FloatingFeather
Programmer
Hi. I have been measuring the scalability of some codes I have written in Fortran, and parallelized with MPI.
The thing is if I have been doing it properly (the code scales, however, I would expect it to scale a little bit more with the number of processors).
I am using an amd ryzen 7 1700 8 core processor, but I have also tested the code in a clustar with up to 80 cores.
The doubt I have is if when I run my code in the 8 processors in my ryzen cpu, each core will be working at the same frequency than working only with one. And I think that actually it doesn't, because I've read many times that the ryzen cpus, and threadripper, have some "boost" frequency (I think that you reach that by overclocking), and a base frequency, and even the benchmarks show that multitasking doesn't score as much as multiplying the performance of one clock per the number of clocks.
So, if I want to study scalability correctly, I should actually know the clock speed at which the cpu is working in each case. Also, I think it isn't the same if I put the cpu to work with the 8 cores, and let it run while I use the internet, that if it is running with nothing else going around.
Has somebody here made this type of scalability studies?
Thanks in advance.
The thing is if I have been doing it properly (the code scales, however, I would expect it to scale a little bit more with the number of processors).
I am using an amd ryzen 7 1700 8 core processor, but I have also tested the code in a clustar with up to 80 cores.
The doubt I have is if when I run my code in the 8 processors in my ryzen cpu, each core will be working at the same frequency than working only with one. And I think that actually it doesn't, because I've read many times that the ryzen cpus, and threadripper, have some "boost" frequency (I think that you reach that by overclocking), and a base frequency, and even the benchmarks show that multitasking doesn't score as much as multiplying the performance of one clock per the number of clocks.
So, if I want to study scalability correctly, I should actually know the clock speed at which the cpu is working in each case. Also, I think it isn't the same if I put the cpu to work with the 8 cores, and let it run while I use the internet, that if it is running with nothing else going around.
Has somebody here made this type of scalability studies?
Thanks in advance.