On 6/18/2011 4:49, Warp wrote:
> Darren New<dne### [at] sanrrcom>  wrote:
>> Certainly having more cores than the number of instructions the program
>> executes is going to be pointless, so mathematically speaking, the law
>> *always* applies, even if it's silly in practice to take it to that
>> exaggeration.

>    Perhaps you meant that having more cores than the amount of generated data

I was speaking entirely theoretically. Amahl's law says you only get so much 
speed-up, no matter how many CPUs you throw at it. If the sequential program 
executes 100 billion instructions over the course of solving the problem, 
giving it more than 100 billion processors cannot possibly speed it up any 
farther.

Sort of like an O(n) question ignoring the constant factor sort of discussion.

> would be useless (because each processor has to output at least one bit of
> data to be useful). Since generated data typically has quite many bits, it's
> not like you would hit this limit any time soon.

Exactly. That's why I said "mathematically speaking."

When I studied SIMD algorithms in school, the assumption was that you always 
had N processors, where N was the same number you were giving to O(N) sorts 
of calculations.

-- 
Darren New, San Diego CA, USA (PST)
   "Coding without comments is like
    driving without turn signals."

Post a reply to this message