damien_at_[hidden] wrote:
> Also, if CG solver = Conjugate Gradient, you have a very fast algorithm
> compared to say a direct solver, so you might be loading the processor in
> shorter spikes which is harder for the kernel to schedule.
In my limited experience of CG solvers, there is likely to be a high
cache miss rate. With MPI CG solvers, on a machine which shares a
memory bus between processes, bus contention is likely to slow down the
rate at which cache misses can be satisfied. This could work in favor
of the performance with a single process per node.
|
