[RC5] RC5-72 progress

Jonas Maebe jonas.maebe at ugent.be
Sat Jul 31 07:46:30 EDT 2004


On 31 jul 2004, at 13:22, adrian wrote:

>  would assume that the  g5 would be quite a bit faster by the clock 
> speed (2- 2.5 ghz)
>
> although it may take a better g5 dnetc core to extract the best.... i 
> havent heard of anyone getting 50 mkeys/sec out of their g5 !  yet..

My dual G5/1.8GHz gets 13,574,491 keys/sec per processor, so that seems 
to be about the same as your G4/1.25GHz... When looking at an analysis 
by shark of the code, it doesn't look like there are any stalls that 
can be optimised away (there's only 1 stall of 1 cycle).

I think it's simply the fact that the G5 is slower clock for clock than 
the G4 as far as raw calculating is concerned (most instructions take 
more cycles to complete than on a G4). Of course, thanks to its much 
larger memory bandwidth, more out-of-order execution logic and much 
better branch prediction it's a lot faster than a G4 in most real-world 
applications, but RC5 cracking requires very little memory bandwidth, 
the instructions are already optimally ordered and there are barely any 
branches...


Jonas



More information about the rc5 mailing list