[RC5] Why is CSC bitslice MMX only?

Basil A. Daoust basildaoust at home.com
Tue Dec 21 13:14:38 EST 1999



Bruce Ford wrote:
> 
> >>The original core (indeed all d.net cores for CSC) are bitslice.  The
> >>reason the MMX core is twice as fast is that it processes 64 keys per
> >>instruction vs 32 for the ALU core.
> >
> >
> >That's strange, because the fastest core pre-MMX for me (P2 400Mhz) was 326
> >kkeys/s, but after the Bitslice core was released, that jumped to between
> >950 and 1000 kkeys/s.  So it's obviously more than twice as fast.  Hmmm....
> 
> Okay.  By the very nature of the beast the MMX cores have alot more
> assembly code than the original C cores.  Enjoy the > 2x speed up.
> 

Bruce, 
   first sounds like his system was hurting when it was only doing 326.  That
was
poor performance from a P2-400 as that is slower than my P2-300 laptop which was 
getting around 339 kkeys/s.  

   However I did see improvements around 2.3x faster so the extra machine code 
must help.

   Question about the latest .450 clients,

from the changes.txt file:
2.8004.450  new: macos port including Altivec core support (twice as
                 fast as MMX on otherwise comparable hardware)

   Is this just for the NEWER CPU's, PIII's, Celerons, AMD K6-III's, and
Athlons?

   I tried the client on P2's amd a AMD k5 without any difference.
   If its worth it I could try it on a cyrix 6x86 150+ cpu?

Wondering.

Basil


--
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest



More information about the rc5 mailing list