From f65d0f9c5d95d814ead5c3a3364b492807f513a4 Mon Sep 17 00:00:00 2001
From: Nicolas Richart <nicolas.richart@epfl.ch>
Date: Tue, 23 Jul 2024 15:38:29 +0200
Subject: [PATCH] nccl config in modules
---
stacks/pinot-noir/modules_kuma_h100.yaml | 10 +++++-----
stacks/pinot-noir/modules_kuma_l40s.yaml | 5 +++++
2 files changed, 10 insertions(+), 5 deletions(-)
diff --git a/stacks/pinot-noir/modules_kuma_h100.yaml b/stacks/pinot-noir/modules_kuma_h100.yaml
index 997a2e1..4599ea4 100644
--- a/stacks/pinot-noir/modules_kuma_h100.yaml
+++ b/stacks/pinot-noir/modules_kuma_h100.yaml
@@ -7,8 +7,8 @@ modules:
OMPI_MCA_pml: 'ucx'
OMPI_MCA_osc: 'ucx'
UCX_NET_DEVICES: 'mlx5_2:1,mlx5_3:1'
- nccl:
- environment:
- set:
- NCCL_IB_HCA: 'mlx5_2:1,mlx5_3:1'
- NCCL_SOCKET_IFNAME: 'bond0'
+ nccl:
+ environment:
+ set:
+ NCCL_IB_HCA: 'mlx5_2:1,mlx5_3:1'
+ NCCL_SOCKET_IFNAME: 'bond0'
diff --git a/stacks/pinot-noir/modules_kuma_l40s.yaml b/stacks/pinot-noir/modules_kuma_l40s.yaml
index 7be541d..1f147c1 100644
--- a/stacks/pinot-noir/modules_kuma_l40s.yaml
+++ b/stacks/pinot-noir/modules_kuma_l40s.yaml
@@ -7,3 +7,8 @@ modules:
OMPI_MCA_pml: 'ucx'
OMPI_MCA_osc: 'ucx'
UCX_NET_DEVICES: 'mlx5_0:1,mlx5_1:1'
+ nccl:
+ environment:
+ set:
+ NCCL_IB_HCA: 'mlx5_0:1,mlx5_1:1'
+ NCCL_SOCKET_IFNAME: 'bond0'
--
GitLab