Dear All,
We have a Intel Xeon E5-2690 processor
http://ark.intel.com/products/64596/Intel-Xeon-Processor-E5-2690-20M-Cache-2_90-GHz-8_00-GTs-Intel-QPI
I used OMP to run the tri example in LAMMPS ( win 64-bit 20140528)
1 thread takes 108 min and 8 threads take 31 min.
Any suggestions to further accelerate by using 8 threads?
Best Wishes!
Dear All,
We have a Intel Xeon E5-2690 processor
I used OMP to run the tri example in LAMMPS ( win 64-bit 20140528)
1 thread takes 108 min and 8 threads take 31 min.Any suggestions to further accelerate by using 8 threads?
use MPI or implement a multi-threaded version of fix srd.
getting 3x speedup with 8 threads is quite good for that kind of setup,
actually.
please check the timing breakdown at the end of each run section and how it
changes (or not) when you gradually increase the number of threads (or not).
amdahl's law at work...
axel.
Dear Alex,
Thanks for your email.
I tested the multi-thread version of fix srd
1 thread: 108 min
2 thread: 66 min
4 thread: 43 min
8 thread: 31 min
If this is the normal case, I think I am ok.
Best Wishes!
Dear Alex,
Thanks for your email.
I tested the multi-thread version of fix srd
You misread what i wrote. There is no multi-thread version of fix SRD.
1 thread: 108 min
2 thread: 66 min
4 thread: 43 min
8 thread: 31 min
These are not the timings I was talking about.
If this is the normal case, I think I am ok.
I already gave you advice and will not repeat it until you pay attention.
Axel
Dear Alex,
Thanks for the email.
I misunderstood your previous email.
So I have to code myself to get a multi thread version of fix srd?
when you said the timing breakdown, you mean the individual time, not the overall timing?
Best Wishes!
Dear Alex,
Thanks for the email.
I misunderstood your previous email.
So I have to code myself to get a multi thread version of fix srd?
Yes
when you said the timing breakdown, you mean the individual time, not the overall timing?
Yes. Fix srd is part of Modify, lj/tri/omp in Pair