Risc System/6000

[Dick Dunn]

ron at woan.austin.ibm.com (Ronald S. Woan/2100000) writes:
...[price info deleted]
> Not a bad price, even so, considering 27.5 MIPS (peak 100) and
> 7.5MFLOPS. That works out to be $543.45/MIP system cost which is the
> lowest in the industry, even among name brand 386/486 PCs, I believe.

Let me start by saying that the new IBM boxes look like they'll actually
come in at believable prices when they're available.  That nicety out of
the way,...

27.5 MIPS is a junk number.  There is no useful meaning to "MIPS" which
justifies expressing three significant digits, especially with a variable
instruction-issue rate.  ONE significant digit is pushing it.  $543.45
(FIVE figures?!?!?!?) is doubly bogus, since it's calculated by bashing the
barely-meaningful MIPS number against an even-less-relevant cost.  Think
about it:  The easiest way to reduce the $/MIP ratio is to offer a more
stripped-down configuration!

Then there's the business of reporting Dhrystone 1.1 numbers.  That's just
a disgrace.  I have to wonder if it took some special effort to dig up a
copy of the old Dhrystone 1.1 benchmark instead of using a "current" (can
we call it current even tho it's more than a year old?) version!  And, it
seems, the MIPS numbers are calculated off these lousy Dhrystone values.
(I *hope*--for IBM's sake--that they're lousy!  Their compiler had better
be good enough to over-optimize Dhrystone 1.1.:-)

If we're going to discuss benchmarks on these machines, let's start with
something halfway reasonable, like SPEC.

If you want some meaningful numbers, you're going to have to think a little
bit.  You may even find--horror of horrors!--that you won't be able to
describe the cost-effectiveness of a machine, compared to all others in the
market, with a single number.
-- 
Dick Dunn     rcd at ico.isc.com    uucp: {ncar,nbires}!ico!rcd     (303)449-2870
   ...Don't lend your hand to raise no flag atop no ship of fools.