diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/blockScaleMoe/DevKernel.cu b/cpp/tensorrt_llm/kernels/trtllmGenKernels/blockScaleMoe/DevKernel.cu
index 65cd5f3c59..0115f3b018 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/blockScaleMoe/DevKernel.cu
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/blockScaleMoe/DevKernel.cu
@@ -186,14 +186,14 @@ void run(Data const& data, void* stream)
     if (data.mUseDeepSeekFp8)
     {
         int const numThreads = 128;
-        const dim3 grid(data.innerDim / 128, data.topK, data.numTokens);
+        const dim3 grid(data.innerDim / 128, data.topK, std::min(8192, data.numTokens));
 
         LAUNCH(data, activationDeepSeekKernel, grid, numThreads, 0, stream);
     }
     else
     {
         int const numThreads = 256;
-        const dim3 grid(data.innerDim / 128, data.topK, data.numTokens);
+        const dim3 grid(data.innerDim / 128, data.topK, std::min(8192, data.numTokens));
 
         LAUNCH(data, activationKernel, grid, numThreads, 0, stream);
     }
@@ -371,7 +371,7 @@ void run(Data const& data, void* stream)
     constexpr int VecSize = 4;
     int const numThreads = 128;
     int const numBlocksX = (data.hiddenDimSf / VecSize - 1 + numThreads) / numThreads;
-    int const numBlocksY = data.numTokens;
+    int const numBlocksY = std::min(8192, data.numTokens);
     dim3 numBlocks(numBlocksX, numBlocksY);
 #define CONVERT_FP4_SF_LAUNCH(LayoutSrc, LayoutDst)                                                                    \
     if (data.sfLayoutSrc == tg::SfLayout::LayoutSrc && data.sfLayoutDst == tg::SfLayout::LayoutDst)                    \
@@ -457,7 +457,7 @@ void run(Data const& data, void* stream)
 {
     int const numThreads = 256;
     int const numBlocksX = (data.hiddenDim - 1 + numThreads) / numThreads;
-    int const numBlocksY = data.numTokens;
+    int const numBlocksY = std::min(8192, data.numTokens);
     dim3 numBlocks(numBlocksX, numBlocksY);
 
     LAUNCH(data, permuteKernel, numBlocks, numThreads, 0, stream);
diff --git a/jenkins/L0_Test.groovy b/jenkins/L0_Test.groovy
index ccd86a937c..9ec7e2821d 100644
--- a/jenkins/L0_Test.groovy
+++ b/jenkins/L0_Test.groovy
@@ -2093,10 +2093,11 @@ def launchTestJobs(pipeline, testFilter)
 
