diff --git a/scripts/performance/configs/nemotronh/nemotron_3_workload_base_configs.py b/scripts/performance/configs/nemotronh/nemotron_3_workload_base_configs.py
index 2cafda9789..18ed8d8096 100644
--- a/scripts/performance/configs/nemotronh/nemotron_3_workload_base_configs.py
+++ b/scripts/performance/configs/nemotronh/nemotron_3_workload_base_configs.py
@@ -41,7 +41,7 @@
 
 NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB300_BF16_V1 = replace(
     BASE_NEMOTRON_3_NANO_CONFIG,
-    micro_batch_size=4,
+    micro_batch_size=2,
 )
 NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB300_FP8_MX_V1 = NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB300_BF16_V1
 NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB300_NVFP4_V1 = NEMOTRON_3_NANO_PRETRAIN_CONFIG_GB300_BF16_V1
@@ -59,7 +59,7 @@
 
 NEMOTRON_3_NANO_PRETRAIN_CONFIG_B300_BF16_V1 = replace(
     BASE_NEMOTRON_3_NANO_CONFIG,
-    micro_batch_size=4,
+    micro_batch_size=2,
 )
 NEMOTRON_3_NANO_PRETRAIN_CONFIG_B300_FP8_MX_V1 = NEMOTRON_3_NANO_PRETRAIN_CONFIG_B300_BF16_V1
 NEMOTRON_3_NANO_PRETRAIN_CONFIG_B300_NVFP4_V1 = NEMOTRON_3_NANO_PRETRAIN_CONFIG_B300_BF16_V1
diff --git a/scripts/performance/configs/qwen/qwen3_workload_base_configs.py b/scripts/performance/configs/qwen/qwen3_workload_base_configs.py
index 9cb42e6722..d9397fc7c9 100644
--- a/scripts/performance/configs/qwen/qwen3_workload_base_configs.py
+++ b/scripts/performance/configs/qwen/qwen3_workload_base_configs.py
@@ -325,7 +325,7 @@
 QWEN3_30B_A3B_PRETRAIN_CONFIG_GB300_BF16_V1 = replace(
     BASE_QWEN3_30B_A3B_CONFIG,
     num_gpus=8,
-    micro_batch_size=8,
+    micro_batch_size=4,
     moe_flex_dispatcher_backend="hybridep",
     cuda_graph_impl="transformer_engine",
     cuda_graph_scope=["moe_router", "moe_preprocess"],
@@ -335,7 +335,7 @@
 QWEN3_30B_A3B_PRETRAIN_CONFIG_GB300_FP8_CS_V1 = replace(
     BASE_QWEN3_30B_A3B_CONFIG,
     num_gpus=8,
-    micro_batch_size=8,
+    micro_batch_size=4,
     moe_flex_dispatcher_backend="hybridep",
     cuda_graph_impl="transformer_engine",
     cuda_graph_scope=["moe_router", "moe_preprocess"],
@@ -382,7 +382,7 @@
 QWEN3_30B_A3B_PRETRAIN_CONFIG_B300_BF16_V1 = replace(
     BASE_QWEN3_30B_A3B_CONFIG,
     num_gpus=8,
-    micro_batch_size=8,
+    micro_batch_size=4,
     moe_flex_dispatcher_backend="hybridep",
     cuda_graph_impl="transformer_engine",
     cuda_graph_scope=["moe_router", "moe_preprocess"],
@@ -392,7 +392,7 @@
 QWEN3_30B_A3B_PRETRAIN_CONFIG_B300_FP8_CS_V1 = replace(
     BASE_QWEN3_30B_A3B_CONFIG,
     num_gpus=8,
-    micro_batch_size=8,
+    micro_batch_size=4,
     moe_flex_dispatcher_backend="hybridep",
     cuda_graph_impl="transformer_engine",
     cuda_graph_scope=["moe_router", "moe_preprocess"],
diff --git a/scripts/performance/configs/qwen_vl/qwen3_vl_workload_base_configs.py b/scripts/performance/configs/qwen_vl/qwen3_vl_workload_base_configs.py
index e81cef05f4..07e9934d9a 100644
--- a/scripts/performance/configs/qwen_vl/qwen3_vl_workload_base_configs.py
+++ b/scripts/performance/configs/qwen_vl/qwen3_vl_workload_base_configs.py
@@ -144,7 +144,7 @@
 QWEN3_VL_30B_A3B_PRETRAIN_CONFIG_GB300_BF16 = replace(
     BASE_QWEN3_VL_30B_A3B_CONFIG,
     num_gpus=8,
-    micro_batch_size=8,
+    micro_batch_size=4,
     moe_flex_dispatcher_backend="hybridep",
     cuda_graph_impl="transformer_engine",
     cuda_graph_scope=["moe_router", "moe_preprocess"],
@@ -154,7 +154,7 @@
 QWEN3_VL_30B_A3B_PRETRAIN_CONFIG_GB300_FP8_CS = replace(
     BASE_QWEN3_VL_30B_A3B_CONFIG,
     num_gpus=8,
-    micro_batch_size=8,
+    micro_batch_size=4,
     moe_flex_dispatcher_backend="hybridep",
     cuda_graph_impl="transformer_engine",
     cuda_graph_scope=["moe_router", "moe_preprocess"],