Doug McDonald <mcdonald at (no spam) NoSpAmscs.uiuc.edu> writes:
the powerpc was a good chip. The 68000 was a major disaster
because of pipeline bottlenecks.

In 1984 (well, even earlier actually, as they had to design it before
selling it)?! Most microprocessors were not pipelined at all then.
[Were _any_? Mainframes were, but micros?]

At that time (early-mid 80s), the 68000 was considered a _much_ better
processor. It had far more registers, wider registers, a more regular
design (so easier for compiler writers), a flat memory model, and was in
many ways far more forward-thinking than the rather wacky and clunky
8086 architecture.

I don;t really know about the Mac and its failure to pick ... TWICE ...
the winning chip for its CPU, but I do know Windows and its progression
though new generations chips and various versions of Windows.

Obviously they should have planned for Snow Leopard several years ago
when they were doing CS2 development.

Exactly. And that is the big point.

I don;t really know about the Mac

and its failure to pick ... TWICE ...
the winning chip for its CPU,

but I do know Windows and its progression
though new generations chips and various versions of Windows.

I have written lots and lots of programs for Windows.

Each and every one,
from 16 bit Windows to now, still runs perfectly, on each version
of Windows, with one partial exception: one written for 16 bit Windows
that explicitly wrote directly to registers on a graphics card ... and,
of course, that one didn't even run on the original Windows version
if it had the wrong card, so it was know to be of
limited lifetime.

Anybody else could do the same thing if they wished. Just write to the
rules and don't cheat.

and its failure to pick ... TWICE ...
the winning chip for its CPU,

68k and powerpc were *much* better chips than x86 in many ways.
unfortunately, better does not always mean commercial success.

the powerpc was a good chip. The 68000 was a major disaster
because of pipeline bottlenecks.

Obviously they should have planned for Snow Leopard several years ago
when they were doing CS2 development.

Ray Fischer wrote:

Obviously they should have planned for Snow Leopard several years ago
when they were doing CS2 development.

Exactly. And that is the big point.

I don;t really know about the Mac and its failure to pick ... TWICE ...
the winning chip for its CPU, but I do know Windows and its progression
though new generations chips and various versions of Windows.

I have written lots and lots of programs for Windows. Each and every one,
from 16 bit Windows to now, still runs perfectly,

Doug McDonald <mcdonald at (no spam) NoSpAmscs.uiuc.edu> writes:
I don;t really know about the Mac and its failure to pick ... TWICE
... the winning chip for its CPU, but I do know Windows and its
progression though new generations chips and various versions of
Windows.

Er, well hindsight is everything. Remember, the original Mac was
released in 1984, when it was far from obvious that Intel and
Intel-Compatible CPUs would dominate so much (and of course it was
_designed_ earlier than that). At that time, there was _far_ more
variation in computer designs than there is now.

When Apple switched to the PPC, things had consolidated a bit around
the "PC compatible", but Intel's CPUs were still pretty poor, and the
general wisdom at the time was that better CPU architectures would
increasingly dominate in the future.

Microsoft doesn't build hardware so they just kind of go with whatever
the market throws at them (I guess they've tried to nudge things
occasionally -- e.g. porting NT to the Alpha and Mips processors --
but never very hard).

Doug McDonald <mcdonald at (no spam) NoSpAmscs.uiuc.edu> writes:
the powerpc was a good chip. The 68000 was a major disaster
because of pipeline bottlenecks.

In 1984 (well, even earlier actually, as they had to design it before
selling it)?! Most microprocessors were not pipelined at all then.
[Were _any_? Mainframes were, but micros?]

At that time (early-mid 80s), the 68000 was considered a _much_ better
processor. It had far more registers, wider registers, a more regular
design (so easier for compiler writers), a flat memory model, and was
in many ways far more forward-thinking than the rather wacky and
clunky 8086 architecture.

Intel's main advantages were considered to be cost and a vague
"compatibility" (heh) with earlier 8-bit intel cpus.

[Anyway, what is it about the 68k that you consider less
pipeline-friendly? The 8086 with its dearth of registers and
excessive use of dedicated registers seems far worse in that respect.]

-Miles

Doug McDonald <mcdonald at (no spam) NoSpAmscs.uiuc.edu> writes:
the powerpc was a good chip. The 68000 was a major disaster
because of pipeline bottlenecks.

In 1984 (well, even earlier actually, as they had to design it before
selling it)?! Most microprocessors were not pipelined at all then.
[Were _any_? Mainframes were, but micros?]

