topic Re: Core count limitations for the shared memory transport in Intel® MPI Library

Core count limitations for the shared memory transport

drmiket7777 — Mon, 09 Feb 2026 18:17:56 GMT

I was wondering if the shared memory transport in the latest IntelMPI has any core count limitation. I was trying to run on an Azure HBv5 (MI300C0) node with 368 cores but it crashes while doing the

MPI_Init

Thanks

Michael

Re: Core count limitations for the shared memory transport

Sergey_K_Intel3 — Mon, 09 Feb 2026 21:50:07 GMT

Please check your number of open files ulimit settings.

Run "ulimit -a" to report all current settings or "ulimit -n" just to report the number of open files limit.

Run "ulimit -Sn <number>" to set a new limit.

Intel MPI uses around 3 file descriptors for each rank.

Re: Core count limitations for the shared memory transport

drmiket7777 — Mon, 09 Feb 2026 21:54:19 GMT

I see, thanks.

There is no other build-in resource limitation in the code per core besides the number of file descriptors, right?

Thanks

Re: Core count limitations for the shared memory transport

drmiket7777 — Tue, 10 Feb 2026 16:56:18 GMT

Our ulimit (soft limits are the same) is

$ ulimit -Ha core file size (blocks, -c) unlimited data seg size (kbytes, -d) unlimited scheduling priority (-e) 0 file size (blocks, -f) unlimited pending signals (-i) 442544 max locked memory (kbytes, -l) unlimited max memory size (kbytes, -m) unlimited open files (-n) 65536 pipe size (512 bytes, -p) 8 POSIX message queues (bytes, -q) 819200 real-time priority (-r) 0 stack size (kbytes, -s) unlimited cpu time (seconds, -t) unlimited max user processes (-u) 32768 virtual memory (kbytes, -v) unlimited file locks (-x) unlimited

Can you think of any other resource that could become sparse when we use IntelMPI on very high core count nodes? We are using nodes with 368 cores each.