Speeding up the 9100...
A few people have noted how much slower the 9100 is than the 9100FT, so it got me to wondering what might be done to improve things. First of all the 68010 CPU is pin for pin compatible upgrade and apparently Atari ST and Amiga folks used to replace the 68000 with the slightly faster 68010 (around 10% improvement or more - some code restrictions that may or may not matter). Second, the 68000 had clock speeds up to 16.67 (the 9100 uses 8mHz) so I'm curious if a little bit of wiring mods might get a gain there. Mostly I'm throwing these ideas out there in the hope that someone will dig further and may actually have some results, I'm a bit busy right now with business stuff... Discussion? John :-#)# _______________________________________________ Techtoolslist mailing list Techtoolslist@flippers.com http://seven.pairlist.net/mailman/listinfo/techtoolslist FTP site is: ftp://ttl.arcadetech.org/TTL/Test_Equipment Archive site: http://seven.pairlist.net/pipermail/techtoolslist/
The Amiga AdSpeed would be an interesting experiment. Its a 14MHz 68000 with a tiny CPU cache which iirc had to be enabled in software so in the 9100 it would be inactive. It supposedly had incompatibilities with certain hard drive interfaces so that could be a dealbreaker here. I suppose you could find one of the 68020/68020 ->68000 adapters with onboard 32 bit RAM but IMHO that probably wouldn't work in a 9100 without major hacking. I had a Microbotics VXL-30 w/o the 32 bit RAM and the improvement was minimal over stock in my Amiga 500. ________________________________ From: John Robertson <jrr@flippers.com> To: "techtoolslist@flippers.com" <techtoolslist@flippers.com> Sent: Wednesday, June 26, 2013 6:17 PM Subject: [Techtoolslist] Speeding up the 9100... A few people have noted how much slower the 9100 is than the 9100FT, so it got me to wondering what might be done to improve things. First of all the 68010 CPU is pin for pin compatible upgrade and apparently Atari ST and Amiga folks used to replace the 68000 with the slightly faster 68010 (around 10% improvement or more - some code restrictions that may or may not matter). Second, the 68000 had clock speeds up to 16.67 (the 9100 uses 8mHz) so I'm curious if a little bit of wiring mods might get a gain there. Mostly I'm throwing these ideas out there in the hope that someone will dig further and may actually have some results, I'm a bit busy right now with business stuff... Discussion? John :-#)# _______________________________________________ Techtoolslist mailing list Techtoolslist@flippers.com http://seven.pairlist.net/mailman/listinfo/techtoolslist FTP site is: ftp://ttl.arcadetech.org/TTL/Test_Equipment Archive site: http://seven.pairlist.net/pipermail/techtoolslist/
The way to make 68K stuff go fast is with a much higher 'local' CPU clock and then cache RAM. Depending on the 9100 architecture (and where the bottle neck is) you might be able to get away some tricks if it's executing from a particular memory that's known to be quicker than 'necessary'. IIRC, one of the old Atari/Amiga tricks was to look at a couple of the 68K signals and basically switch in a couple of quicker clock cycles (like if it's natively ~14MHz, you might sneak it a couple @ ~28MHz when it's just doing 'internal' cycles and not out on memory. Again, IIRC, you want to use the 68HC000 instead because they were *very* happy being overclocked in general. Our Amiga acceleratorr (the SupraTurbo 28) actually ran a 16MHz 'HC000 at 40MHz pretty reliably and 36MHz 'almost always'. We ended up backing it off to 28MHz for production just because that was 2x Amiga 500/2000 speeds and was totally safe. The key was to have some local 'fast' RAM (SRAMs) and some cache tag RAM (IDT7174? That's from memory-- could be wrong). When an address came along and the value wasn't in cache it'd be stored in the fast local SRAM and the cache tag set, then next time if that same address was a 'match' on the tag RAM you knew you could substitute in the fast (28MHz) clock for at least as long as it takes to fetch and deal with the cached values. So doing that you could cache things like ROM space (slowest), RAM (slowish), but not I/O. Overall it did make a pretty big difference on compute intensive stuff. Something like a 68020 would help a little too since it has sort-of a mini-cache built in (like three long words?) so it'll speed up loops and the like, and there *may* be fewer clock cycles on some instructions vs. the 68000/010 too. -Clay _______________________________________________ Techtoolslist mailing list Techtoolslist@flippers.com http://seven.pairlist.net/mailman/listinfo/techtoolslist FTP site is: ftp://ttl.arcadetech.org/TTL/Test_Equipment Archive site: http://seven.pairlist.net/pipermail/techtoolslist/
Er... Don't know where my brain was-- 28MHz is more like 4x original Amiga speeds. ;-) -Clay _______________________________________________ Techtoolslist mailing list Techtoolslist@flippers.com http://seven.pairlist.net/mailman/listinfo/techtoolslist FTP site is: ftp://ttl.arcadetech.org/TTL/Test_Equipment Archive site: http://seven.pairlist.net/pipermail/techtoolslist/
participants (3)
-
Alan McCormick -
Clay Cowgill -
John Robertson