At that time (early-mid 80s), the 68000 was considered a _much_ better
processor.

It had far more registers, wider registers, a more regular
design (so easier for compiler writers), a flat memory model, and was in
many ways far more forward-thinking than the rather wacky and clunky
8086 architecture.

Intel's main advantages were considered to be cost and a vague
"compatibility" (heh) with earlier 8-bit intel cpus.

[Anyway, what is it about the 68k that you consider less
pipeline-friendly? The 8086 with its dearth of registers and excessive
use of dedicated registers seems far worse in that respect.]

Actually NT was designed from the ground up to be portable--they saw it as a
competitor to Unix. However there wasn't really much interest in the
non-Intel versiona and finally Microsoft decided that supporting them was a
losing proposition. Windows 2000 and Server 2K3 and 2K5 were ported to the
Itanic without much fuss.

At that time (early-mid 80s), the 68000 was considered a _much_ better
processor.

By Mac heads.

It had far more registers, wider registers, a more regular
design (so easier for compiler writers), a flat memory model, and was in
many ways far more forward-thinking than the rather wacky and clunky
8086 architecture.

You have the key point exactly wrong: it was NOT forward looking.
Forward looking would have meant looking for best design in
a ruthlessly pipelined chip.

Intel's main advantages were considered to be cost and a vague
"compatibility" (heh) with earlier 8-bit intel cpus.

[Anyway, what is it about the 68k that you consider less
pipeline-friendly? The 8086 with its dearth of registers and excessive
use of dedicated registers seems far worse in that respect.]

It appears to be in the details rather than the overall idea. The
"dearth of registers and excessive use of dedicated registers"
is of course completely harmless. The Intel use of a stack
floating point unit was not good for programming, but
apparently is harmless for modern chip design.

The real point is that "neatness" of instruction set is completely
unimportant. The 68000 had a neat set. What matters is how well
it works pipelined, which is understood only by specialists.

The 68k obviously had bad points too,

"mcdonaldREMOVE TO ACTUALLY REACH ME" at (no spam) scs.uiuc.edu writes:
At that time (early-mid 80s), the 68000 was considered a _much_ better
processor.

By Mac heads.

No. Generally. Before the mac came out. By designers, not users.

The 68k and derivatives were very popular amongst workstation makers in
the early 80s (e.g., apollo, later sun, ...), and was generally thought
of as being a good and modern architecture -- unlike the 8086.

It had far more registers, wider registers, a more regular
design (so easier for compiler writers), a flat memory model, and was in
many ways far more forward-thinking than the rather wacky and clunky
8086 architecture.

You have the key point exactly wrong: it was NOT forward looking.
Forward looking would have meant looking for best design in
a ruthlessly pipelined chip.

The 68k _did_ have many attributes that were forward thinking -- larger
registers, more registers, fewer dedicated registers, etc. (all of which
are good for making hardware fast, and easily adapting to larger
problems). It was also overly complicated in terms of things like
addressing modes etc -- but then, so was the 8086.

Intel's main advantages were considered to be cost and a vague
"compatibility" (heh) with earlier 8-bit intel cpus.

[Anyway, what is it about the 68k that you consider less
pipeline-friendly? The 8086 with its dearth of registers and excessive
use of dedicated registers seems far worse in that respect.]

It appears to be in the details rather than the overall idea. The
"dearth of registers and excessive use of dedicated registers"
is of course completely harmless. The Intel use of a stack
floating point unit was not good for programming, but
apparently is harmless for modern chip design.

The real point is that "neatness" of instruction set is completely
unimportant. The 68000 had a neat set. What matters is how well
it works pipelined, which is understood only by specialists.

Sure, pipelining is important -- but what is it about the 68k that makes
it worse for pipelining than the 8086?

The points I mentioned _do_ actually make a difference, and it's not in
the 8086's favor: to take advantage of pipelining you want to avoid
artificial (unneeded) dependencies -- and limited registers and
dedicated registers _create_ artificial dependencies.

The 68k obviously had bad points too, and modern RISC designs are better
in many ways; however, the 68k was at the least, better than the 8086
(which had most of the same problems, and many more).

Miles Bader wrote:

The 68k obviously had bad points too,

such as addressing modes that were, in detail, bad
for pipelining.

The 68000 was influenced by the PDP-11, which was
a "clean" "symmetric" design which had fatal flaws
in the overcomplicated addressing. The RISC designs
used much cleaner addressing, but requiring more instructions.

But time has told the tale, and the Intel designs won, big time.