Hi Shuwei, The MPICH library does not implement a hard limit on the number of ranks. That being said, the rank is stored as a 32bit integer, therefore subject to the limit of MAX_INT. In practice, it is more likely that the library will run out-of-memory before hitting that limit. MPICH is optimized for large scale use cases than contains hundreds of thousands processes, so it should be fine for your use cases. Best, Yanfei Guo Assistant Computer Scientist MCS, ANL From: Shuwei Zhao via discuss <[email protected]> Reply-To: "[email protected]" <[email protected]> Date: Monday, October 29, 2018 at 3:43 PM To: "[email protected]" <[email protected]> Cc: Shuwei Zhao <[email protected]> Subject: [mpich-discuss] Maximum number of processes in compute farm Hi, all I have several questions about parallel scale of mpich: Is there a hard limit on the number of processes that mpich can start in a compute farm? As the number of process going up , for a large-scale(1000-10000) loose-communication application(the communication overhead is minimal compared to the runtime), is it possible that the performance of the mpich will go down because the message passing will take longer time? Thank you very much, Shuwei