ML NCSHMEM: Difference between revisions
No edit summary |
No edit summary |
||
| Line 4: | Line 4: | ||
Description: Sets the number of ranks that share a shared memory segment. | Description: Sets the number of ranks that share a shared memory segment. | ||
---- | ---- | ||
The number of memory segments that are created are equal to the number of cores per node divided by {{TAG|ML_NCSHMEM}}. Each memory segement has the same size in memory. Hence, the more segments there are the higher the total memory consumption is. However on some machines, especially with a high number of NUMA domains the performance for inference of the MLFFs can drastically degrade if all NUMA domains access the same memory segment. Ideally each domain should have it's own shared memory segment. | |||
== Related tags and articles == | == Related tags and articles == | ||
Revision as of 13:37, 2 February 2026
ML_NCSHMEM = [integer]
Default: ML_NCSHMEM = Number of available ranks per node
Description: Sets the number of ranks that share a shared memory segment.
The number of memory segments that are created are equal to the number of cores per node divided by ML_NCSHMEM. Each memory segement has the same size in memory. Hence, the more segments there are the higher the total memory consumption is. However on some machines, especially with a high number of NUMA domains the performance for inference of the MLFFs can drastically degrade if all NUMA domains access the same memory segment. Ideally each domain should have it's own shared memory segment.
Related tags and articles
ML_LMLFF, ML_MODE, Shared memory