[mkroot] To the people who keeping pinging me off list...

[Rob Landley]

On 03/08/2018 04:04 PM, Alain Toussaint wrote:
> 
>> Sigh, I thought there was an open 486 but
>> https://en.wikipedia.org/wiki/Zet_(hardware) is 16 bits only.
> 
> Forget about x86 for new designs. I'd be more interested into putting to good uses what are deemed
> useless boxes.

For regression testing on real hardware, being able to load a 486 into an FPGA
would be useful. (In the meantime there's qemu, but qemu often runs stuff real
hardware won't.)

> New designs of anything, I'd base it on risc-v or any of its spinoff (lowrisc for example) upon
> which, I'd plug in coprocessors to do the necessary calculations to do drug development and build a
> cluster of compute boxes in a rack using said design to work on drug development. I do have the
> Harris & Harris book but I'm still a long way to go from designing such coprocessor.

I've worked on https://j-core.org since 2014 so I may be biased, but I have zero
interest in risc-v. It's the itanium of the open hardware world. Where they're
light years ahead of everybody else is _marketing_, not technology.

Every time somebody points out a new corner case they lose at (even against x86
or arm) they invent another instruction set extension or mode that's
incompatible with the previous ones just to win THAT benchmark, which don't
combine in any useful way but they win each and every benchmark one by one, in
isolation, essentially with different compiler options. And then they go "look
how simple we are, you can build this chip to exclude the extensions you're not
using! Compatibility? What's that?"

*shrug* I expect a large wave of dot-com style investment in them, and to see it
crash and burn in 5-7 years, and then maybe something useful will emerge from
the wreckage? (I doubt it, but it's possible.)

It's sort of an
https://www.joelonsoftware.com/2000/05/12/strategy-letter-i-ben-and-jerrys-vs-amazon/
thing where they are so desperate to emerge as the new STANDARD and drive
everyone else from the field, except open source and webassembly (heck, it was
the point of java's "write once run anywhere") erode the importance of a
specific instruction set, and the dominant one by unit volume is going to be arm
for at least the next decade anyway (not just installed base but new
deployments) because things like raspberry pi are leveraging the unit volume of
phones (the way a "blade server" was a laptop chipset stuck in a server rack)
and semiconductor manufacturing cost is almost entirely a question of unit
volume and ASIC fabs continue to be large proprietary multi-billion dollar
assets where the mask layout software is essentially a giant proprietary
hand-rolled VHDL/verilog compiler that's different for every single fab (the old
"every hardware platform has its own C compiler and libc provided by the vendor"
days) they won't even let you see the SPECS for until you sign an NDA, and they
have a dialect of java that generates verilog as an output...

Sigh. Can of worms. Linux has 30 architectures, this one has a microsoft-style
marketing campaign behind it but so did itanium, what got microsoft its monopoly
was the predatory distribution contracts it negotiated tying up the supply
chain, and pushing back against that linux development didn't have to spend 6
figures to compile a kernel but even as a free download we had 20 consecutive
"year of the linux desktop"s with <2% market share on that because it couldn't
get preinstalls...

Ahem. Can. Of. Worms.

> Alain

Rob

P.S. Did the US Navy ever work its way through their stockpile of HST modems?
http://tycho.usno.navy.mil/ads.html says it's still using them...