[RC5] Athlon core contribution

RogerKIWI at aol.com RogerKIWI at aol.com
Fri Dec 6 10:41:22 EST 2002


In a message dated 12/6/02 9:59:50 AM, dburg at nero.com writes:

> So, let's check the XP vs G4 status. ;-)
> 
> Athlon XP:
> 
> http://n0cgi.distributed.net/speed/query.cgi?cputype=101&arch=all&
> contest=rc
> 572&multi=0&recordid=0
> 
> Power PC G4 and G3:
> 
> http://n0cgi.distributed.net/speed/query.cgi?cputype=all&arch=2&
> contest=rc57
> 2&multi=0&recordid=0
> 
> Fastest XP is nearly 3 times faster than fastest G4... But I actually did
> not plan to purchase any power pc-based computer for so long I can't say
> what in price in equivalent for a XP 2400 in the power pc world. But it
> looks like Power PC are pretty much behind anyway. Let me guess that risc 
> is
> not very good for floating point calculation?
> 

The problem is that currently the RC5-72 core for G4 is not optimised for the 
AltiVec unit in the G4 chips, so doesn't use it.

If you look at the RC5-64 stats you'll see that a dual 800 MHz G4 computer 
could do nearly 17,000,000 keys/sec (8,500,000 keys/sec per processor) 

Power PC 7450/7455 G4     1600     16,991,648     

from 
http://n0cgi.distributed.net/speed/query.cgi?cputype=all&arch=2&contest=rc564

while this page

http://n0cgi.distributed.net/speed/query.cgi?cputype=all&arch=0&contest=rc564

shows that there wasn't any Intel/AMD/x86 chip that got anywhere near that.

The fastest I see on that page is

AMD Athlon XP (Palomino)     2200     6,723,741

Now the G4 chips in PowerMacs are running 1.25 GHz, so one would reasonably 
expect one of these dual processor machines running RC5-64 would achieve 
1.25/.8 = 1.56 times the keyrate - about 26,000,000 keys/sec. (13 
megakeys/sec on each processor).

There's no reason to think that when a G4 AltiVec core for RC5-72 is created 
that the G4 won't be significantly faster again.

Cheers

Roger
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.distributed.net/pipermail/rc5/attachments/20021206/6618027a/attachment-0001.htm


More information about the rc5 mailing list