quadibloc <quadibloc@gmail.com> writes:

On Fri, 20 Jun 2025 21:34:24 +0000, MitchAlsup1 wrote:

The attack of the Killer Micro's did not appear until circa 1977.

[...]

But 8-bit microprocessors didn't kill minis and mainframes. They weren't >powerful enough to compete. When did micros really become killers?

Well, they certainly were killers when the Pentium II came out in 1997,
but I'd say that's rather a late date.

Instead, micros were lethal to a lot of larger systems even before they >reached that level of performance. In 1987, halfway between those two
dates, Intel came out with the 387. Hardware floating point for a 32 bit >system? It's about at that point that anything larger became
questionable.

The 387 was a pretty weak FPU compared to, e.g., the MIPS R3010, which
was available at the same time.

Concerning your question, we can easily see that through the
transition of various computer manufacturers from discrete bipolar
CPUs to integrated MOS CPUs.

HP implemented the original (16-bit) HP3000 architecture in TTL, then
in 1979 with Silicon on Sapphire starting in 1979 (probably still multiple-chip). In 1982 HP launched the HP 9000/500 with the HP FOCUS
CPU (a 32-bit variant of the 16-bit HP3000 architecture) as a single
chip, so that might be seen as an early case. They did the first
HP-PA implementation (TS-1 HP 3000/930 aka HP 9000/840) in TTL
(apparently better for prototyping), but replaced it with a faster MOS implementation in 1987; that MOS implementation still was multi-chip,
so probably does not count as a microprocessor.

DEC had MicroVAX pretty early on, but it was considered a low-end implementation. MicroVAX II too AFAIK. They introduced the
MIPS-based DECStations in early 1989 (they had problems keeping the
Unix customers with their VAXen). They came out with the single-chip
CMOS NVAX in 1991 and it had better performance than the earlier
VAXen. At the same time VAX invested a lot in the ECL-based VAX 9000,
which also came out in 1991 with roughly the same performance; the VAX
9000 flopped, so obviously the killer micros had won.

IBM's high-end ES/9000 models used ECL into the early 1990s, but with
the introduction of the 9672 in 1994 they transitioned to CMOS; the
CMOS versions were slower at first (they were single-issue in-order
CPUs, while the high-end ECL ones were superscalar OoO), but the CMOS
speed boost of the 1990s made them outperform the ECL variants after a
few years.

Cray ended their ECL line with the Cray T90, first shipped in 1995.
They implemented the same (Cray Y-MP) architecture in CMOS in two
chips in the Cray J90 in 1994, but that was slower than the T90; the
CMOS Cray SV1 (1998) was also a vector machine; I don't find how it
relates to the T90 in performance, but apparently the memory subsystem
of the T90 was superior. Cray also introduced the T3D with up to 2048 microprocessors (150MHz 21064s) in 1993, but that had quite a
different architecture compared to their vector supercomputers, and
did not replace them. Finally, the Cray X1 (introduced in 2003) is
described as the unification of T90, SV1, and T3E; in particular, the
CPU part is a successor of the SV1, so one might say that Cray has
transitioned to microprocessors with the SV1 at the latest.

Another interesting case is the NEC SX-6 (2001), which is a
single-chip vector CPU, whereas the SX-5 was multi-chip (don't know
the technologty, though).

Why were vector supercomputers so late in making the transition to microprocessors? My guess is that

1) they wanted to have many FUs in the processor, and it took longer
to get that into a single chip.

2) When you need a big expensive memory subsystem, the savings of
having the CPU in one chip rather than many are not so important.

3) At least the Cray people may have been particularly good at getting
very good performance out of ECL, and had less experience with
CMOS, so they delayed the transition, and aparently were able to
keep up with CMOS for a few more years than others (ECL Cray T90
with 2.2ns clock cycle (450MHz) in 1995, whereas the J90 has a 10ns
clock (100MHz) in 1994; for comparison, the 21064A had 275MHz in
1993, and the 21164 had 333MHz in 1995). Note how the ECL MIPS
R6000 (around 1990, 80MHz) was quickly dropped and replaced with
R4000 (1991) and its successors, or how the ECL VAX 9000 did not
perform better than the CMOS NVAX (1991, 91MHz).

- anton
--
'Anyone trying for "industrial quality" ISA should avoid undefined behavior.'
Mitch Alsup, <c17fcd89-f024-40e7-a594-88a85ac10d20o@googlegroups.com>

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)