Understanding Parallel Computing: Amdahl's Law

More cores mean better performance, right? That's not what Amdahl says. Learn one of the foundations of parallel computing in "Amdahl's Law." Prepare yourself for math. And lawn mowing.

Category:

Science & Technology

Tags:

• linux
• magazine
• parallel
• computing
• hpc
• amdahl
• law

[achuou]

Hi Douglas, I am too bored to read all that text about this law and verbally it sound simple and easy.....Very nice explanation, thanks for that...and kind of personal comment...you kind of look like "al pacino" and sound like him too...you know that ?

[locanix]

nice shirt

[themassau]

@xpeedX because you increase the road 2 cars can go over the road so you double the speed. but if you increase the single treat performance you could increase the clock = speed or you could increase the instructions/clock look at it as using a car that has 6 seats instead of 4. and HT is just getting all the cars full instead of leaving one guy whit one car. turbo is like increasing the of the roads when there are a low amount of cars at the road.

[themassau]

@xpeedX i think you forgot that it is (somthing)+p/n so the n can only make p/n zero and the equation that is left whit like 1000core would be 1/(1-p). so you can only speed up than by making more of the core parallel or increasing the single thread performance.

look at it as driving form A to B but you got a speed limit of 120km/h (and you can't go faster) the distance between it is 240km so it takes 2 hours. no matter what you do. but you can get more cars trough the road by laying more roads

[xpeedX]

Dear Sir,

i think something is not like you explain it, i do not know much about Parelelism nor HPC but i do on math, and when P has only one value and N increases the remaining equation would be 1/(1/N) that always will be N no matter what N value be (excluding Zero and Infinite). so, if we graph that equation, it will be a linear increase, not an asintotic increase. could you explain why? or where can i be missing your point ?

[kashifkhi]

what is the end of N?

[lennyogarc]

Gustafson's law (weak scaling) might be interesting to discuss, as some people see it as more relevant than Amdahl's Law.

[canadafreakazoid]

Math ftw!