diff --git a/configs/runners.yaml b/configs/runners.yaml index 6f0b35e9c..651ab1cc8 100644 --- a/configs/runners.yaml +++ b/configs/runners.yaml @@ -30,18 +30,18 @@ labels: h200: - h200-cw_00 - h200-cw_01 - - h200-nb_0 - - h200-nb_1 - - h200-dgxc-slurm_0 - - h200-dgxc-slurm_1 - - h200-dgxc-slurm_2 - - h200-dgxc-slurm_3 - - h200-dgxc-slurm_4 - - h200-dgxc-slurm_5 - - h200-dgxc-slurm_6 - - h200-dgxc-slurm_7 - - h200-dgxc-slurm_8 - - h200-dgxc-slurm_9 + - h200-nb_00 + - h200-nb_01 + - h200-dgxc-slurm_00 + - h200-dgxc-slurm_01 + - h200-dgxc-slurm_02 + - h200-dgxc-slurm_03 + - h200-dgxc-slurm_04 + - h200-dgxc-slurm_05 + - h200-dgxc-slurm_06 + - h200-dgxc-slurm_07 + - h200-dgxc-slurm_08 + - h200-dgxc-slurm_09 - h200-dgxc-slurm_10 - h200-dgxc-slurm_11 - h200-dgxc-slurm_12 @@ -53,8 +53,8 @@ labels: b200: - b200-cw_00 - b200-cw_01 - - b200-nb_0 - - b200-nb_1 + - b200-nb_00 + - b200-nb_01 - b200-dgxc_00 - b200-dgxc_01 - b200-dgxc_02 @@ -77,29 +77,31 @@ labels: - b200-dgxc_08 - b200-dgxc_09 b200-multinode: - - b200-dgxc-slurm_7 - - b200-dgxc-slurm_8 - - b200-dgxc-slurm_9 + - b200-dgxc_06 + - b200-dgxc_07 + - b200-dgxc_08 b200-new: - - b200-nscale-slurm_0 - - b200-nscale-slurm_1 - - b200-nscale-slurm_2 - - b200-nscale-slurm_3 - - b200-nscale-slurm_4 - - b200-nscale-slurm_5 - - b200-nscale-slurm_6 - - b200-nscale-slurm_7 - - b200-nscale-slurm_8 - b200-nscale: - - b200-nscale-slurm_0 - - b200-nscale-slurm_1 - - b200-nscale-slurm_2 - - b200-nscale-slurm_3 - - b200-nscale-slurm_4 - - b200-nscale-slurm_5 - - b200-nscale-slurm_6 - - b200-nscale-slurm_7 - - b200-nscale-slurm_8 + - b200-nscale-slurm_00 + - b200-nscale-slurm_01 + - b200-nscale-slurm_02 + - b200-nscale-slurm_03 + - b200-nscale-slurm_04 + - b200-nscale-slurm_05 + - b200-nscale-slurm_06 + - b200-nscale-slurm_07 + - b200-nscale-slurm_08 + - b200-nscale-slurm_09 + cluster:b200-nscale: + - b200-nscale-slurm_00 + - b200-nscale-slurm_01 + - b200-nscale-slurm_02 + - b200-nscale-slurm_03 + - b200-nscale-slurm_04 + - b200-nscale-slurm_05 + - b200-nscale-slurm_06 + - b200-nscale-slurm_07 + - b200-nscale-slurm_08 + - b200-nscale-slurm_09 mi300x: - mi300x-amds_00 - mi300x-amds_01 @@ -142,6 +144,7 @@ labels: - mi355x-amds_06 - mi355x-amds_07 - mi355x-amds_08 + - mi355x-amds_09 mi355x-disagg: - mi355x-amds_06 - mi355x-amds_07 @@ -153,10 +156,26 @@ labels: - mi355x-amds_18 - mi355x-amds_19 gb200: - - gb200-nv_0 - - gb200-nv_1 - - gb200-nv_2 + - gb200-nv_00 + - gb200-nv_01 + - gb200-nv_02 + - gb200-nv_03 + - gb200-nv_04 + - gb200-nv_05 + - gb200-nv_06 + - gb200-nv_07 + - gb200-nv_08 + - gb200-nv_09 + - gb200-nv_10 + - gb200-nv_11 + - gb200-nv_12 + - gb200-nv_13 + - gb200-nv_14 + - gb200-nv_15 + - gb200-nv_16 + - gb200-nv_17 b300: + - b300-nv_00 - b300-nv_01 - b300-nv_02 - b300-nv_03 @@ -181,9 +200,24 @@ labels: - b300-nv_16 - b300-nv_17 gb300: - - gb300-nv_0 - - gb300-nv_1 - - gb300-nv_2 + - gb300-nv_00 + - gb300-nv_01 + - gb300-nv_02 + - gb300-nv_03 + - gb300-nv_04 + - gb300-nv_05 + - gb300-nv_06 + - gb300-nv_07 + - gb300-nv_08 + - gb300-nv_09 + - gb300-nv_10 + - gb300-nv_11 + - gb300-nv_12 + - gb300-nv_13 + - gb300-nv_14 + - gb300-nv_15 + - gb300-nv_16 + - gb300-nv_17 rtx6000pro: - rtx6000pro-lat_00 rtx6000pro-lat: @@ -216,22 +250,23 @@ labels: - h200-cw_00 - h200-cw_01 cluster:h200-dgxc: - - h200-dgxc-slurm_0 - - h200-dgxc-slurm_1 - - h200-dgxc-slurm_2 - - h200-dgxc-slurm_3 - - h200-dgxc-slurm_4 - - h200-dgxc-slurm_5 - - h200-dgxc-slurm_6 - - h200-dgxc-slurm_7 - - h200-dgxc-slurm_8 - - h200-dgxc-slurm_9 + - h200-dgxc-slurm_00 + - h200-dgxc-slurm_01 + - h200-dgxc-slurm_02 + - h200-dgxc-slurm_03 + - h200-dgxc-slurm_04 + - h200-dgxc-slurm_05 + - h200-dgxc-slurm_06 + - h200-dgxc-slurm_07 + - h200-dgxc-slurm_08 + - h200-dgxc-slurm_09 - h200-dgxc-slurm_10 - h200-dgxc-slurm_11 - h200-dgxc-slurm_12 - h200-dgxc-slurm_13 cluster:h200-nb: - - h200-nb_0 + - h200-nb_00 + - h200-nb_01 cluster:b200-dgxc: - b200-dgxc_00 - b200-dgxc_01 @@ -243,13 +278,19 @@ labels: - b200-dgxc_07 - b200-dgxc_08 - b200-dgxc_09 - cluster:b300-cw: - - b300-cw_0 - - b300-cw_01 - - b300-cw_02 - - b300-cw_03 - - b300-cw_04 + cluster:b200-nscale: + - b200-nscale-slurm_00 + - b200-nscale-slurm_01 + - b200-nscale-slurm_02 + - b200-nscale-slurm_03 + - b200-nscale-slurm_04 + - b200-nscale-slurm_05 + - b200-nscale-slurm_06 + - b200-nscale-slurm_07 + - b200-nscale-slurm_08 + - b200-nscale-slurm_09 cluster:b300-nv: + - b300-nv_00 - b300-nv_01 - b300-nv_02 - b300-nv_03 @@ -268,14 +309,43 @@ labels: - b300-nv_16 - b300-nv_17 cluster:gb200-nv: - - gb200-nv_0 - - gb200-nv_1 - - gb200-nv_2 - - gb200-nv_3 + - gb200-nv_00 + - gb200-nv_01 + - gb200-nv_02 + - gb200-nv_03 + - gb200-nv_04 + - gb200-nv_05 + - gb200-nv_06 + - gb200-nv_07 + - gb200-nv_08 + - gb200-nv_09 + - gb200-nv_10 + - gb200-nv_11 + - gb200-nv_12 + - gb200-nv_13 + - gb200-nv_14 + - gb200-nv_15 + - gb200-nv_16 + - gb200-nv_17 cluster:gb300-nv: - - gb300-nv_0 - - gb300-nv_1 - - gb300-nv_2 + - gb300-nv_00 + - gb300-nv_01 + - gb300-nv_02 + - gb300-nv_03 + - gb300-nv_04 + - gb300-nv_05 + - gb300-nv_06 + - gb300-nv_07 + - gb300-nv_08 + - gb300-nv_09 + - gb300-nv_10 + - gb300-nv_11 + - gb300-nv_12 + - gb300-nv_13 + - gb300-nv_14 + - gb300-nv_15 + - gb300-nv_16 + - gb300-nv_17 cluster:rtx6000pro-lat: - rtx6000pro-lat_00 cluster:mi300x-amds: @@ -301,7 +371,6 @@ labels: - mi325x-amds_06 - mi325x-amds_07 - mi325x-amds_08 - - mi325x-amds_09 cluster:mi355x-amds: - mi355x-amds_00 - mi355x-amds_01 @@ -312,6 +381,7 @@ labels: - mi355x-amds_06 - mi355x-amds_07 - mi355x-amds_08 + - mi355x-amds_09 hardware: cluster:h100-dgxc: available-cpu-dram-mib: 2_063_837