[nylug-talk] LOWMEM vs. HIGHMEM performance advantage?

[Bryan J. Smith]

On Sun, 2008-04-20 at 00:13 +0000, wdg3rd at comcast.net wrote:
> Hard to say.  Half or more of the problems with the current 32 and 64
> bit Intel-oid CPUs is legacy from the 8/16 bit roots they came from.
> (And yes, there are also remnants of the 8080 and even the 4004 in the
> spiffy new things -- read the assembly language instruction set
> manuals).

Actually, the Intel 80386 mitigated a lot of legacy issues, unlike
earlier Intel designs before it.  About the only thing not mitigated was
the use of the segments, although some people consider that an advantage
(and even a complaint against x86-64, at least before the NX bit on
pages).

Linus himself virtually opened the Intel 80386 programmer's manual when
he went to write his 32-bit OS, forgoing a lot of considerations at the
time.  Most of his assumptions, based on several, advanced features of
the 80386 that were not available in most other (and especially not
legacy) processors earned him a rebuke from many, highly regarded
experts on OS design (including Tannebaum himself).

> Damn, but I wish Motorola had continued the 68k line instead of going
> (and eventually failing) the RISC route -- much better CPUs IMAO.

Some would argue that the 80386, although designed later, had many
architecture advantages over the 68000 line.  But that aside ...

I would strongly argue IBM's POWER RISC architecture to be very superior
to the 68K, and it's not surprising that Motorola partnered with IBM
when the MIPS unification efforts of the late '80s / early '90s fizzled
out.  Motorola's execution was the problem, not the architecture.

I think the defining moment of this was when the XBox 360 first came
out.  A triple core POWER processor with a price-performance that just
totally killed the PC.  IBM had finally evolved into what it should had
always been, a commodity, high-volume foundry with a powerful core
architecture.

POWER is the dominate architecture now.  Heck, even Intel sold its
XScale arm to Marvell, and has decided the only way to get back in the
game is by leveraging x86 compatibility (such as with the new Atom).
The downside is that x86 will never address some market aspects.

> Definitely better for *ix than the crud that is presently hogging the
> market.

Beyond POWER, I would be quick to argue that the Standard MIPS and
Berkeley RISC (SPARC) lineages are far superior to 68K.

It's sad where MIPS is today.  Almost the "standard" architecture 2
decades ago.  Now they are regulated to a losing marketshare with
Broadcom being almost their sole, major vendor.

As far as SPARC, Sun isn't giving up.  The architecture isn't bad at
all, and they've found areas to innovate with it.  But the
commoditization of x86 (or even POWER for that matter) is hard to beat.

But the biggest tragedy is Alpha.  But I won't revisit that.

> But I may be prejudiced, I started with Xenix on the 68000, on a
> little machine known as the TRS-80 Model 16.  (An officially licensed
> port of Unix by Microsoft from AT&T, Microsoft wasn't competent to
> make it an end-user product so they sublet it to Altos, IBM, (the
> original) SCO and Tandy (back in the days when I was doing training
> and tech support at various Radio Shack Computer Centers in the far
> southwest).

Caldera is yet another tragedy.


-- 
Bryan J  Smith              Professional, Technical Annoyance
mailto:b.j.smith at ieee.org  http://www.linkedin.com/in/bjsmith
-------------------------------------------------------------
           Fission Power:  An Inconvenient Solution