thinking of the "ultimate" retro x86 PCs - what bits to seek/keep ?

[Liam Proven]

On 3 June 2016 at 07:32, Sam O'nella <barythrin at gmail.com> wrote:
>
> Definitely stay away from Cyrix processors. Most computer stores i knew in the 486 era wouldn't even sell them or take them as trade ins. Comparability issues and overheating seemed to be common features.


Really? For a while, I recommended 'em and used 'em myself. My own
home PC was a Cyrix 6x86 P166+ for a while. Lovely machine -- a 133MHz
processor that performed about 30-40% better than an Intel Pentium MMX
at the same clock speed.

My then-employer, PC Pro magazine, recommended them too.

I only ever hit one problem: I had to turn down reviewing the latest
version of Aldus PageMaker because it wouldn't run on a 6x86. I
replaced it with a Baby-AT Slot A Gigabyte motherboard and a Pentium
II 450. (Only the 100MHz front side bus Pentium IIs were worth
bothering with IMHO. The 66MHz FSB PIIs could be outperformed by a
cheaper SuperSocket 7 machine with a Cyrix chip.)  It was *very*
difficult to find a Baby-AT motherboard for a PII -- the market had
switched to ATX by then -- but it allowed me to keep a case I
particularly liked, and indeed, most of the components in that case,
too.

The one single product that killed the Cyrix chips was id Software's Quake.

Quake used very cleverly optimised x86 code that interleaved FPU and
integer instructions, as John Carmack had worked out that apart from
instruction loading, which used the same registers, FPU and integer
operations used different parts of the Pentium core and could
effectively be overlapped. This nearly doubled the speed of
FPU-intensive parts of the game's code.

The interleaving didn't work on Cyrix cores. It ran fine, but the
operations did not overlap, so execution speed halved.

On every other benchmark and performance test we could device, the
6x86 core was about 30-40% faster than the Intel Pentium core -- or
the Pentium MMX, as nothing much used the extra instructions, so
really only the additional L1 cache helped. (The Pentium 1 had 16 kB
of L1; the Pentium MMX had 32 kB.)

But Quake was extremely popular, and everyone used it in their
performance tests -- and thus hammered the Cyrix chips, even though
the Cyrix was faster in ordinary use, in business/work/Windows
operation, indeed in every other game *except* Quake.

And ultimately that killed Cyrix off. Shame, because the company had
made some real improvements to the x86-32 design. Improving
instructions-per-clock is more important than improving the raw clock
speed, which was Intel's focus right up until the demise of the
Netburst Pentium 4 line.

AMD with the 64-bit Sledgehammer core (Athlon 64 & Opteron) did the
same to the P4 as Cyrix's 6x86 did to the Pentium 1. Indeed I have a
vague memory some former Cyrix processor designers were involved.

Intel Israel came back with the (Pentium Pro-based) Pentium M line,
intended for notebooks, and that led to the Core series, with IPC
speeds that ultimately beat even AMD's.

-- 
Liam Proven • Profile: http://lproven.livejournal.com/profile
Email: lproven at cix.co.uk • GMail/G+/Twitter/Flickr/Facebook: lproven
MSN: lproven at hotmail.com • Skype/AIM/Yahoo/LinkedIn: liamproven
Cell/Mobiles: +44 7939-087884 (UK) • +420 702 829 053 (ČR)