I’ve noticed that now simulations take much longer than they used to (I’m running annual daylight simulations on a room with 1798 test points, and it takes 1 hour to complete one). Has anything happened with new licenses/reduction of CPUs available?
Thanks in advance!
Hi @emilianoadam, Sorry for the delayed reply.
As I commented here there is some work to be done on our side to help with some performance issues.
But in your case, you have started several 100s of runs and each of them is using several CPUs - as a result, your runs will be queued once your account reaches the 100 CPU limit which is currently set for beta testing. The runs will optimize for the whole job to finish faster and because of that, a single run might take longer than expected waiting for the resources to be available. You should calculate the time for each run as the total time divided by the total number of runs in the job.
Just to make sure my comment about “optimize for the whole job to finish faster” is clear here is an example. Say you start a job with 118 runs. The first one starts the first step (which is creating the folder) and by the time that it finishes that all the other CPUs are being used by other runs so now it has to wait for CPUs to be available for the next step before it can continue which will be available after one of the other runs finishes their current step of the simulation.
With this design the time that it takes for the single run will look longer than expected (since it includes the waiting time) but that means all the available resources are being used efficiently by other runs. From the cost perspective, your account will not be charged for the wait time. We only charge the compute time when it is using the resources.