diff --git a/benchmarks/kernels/benchmark_cutlass_moe_nvfp4.py b/benchmarks/kernels/benchmark_cutlass_fp4_moe.py
similarity index 98%
rename from benchmarks/kernels/benchmark_cutlass_moe_nvfp4.py
rename to benchmarks/kernels/benchmark_cutlass_fp4_moe.py
index d6b5820a5b41..7982cbb1422c 100644
--- a/benchmarks/kernels/benchmark_cutlass_moe_nvfp4.py
+++ b/benchmarks/kernels/benchmark_cutlass_fp4_moe.py
@@ -21,7 +21,6 @@
 from vllm.model_executor.layers.fused_moe.fused_moe import fused_experts, fused_topk
 from vllm.scalar_type import scalar_types
 from vllm.utils.argparse_utils import FlexibleArgumentParser
-from vllm.v1.worker.workspace import init_workspace_manager
 
 WEIGHT_SHAPES_MOE = {
     "nvidia/DeepSeek-R1-FP4": [
@@ -442,10 +441,6 @@ def replay_graph(graph, num_repeats):
 
 
 def main(args):
-    # Initialize workspace manager (required for CUTLASS MoE kernels)
-    device = torch.device("cuda:0")
-    init_workspace_manager(device)
-
     print("Benchmarking models:")
     for i, model in enumerate(args.models):
         print(f"[{i}]  {model}")
diff --git a/benchmarks/kernels/benchmark_cutlass_moe_fp8.py b/benchmarks/kernels/benchmark_cutlass_moe_fp8.py
index 626b3b160044..e07d6c776bc0 100644
--- a/benchmarks/kernels/benchmark_cutlass_moe_fp8.py
+++ b/benchmarks/kernels/benchmark_cutlass_moe_fp8.py
@@ -15,7 +15,6 @@
 from vllm.model_executor.layers.fused_moe.fused_moe import fused_experts, fused_topk
 from vllm.platforms import current_platform
 from vllm.utils.argparse_utils import FlexibleArgumentParser
-from vllm.v1.worker.workspace import init_workspace_manager
 
 # Weight shapes for different models: [num_experts, topk, hidden_size,
 # intermediate_size]
@@ -298,10 +297,6 @@ def bench_cuda_graph(graph, num_warmup=5, num_iters=100):
 
 
 def main(args):
-    # Initialize workspace manager (required for CUTLASS MoE kernels)
-    device = torch.device("cuda:0")
-    init_workspace_manager(device)
-
     print("Benchmarking models:")
     for i, model in enumerate(args.models):
         print(f"[{i}]  {model}")
diff --git a/benchmarks/kernels/benchmark_grouped_gemm_cutlass.py b/benchmarks/kernels/benchmark_grouped_gemm_cutlass.py
index 4390be8770c1..9b426d8d5f77 100644
--- a/benchmarks/kernels/benchmark_grouped_gemm_cutlass.py
+++ b/benchmarks/kernels/benchmark_grouped_gemm_cutlass.py
@@ -14,7 +14,6 @@
     fused_topk,
 )
 from vllm.utils.argparse_utils import FlexibleArgumentParser
-from vllm.v1.worker.workspace import init_workspace_manager
 
 DEFAULT_MODELS = [
     "mistralai/Mixtral-8x7B-Instruct-v0.1",
@@ -365,10 +364,6 @@ def replay_graph(graph, num_repeats):
 
 
 def main(args):
-    # Initialize workspace manager (required for CUTLASS MoE kernels)
-    device = torch.device("cuda:0")
-    init_workspace_manager(device)
-
     print("Benchmarking models:")
     for i, model in enumerate(args.models):
         print(f"[{i}]  {model}")