From cb076816ea9ca97b788770e5ffb968f572c32854 Mon Sep 17 00:00:00 2001 From: Pier Fiedorowicz <117680821+fiedorowicz1@users.noreply.github.com> Date: Fri, 27 Sep 2024 14:40:50 -0700 Subject: [PATCH] Add new LC systems (#2474) --- python/lbann/contrib/lc/launcher.py | 2 +- python/lbann/contrib/lc/systems.py | 2 ++ 2 files changed, 3 insertions(+), 1 deletion(-) diff --git a/python/lbann/contrib/lc/launcher.py b/python/lbann/contrib/lc/launcher.py index 5a0231acaac..9b5ce42f9ba 100644 --- a/python/lbann/contrib/lc/launcher.py +++ b/python/lbann/contrib/lc/launcher.py @@ -94,7 +94,7 @@ def prepend_environment_path(key, prefix): set_environment('MV2_USE_THREAD_WARNING', 0) # Optimizations for Tioga - if system in ('tioga', 'rzvernal'): + if system in ('tioga', 'rzvernal', 'tuolumne', 'rzadams'): #set_environment('NCCL_SOCKET_IFNAME', 'hsi') set_environment('NCCL_NET_GDR_LEVEL', '3') # From HPE to avoid hangs set_environment('MIOPEN_DEBUG_DISABLE_FIND_DB', '0') diff --git a/python/lbann/contrib/lc/systems.py b/python/lbann/contrib/lc/systems.py index 0daeb63ce28..4a7370879c8 100644 --- a/python/lbann/contrib/lc/systems.py +++ b/python/lbann/contrib/lc/systems.py @@ -24,6 +24,8 @@ def __init__(self, cores_per_node, gpus_per_node, scheduler): 'rzvernal': SystemParams(64, 8, 'flux'), 'sierra': SystemParams(44, 4, 'lsf'), 'tioga': SystemParams(64, 8, 'flux'), + 'tuolumne': SystemParams(96, 4, 'flux'), + 'rzadams': SystemParams(96, 4, 'flux'), } # Detect system