[rc5] Processor Core Autoselection

Joe Zbiciak j-zbiciak1 at ti.com
Mon Nov 3 03:59:29 EST 1997


'Justin Cave' said previously:
 
| When the client starts up, it runs a quick test with each of the cores and
| uses the fastest one.  This may vary from the expected, as in your case,
| depending on what your system is doing.  [...]

During RC5-56, with the V2.005 client, I consistently got higher
Kkeys/sec with the AMD K6 core on my Pentium II 300MHz under Linux.
(760K for K6 core vs. 740K for the P-II core.)  That core is the one
the autoselect chose initially, but separate benchmarking confirmed
that it was the faster core.  Now, with the 64-bit client, I get wildly
different results -- 750K for the P-II core vs. 540K for the K6 core.
:-)  Similarly, the 56-bit K6 core was faster than any of the others (in
v2.005) on my 486DX2/66 and AMD 5x86-133 (not a K5) -- but slower when
the 64-bit client came out.

Basically, the way I chose which to run was to run the short benchmark
once with each core to pick the top three cores, and then run the long
benchmark to pick the fastest of these cores as the core I'd run.  As
long as you do this on an unloaded machine, it seems to be a pretty
effective way of picking the highest Kkey/sec producing core for a
given piece of hardware and version of the software.

Regards,

--Joe

-- 
 +------------ Joseph Zbiciak -----------+                                    
 |- - - - -  j-zbiciak1 at ti.com  - - - - -|  "Reality is that which, when      
 | - http://www.primenet.com/~im14u2c/ - |   you stop believing in it,        
 |- - - -Texas Instruments, Dallas- - - -|   doesn't go away" -- Philip Dick  
 +------#include <std_disclaimer.h>------+                                    
----
To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the body.



More information about the rc5 mailing list