit that
> > on (HP I think) use a different 64 bit number format.  This has an
effect
> > when bit slicing, and maksing of hardware registers.  This should only 
> > really be and issue for kernel hackers.  App-land hackers should write
in
> > a way that don't cause problems - portablity is normally more important 
> > for applications than abolulte raw speed - of course there are always
> > exceptions.
> 
> Yup, this is along the same lines as what I was saying.  Funnily enough it
> is difficult to write an ICCCM compliant window manager for a 64-bit
system
> because the standard calls for 32-bit unsigned integers where you are
going
> to recieve 64-bit ones from X... this is one reason it would have been
nice
> to have a 64-bit system to work on.

Funny we wee agree on stuff like this.  And to me this is a kernel hacking
problem.  Okay the X system is user-land based, but it the kernel is giving 
direct access to the hardware for performance.

So, as you say, if you're going to be hacking the kernel, X, or any other
system that talks directly to the hardware, is the extra cost of a 64-bit
system worth it.  For me it is not (as much as I would like one), for you
I leave you to have the argument with your wallet.

Steve