diff --git a/custom_ops/gpu_ops/moe/fast_hardamard_kernel.cu b/custom_ops/gpu_ops/moe/fast_hardamard_kernel.cu
index 66d9f72fe..0bda82f29 100644
--- a/custom_ops/gpu_ops/moe/fast_hardamard_kernel.cu
+++ b/custom_ops/gpu_ops/moe/fast_hardamard_kernel.cu
@@ -718,7 +718,6 @@ void MoeFastHardamardImplWrapper(const T *x,
     dim,
     reinterpret_cast<out_type*>(out)
   );
-  CUDA_CHECK(cudaDeviceSynchronize());
 }
 
 template <typename T, typename OutT>