diff --git a/modules/gpu/src/nvidia/NPP_staging/NPP_staging.cu b/modules/gpu/src/nvidia/NPP_staging/NPP_staging.cu
index af65bbf3e3..21593bbd70 100644
--- a/modules/gpu/src/nvidia/NPP_staging/NPP_staging.cu
+++ b/modules/gpu/src/nvidia/NPP_staging/NPP_staging.cu
@@ -280,7 +280,8 @@ __global__ void scanRows(T_in *d_src, Ncv32u texOffs, Ncv32u srcWidth, Ncv32u sr
 
     __shared__ T_out shmem[NUM_SCAN_THREADS * 2];
     __shared__ T_out carryElem;
-    carryElem = 0;
+    if (threadIdx.x == 0)
+        carryElem = 0;
     __syncthreads();
 
     while (numBuckets--)