Add Topologies for 16-GPU gfx942 SuperNode (#1417)

* Add Topologies for 16-GPU gfx942 SuperNode

- Add GigaIO topologies to tools/topo_expl for dev and testing
- Add GigaIO Columba 16 GPU romeModel and adjust topology
  matching algorithm in rome_models for 16 GPU system
- Fix bug which failed to match Rome Model when using subsets
  of system resources (i.e. ROCR_VISIBLE_DEVICES is set)
- Fixes for topo_expl

* Fix bug w/ 1H16P
Cette révision appartient à :
Benjamin Kitor
2024-12-03 13:12:03 -08:00
révisé par GitHub
Parent 28594b26b3
révision a05329bd0d
8 fichiers modifiés avec 781 ajouts et 71 suppressions
+2
Voir le fichier
@@ -161,6 +161,8 @@ NodeModelDesc model_descs[] = {
{5, "topo_8p_940.xml", "5 nodes gfx940 8P"},
{2, "topo_8p_942.xml", "2 nodes gfx942 8P"},
{2, "topo_8p_942_1.xml", "2 nodes gfx942 8P Alt."},
{1, "topo_16p_gio-1s-1rp-cascade.xml", "GigaIO 16P gfx942"},
{1, "topo_16p_gio-3s-1rp-split-flat.xml", "GigaIO 16P gfx942 1rp-split"},
};
NCCL_PARAM(MaxCTAs, "MAX_CTAS", MAXCHANNELS);