diff --git a/tensorrt_llm/_torch/models/modeling_qwen3_moe.py b/tensorrt_llm/_torch/models/modeling_qwen3_moe.py
index 221478edf8..ba7f6026ed 100644
--- a/tensorrt_llm/_torch/models/modeling_qwen3_moe.py
+++ b/tensorrt_llm/_torch/models/modeling_qwen3_moe.py
@@ -5,6 +5,7 @@ import torch
 from torch import nn
 from transformers import Qwen3MoeConfig
 
+from tensorrt_llm._ipc_utils import can_access_peer
 from tensorrt_llm._torch.models.checkpoints.base_weight_mapper import \
     BaseWeightMapper
 
@@ -187,6 +188,8 @@ class Qwen3MoEDecoderLayer(DecoderLayer):
                                    strategy=model_config.allreduce_strategy)
         self.next_layer_layernorm: RMSNorm = None
 
+        self.is_p2p_supported = can_access_peer(model_config.mapping)
+
         self.fusion_config = EagerFusionConfig()
         self.enable_fusion = os.environ.get(
             "TRTLLM_QWEN3_EAGER_FUSION_DISABLED", "0") == "0"
@@ -242,11 +245,11 @@ class Qwen3MoEDecoderLayer(DecoderLayer):
                 hidden_states, residual)
 
         # Note: this fusion pattern is only supported for TRTLLM-nvfp4 backend now
-        do_finalize = not (hidden_states.shape[0]
-                           <= self.moe_allreduce.max_token
-                           and self.fusion_config.POST_MOE_FUSION
-                           and self.model_config.moe_backend == 'TRTLLM'
-                           and self.mlp.experts.has_nvfp4)
+        do_finalize = not (
+            hidden_states.shape[0] <= self.moe_allreduce.max_token
+            and self.fusion_config.POST_MOE_FUSION
+            and self.model_config.moe_backend == 'TRTLLM'
+            and self.mlp.experts.has_nvfp4 and self.is_p2p_supported)
 
         hidden_states = self.mlp(
             hidden_states,
diff --git a/tests/integration/defs/accuracy/test_llm_api_pytorch.py b/tests/integration/defs/accuracy/test_llm_api_pytorch.py
index d8d77b7d37..ef02fcd9c6 100644
--- a/tests/integration/defs/accuracy/test_llm_api_pytorch.py
+++ b/tests/integration/defs/accuracy/test_llm_api_pytorch.py
@@ -2876,13 +2876,17 @@ class TestQwen3_235B_A22B(LlmapiAccuracyTestHarness):
         "tp_size,pp_size,ep_size,attention_dp,cuda_graph,overlap_scheduler,moe_backend,eagle3",
         [
             (8, 1, 8, True, True, True, "CUTLASS", False),
+            (8, 1, 8, False, True, True, "TRTLLM", False),
             (8, 1, 8, True, True, True, "TRTLLM", False),
+            (8, 1, 8, False, True, True, "TRTLLM", True),
             (8, 1, 8, True, True, True, "TRTLLM", True),
         ],
         ids=[
             "latency_moe_cutlass",
             "latency_moe_trtllm",
+            "latency_moe_trtllm_attention_dp",
             "latency_moe_trtllm_eagle3",
+            "latency_moe_trtllm_eagle3_attention_dp",
         ],
     )
     def test_nvfp4(self, tp_size, pp_size, ep_size, attention_dp, cuda_graph,
diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt
index c1ab5b3a1c..e445478bbf 100644
--- a/tests/integration/test_lists/qa/llm_function_core.txt
+++ b/tests/integration/test_lists/qa/llm_function_core.txt
@@ -582,6 +582,7 @@ accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_fp8[throughput_laten
 accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_fp8[latency]
 accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_cutlass]
 accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm]
+accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm_attention_dp]
 accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4_4gpus[latency_moe_trtllm_eagle3]
 accuracy/test_llm_api_pytorch.py::TestKanana_Instruct::test_auto_dtype
 accuracy/test_llm_api_pytorch.py::TestBielik11BInstruct::test_auto_dtype
diff --git a/tests/integration/test_lists/test-db/l0_gb200_multi_nodes.yml b/tests/integration/test_lists/test-db/l0_gb200_multi_nodes.yml
index 2c1d315b99..cae2fc365d 100644
--- a/tests/integration/test_lists/test-db/l0_gb200_multi_nodes.yml
+++ b/tests/integration/test_lists/test-db/l0_gb200_multi_nodes.yml
@@ -35,4 +35,5 @@ l0_gb200_multi_nodes:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_trtllmgen] TIMEOUT (180)
   - accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_cutlass] TIMEOUT (90)
   - accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm] TIMEOUT (90)
+  - accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm_attention_dp] TIMEOUT (90)
   - accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_bf16[multi_gpus_no_cache] TIMEOUT (180)
diff --git a/tests/integration/test_lists/waives.txt b/tests/integration/test_lists/waives.txt
index c9693c512e..27903f1c0a 100644
--- a/tests/integration/test_lists/waives.txt
+++ b/tests/integration/test_lists/waives.txt
@@ -274,6 +274,7 @@ examples/test_phi.py::test_phi_fp8_with_bf16_lora[Phi-3-small-128k-instruct] SKI
 examples/test_phi.py::test_phi_fp8_with_bf16_lora[Phi-3.5-mini-instruct] SKIP (https://nvbugs/5465143)
 examples/test_phi.py::test_phi_fp8_with_bf16_lora[Phi-4-mini-instruct] SKIP (https://nvbugs/5465143)
 examples/test_llama.py::test_llm_llama_v1_2gpu_summary[llama-7b-nb:4-enable_auto_parallel] SKIP (https://nvbugs/5465173)
+test_e2e.py::test_ptp_quickstart_multimodal[llava-v1.6-mistral-7b-llava-v1.6-mistral-7b-hf-image-False] SKIP (https://nvbugs/5444095)
 full:GB200/examples/test_qwen.py::test_llm_qwen_7b_multi_gpus_summary[qwen1.5_7b_chat-enable_fmha_fp32_acc-enable_plugin-tp2pp2-nb:4] SKIP (https://nvbugs/5247837)
 full:GB200/examples/test_qwen.py::test_llm_qwen_7b_multi_gpus_summary[qwen2_7b_instruct-enable_fmha_fp32_acc-enable_plugin-tp2pp2-nb:4] SKIP (https://nvbugs/5247837)
 full:GB200/examples/test_qwen.py::test_llm_qwen_7b_multi_gpus_summary[qwen2_vl_7b_instruct-enable_fmha_fp32_acc-enable_plugin-tp2pp2-nb:4] SKIP (https://nvbugs/5359696)