[RC5] Power PC G4 vs. G5

Ben Payne bpayne at jetheaddev.com
Tue Aug 2 11:31:50 EDT 2005

So I did a bit more checking.  Running the benchmark tells me that I can do 13.3 million k/sec.  However when I run the client in normal mode I only get ~9 million k/sec per processor for a total of ~18 million k/sec, if I'm reading the logging right.  The client starts two crunchers and auto detects two processors.  Activity monitor show the process with 3 thread and taking ~190% of the CPU.  System Info and physical inspection both show that I have 2 processors (as well as my checkbook).  

One thing I find very odd about these results is the discrepancy between the bench numbers and the actual numbers.  All the other systems that I have run on, had very small differences between these two figures (less than 10%).  Of course all the other systems where not dual processors.  I did a bit of playing with the settings and found that I could get the number when running in normal mode to match the bench numbers if I kicked the priority up to 9.  I have other OSX system that don't show this discrepancy.  So I'm not sure what the cause is, but it's nice to find another 6 million k/sec!!!

-> Ben

Normal mode:

[Aug 02 03:44:22 UTC] RC5-72 #a: Completed CB:0B0E3181:00000000 (1.00 stats ...
                       0.00:07:34.09 - [9,458,362 keys/s] [Aug 02 03:44:22 UTC] RC5-72 #a: Loaded CB:0B0E3183:00000000:1*2^32 [Aug 02 03:44:22 UTC] RC5-72: Summary: 5395 packets (5395.00 stats
                       14.11:50:02.41 - [18.50 Mkeys/s] [Aug 02 03:44:22 UTC] RC5-72: 40 packets (40.00 stats units) remain in
                       Projected ideal time to completion: 0.02:27:00.00 [Aug 02 03:44:22 UTC] RC5-72: 8 packets (8.00 stats units) are in
[Aug 02 03:49:41 UTC] RC5-72 #b: Completed CB:0B0E3182:00000000 (1.00 stats ...
                       0.00:07:38.41 - [9,369,249 keys/s] [Aug 02 03:49:41 UTC] RC5-72 #b: Loaded CB:0B0E3184:00000000:1*2^32 [Aug 02 03:49:41 UTC] RC5-72: Summary: 5396 packets (5396.00 stats
                       14.11:55:21.20 - [18.50 Mkeys/s] [Aug 02 03:49:41 UTC] RC5-72: 39 packets (39.00 stats units) remain in
                       Projected ideal time to completion: 0.02:23:19.00 [Aug 02 03:49:41 UTC] RC5-72: 9 packets (9.00 stats units) are in

Bench (dnetc --bench RC5-72 4):
[Aug 02 03:58:37 UTC] Automatic processor type detection found
                       a PowerPC 970 (G5) processor.
[Aug 02 03:58:37 UTC] RC5-72: using core #4 (KKS 7450).
[Aug 02 03:58:57 UTC] RC5-72: Benchmark for core #4 (KKS 7450)
                       0.00:00:16.91 [13,290,099 keys/sec]

-----Original Message-----
From: Jerome Lamarque [mailto:globulle at mac.com] 
Sent: Monday, August 01, 2005 6:38 PM
To: D.net Discussion
Subject: Re: [RC5] Power PC G4 vs. G5

'lo 'veryone
À 16:07 -0700 1/08/2005, Ben Payne tapotait avec ses mimines pleines 
de doigts :

>I have been very surprised by the relative performance of the G4 and 
>the G5 processors.  I'm hoping that some people here can confirm my 
>finding and/or possibly explain why this would be occurring.  I have 
>noticed that the 7450 core is the fastest for both the G4 and G5 
>processors.  I have also noticed that there is a 970 core but it has 
>been disabled.    
>So here are the numbers I'm getting:
>I have a dual processor G5 1.8 GHz it does ~18 million keys a sec
>I have a 1.4 GHz mac mini that does ~15 million keys a sec.
>That all sound good until you add the fact that the G5 has two 
>processors and is actually clock-for-clock less than 50% the key 
>rate of the G4.
>G5: 18 / 1.8 / 2 = 5 million keys per sec per GHz per CPU
>G4: 15 / 1.4 = 10.7 million keys per sec per GHz per CPU

Have you considered the fact that there is a setting in the client to 
tell it to use all the processors of a machine ?
And are you sure you correctly read the figures ?
I have a Dual G5 1.8GHz at work and it says it crunches 18x10^6 k/sec 
per processor.
That's 36x10^6 k/sec.

Either you don't read the figures correctly or you should check if 
you really have two processors in your G5 ;)
Version: 3.1.2
GCM/ED d- s+:- a C++() UB P+ L+ E--- W+ N+(++++) o K- w--- O? M++
V? PS+ PE- Y PGP-- t+ 5++ X++ R !tv b+++ DI D+ G(+++++) e* h+ r- y+
rc5 mailing list
rc5 at lists.distributed.net

More information about the rc5 mailing list