For this class, are we referring to speedup only as the time for a single processor over the time for multiple processors, or can it also be compared between different number of processors? ex. Time(n processors)/Time(m processors)?
Metalbird
@bschumuck, you could compare speedups between n and m processors. You could either calculate 2 speedups, the speedup for using n processors, and the speedup for using m processors, or you could divide the time(n processors)/time(m processors), they end up simplifying to the same thing. From what I can tell, we typically want to see how much speedup we get compared to 1 processor, but there should be nothing to stop you from comparing speedups between n and m processors.
Tiresias
For practical purposes, we said that speedup had to be for algorithms that make sense for that number of processors, right? (i.e.: the code used for 1 processor shouldn't be the same code that was optimized for p processors, because then of course you'll get speedup)
For this class, are we referring to speedup only as the time for a single processor over the time for multiple processors, or can it also be compared between different number of processors? ex. Time(n processors)/Time(m processors)?
@bschumuck, you could compare speedups between n and m processors. You could either calculate 2 speedups, the speedup for using n processors, and the speedup for using m processors, or you could divide the time(n processors)/time(m processors), they end up simplifying to the same thing. From what I can tell, we typically want to see how much speedup we get compared to 1 processor, but there should be nothing to stop you from comparing speedups between n and m processors.
For practical purposes, we said that speedup had to be for algorithms that make sense for that number of processors, right? (i.e.: the code used for 1 processor shouldn't be the same code that was optimized for p processors, because then of course you'll get speedup)