Multiple CPUS versus one fast CPU

TNWestTex mcforum at bellsouth.net
Fri Feb 8 15:38:04 UTC 2008




Leslie Satenstein wrote:
> 
> There is a Queuing theory conclusion that   one cpu at speed 2x is better
> than two cpus at speed x.
> 
> The reason is that it takes overhead to schedule two and also each cpu can
> interfere with the other through locks, or memory access.
> 
> But to reduce power consumption today, the trend is to 2 cpus at speed x.
> Increasing the speed causes a disproportionate increase in power
> consumption. The two cpu solution generally uses less power.
> 
> What is the difference expected in performance?  Frankly, it is hard to
> tell, In most cases, it will be ZERO and in other cases (with a lock
> wanted by cpu a which is held by cpu b ) it could be 1% to 2%.  
> 
> Over an hour, that 1% to 2% would wash out, meaning, you should FORGET
> ABOUT IT.
> 
> Go for 2 cpu, shared cache on the die. (Intel or AMD)
> 
Some parts of a process are inherently sequential.  Those go at the single
processor speed.  Amdahl has a limit theorem for parallel processing.  

Robert McBroom

-- 
View this message in context: http://www.nabble.com/Multiple-CPUS-versus-one-fast-CPU-tp15346010p15357591.html
Sent from the Fedora List mailing list archive at Nabble.com.




More information about the fedora-list mailing list