Hello everyone. Imagine i have multi-core CPU (2 or 4, nevermind). Can i use all cores to accelerate a simulation in Modelsim?
Guest
#4501022
Last time I checked none of the usual suspects managed to do multi core processing of a HDL simulation. This is due to the 'nature law' of event driven simulation architectures. What I tried to speed up verification processes is to split several simulations into different/distributed threads and let them talk over sockets (co-simulation). This is not trivial though and requires some coding work.
I am afraid that EDA vendors see little value in multithreading the simulator kernel (as opposed to, e.g. having separate GUI or logging threads, which is being done). Not only does multithreading cause synchronization overhead between threads, it also raises questions regarding clever partitioning and reproducability (no changes in behavior between two executions of the same simulation, even on different machines). Typical projects the size where it is actually worth your while thinking about potential advantages of multithreading come with many testcases, so that you will get a chance using all your cores at once by running a number of simulations in parallel.