     multiNodesSBSAConfigs = [
         // Each stage test 1 testcase with 8 GPUs and 2 nodes.
-        "GB200-8_GPUs-2_Nodes-PyTorch-1": ["gb200-multi-node", "l0_gb200_multi_nodes", 1, 4, 8, 2],
-        "GB200-8_GPUs-2_Nodes-PyTorch-2": ["gb200-multi-node", "l0_gb200_multi_nodes", 2, 4, 8, 2],
-        "GB200-8_GPUs-2_Nodes-PyTorch-3": ["gb200-multi-node", "l0_gb200_multi_nodes", 3, 4, 8, 2],
-        "GB200-8_GPUs-2_Nodes-PyTorch-4": ["gb200-multi-node", "l0_gb200_multi_nodes", 4, 4, 8, 2],
+        "GB200-8_GPUs-2_Nodes-PyTorch-1": ["gb200-multi-node", "l0_gb200_multi_nodes", 1, 5, 8, 2],
+        "GB200-8_GPUs-2_Nodes-PyTorch-2": ["gb200-multi-node", "l0_gb200_multi_nodes", 2, 5, 8, 2],
+        "GB200-8_GPUs-2_Nodes-PyTorch-3": ["gb200-multi-node", "l0_gb200_multi_nodes", 3, 5, 8, 2],
+        "GB200-8_GPUs-2_Nodes-PyTorch-4": ["gb200-multi-node", "l0_gb200_multi_nodes", 4, 5, 8, 2],
+        "GB200-8_GPUs-2_Nodes-PyTorch-5": ["gb200-multi-node", "l0_gb200_multi_nodes", 5, 5, 8, 2],
         "GB200-8_GPUs-2_Nodes-PyTorch-Post-Merge-1": ["gb200-multi-node", "l0_gb200_multi_nodes", 1, 5, 8, 2],
         "GB200-8_GPUs-2_Nodes-PyTorch-Post-Merge-2": ["gb200-multi-node", "l0_gb200_multi_nodes", 2, 5, 8, 2],
         "GB200-8_GPUs-2_Nodes-PyTorch-Post-Merge-3": ["gb200-multi-node", "l0_gb200_multi_nodes", 3, 5, 8, 2],
diff --git a/tests/integration/defs/accuracy/test_llm_api_pytorch.py b/tests/integration/defs/accuracy/test_llm_api_pytorch.py
index 2d8f80f54c..be178923c9 100644
--- a/tests/integration/defs/accuracy/test_llm_api_pytorch.py
+++ b/tests/integration/defs/accuracy/test_llm_api_pytorch.py
@@ -1918,18 +1918,25 @@ class TestDeepSeekR1(LlmapiAccuracyTestHarness):
     @pytest.mark.skip_less_mpi_world_size(8)
     @skip_pre_hopper
     @pytest.mark.parametrize(
-        "tp_size,pp_size,ep_size,mtp_nextn,fp8kv,attention_dp,cuda_graph,overlap_scheduler,max_batch_size",
-        [(8, 1, 4, 3, False, False, True, True, 1),
-         (8, 1, 8, 0, True, True, True, True, 24),
-         (8, 1, 8, 1, True, True, True, True, 24)],
-        ids=["latency", "throughput", "throughput_mtp"])
+        "tp_size,pp_size,ep_size,mtp_nextn,fp8kv,attention_dp,cuda_graph,overlap_scheduler,max_batch_size,moe_backend",
+        [(8, 1, 4, 3, False, False, True, True, 1, "_DEFAULT"),
+         (8, 1, 8, 0, True, True, True, True, 24, "_DEFAULT"),
+         (8, 1, 8, 1, True, True, True, True, 24, "_DEFAULT"),
+         (8, 1, 8, 1, True, True, True, True, 24, "TRTLLM")],
+        ids=[
+            "latency", "throughput", "throughput_mtp", "throughput_mtp_trtllm"
+        ])
     def test_fp8_blockscale(self, tp_size, pp_size, ep_size, mtp_nextn, fp8kv,
                             attention_dp, cuda_graph, overlap_scheduler,
-                            max_batch_size):
+                            max_batch_size, moe_backend):
+
         if get_sm_version() == 100:
-            moe_config = MoeConfig(backend="DEEPGEMM", max_num_tokens=16384)
+            moe_backend = "DEEPGEMM" if moe_backend == "_DEFAULT" else moe_backend
+            moe_config = MoeConfig(backend=moe_backend, max_num_tokens=16384)
             kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.6)
         else:
+            if moe_backend != "_DEFAULT":
+                pytest.skip("Not supported MoE backend!")
             moe_config = MoeConfig()
             kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.9)
 
diff --git a/tests/integration/test_lists/test-db/l0_dgx_b200.yml b/tests/integration/test_lists/test-db/l0_dgx_b200.yml
index 5d05d14159..924f3026cf 100644
--- a/tests/integration/test_lists/test-db/l0_dgx_b200.yml
+++ b/tests/integration/test_lists/test-db/l0_dgx_b200.yml
@@ -71,6 +71,7 @@ l0_dgx_b200:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_mtp] TIMEOUT (180)
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput] TIMEOUT (180)
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput_mtp] TIMEOUT (180)
+  - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput_mtp_trtllm] TIMEOUT (180)
 - condition:
     ranges:
       system_gpu_count:
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 7cab7535b8..95528a7258 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
@@ -17,6 +17,7 @@ l0_gb200_multi_nodes:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_mtp] TIMEOUT (180)
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput] TIMEOUT (180)
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput_mtp] TIMEOUT (180)
+  - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput_mtp_trtllm] TIMEOUT (180)
 - condition:
     ranges:
       # 2 nodes with each node has 4 GPUs