diff --git a/src/nvfp4_megamoe_kernel/cutlass_nvfp4_gemm/kernel.py b/src/nvfp4_megamoe_kernel/cutlass_nvfp4_gemm/kernel.py
index aad456a3..5396adfe 100644
--- a/src/nvfp4_megamoe_kernel/cutlass_nvfp4_gemm/kernel.py
+++ b/src/nvfp4_megamoe_kernel/cutlass_nvfp4_gemm/kernel.py
@@ -28,10 +28,7 @@ def cutlass_nvfp4_blockscaled_gemm(
     """Single NVFP4 block-scaled GEMM using CUTLASS."""
     if not _CUTLASS_AVAILABLE:
         raise RuntimeError("CUTLASS NVFP4 GEMM extension not available")
-    result = _C.forward(A_packed, SFA, B_packed, SFB, M, N, K, alpha)
-    if result < 0:
-        raise RuntimeError(f"CUTLASS NVFP4 GEMM failed with code {result} — can_implement may have rejected the problem (M={M} N={N} K={K})")
-    return result
+    return _C.forward(A_packed, SFA, B_packed, SFB, M, N, K, alpha)
 
 
 def cutlass_grouped_nvfp4_gemm(