diff --git a/src/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.cpp b/src/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.cpp
index 5789113..621bd2b 100644
--- a/src/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.cpp
+++ b/src/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.cpp
@@ -63,7 +63,7 @@
 
 std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *bias, ITensorInfo *output)
 {
-    constexpr unsigned int num_elems_processed_per_iteration = 16;
+    constexpr unsigned int num_elems_processed_per_iteration = 4;
 
     // Configure kernel window
     Window win = calculate_max_window(*output, Steps(num_elems_processed_per_iteration));