Accelerator boards - no future? Bad business?

Jan Adelsbach jan at janadelsbach.com
Fri Apr 22 15:38:05 CDT 2016


The way one would implement such accelerator boards over an IO subsystem 
bus is to use memory-mapped registers and depending upon the accelerator 
either a single register that executes an instruction on write or a 
small memory for microcode with some way of triggering execution.

Older SGI's have a VME bus and the GIO/GIO64 bus used for I think the 
Indigo's is also documented. This is aside from E(ISA) and PCI cards.
So it is possible but as mentioned before the software is an issue, 
especially if you don't want to explicitly write code for it but have 
the compiler automatically use the accelerator.


On the topic of IRIX emulation:

There used to be an emulator from Stanford which could run IRIX 5.4
https://en.wikipedia.org/wiki/SimOS

Also whilst searching for the latter's name I've found 
https://www.vanheusden.com/miep/

(also Gxemul can emulate the PROM firmware of an O2)

- Jan

On 04/22/2016 09:12 PM, Swift Griggs wrote:
> On Fri, 22 Apr 2016, Guy Sotomayor wrote:
>> [...] emulation (or dynamic translation) is fast enough and with the
>> various virtualization capabilities, it?s not unusual to have multiple
>> different OS?s running on the same HW.
>
> Indeed.  My other wish is that someone far brighter than I will someday
> develop a MIPS emulator capable of emulating IRIX.  The odds of that aren't
> great since untangling the ROM would be a real PITA.
>
>> Apple did this with some success in it?s various CPU transitions.  When
>> they switched from 68K to PPC, the PPC emulated the 68K code.  The same
>> happened when they switched from PPC to x86, again the PPC code was
>> emulated (actually in that case it was dynamically translated).
>
> Yes, and don't forget the NeXT platform which had "quad-fat binaries"
> capable of running on 4 (!) platforms at once.  It might have been a little
> top heavy and impractical at times, but it was soooo cool to me.  I'm sure
> you remember "Rosetta" too.  IRIX on x86 would be heresy to some, but I'd
> welcome it *especially* if it could dynamically translate MIPS ECOFF
> binaries, and run legacy IRIX software. I know the lawyers would never let
> any of that happen, but a man needs to have some dreams.  :-)
>
>> It?s also keeping in character with the old machines.  It?s not adding new
>> capabilities but more replacing old peripherals with something a bit more
>> convenient.
>
> I see your point and I agree.
>
>>   Adding in a new accelerator means not only developing the HW but also
>> writing a boatload of *new* SW in order to be able to take advantage of
>> it.
>
> An oil-tanker sized boat.  Yep.  Unless of course you could keep everything
> pin compatible etc...  which is even harder.  "We're gonna need a bigger
> boat."
>
>> Most of what you see (and what I?m mainly doing these days in terms of
>> ?hobby?) is producing parts to a system in order to keep it running rather
>> than adding completely new capabilities.
>
> Well, I appreciate that fact that anyone is doing anything!  If I ever
> meet "Lotharek" I'm definitely buying him a round.  I get excited just
> looking at some old gear that someone's cleaned up nicely or polished the
> face panels on.  I'd probably die in an apoplectic fit if I ever saw an ad
> for an SGI accel board. If I ever make someone mad, they could totally get
> back at me by handing me a fake flyer for "Press release: Acme Inc
> releases SGI MIPS accelerator board based off high speed GPU". When I
> found out it was bogus I'd go jump off the roof.  Joking aside, I do see
> your point about restoring vs enhancing, and it's well taken.
>
>>   Most companies would rather spend their time and budget doing things for
>> a high ROI and for large and growing markets.
>
> Amen brother, I was just wishing out loud. I needed someone to bring me back
> to Earth.
>
> -Swift
>



More information about the cctech mailing list