Right now, integrator.ode_max_steps = 150000 by default. This is very high, and if we need this many steps, we most likely have a problem (like a bad Jacobian). In many problems, lowering this to 10000 or so and then turning on integrator.use_burn_retry speeds up the execution a lot.
We should lower this default to something smaller like 25000. This will force retries (either at the burner or application code level) and should perform better on GPUs where the divergence in the number of steps different zones needs can be a big slowdown.
Right now,
integrator.ode_max_steps = 150000by default. This is very high, and if we need this many steps, we most likely have a problem (like a bad Jacobian). In many problems, lowering this to10000or so and then turning onintegrator.use_burn_retryspeeds up the execution a lot.We should lower this default to something smaller like
25000. This will force retries (either at the burner or application code level) and should perform better on GPUs where the divergence in the number of steps different zones needs can be a big slowdown.