diff --git a/src/modules/module_03200.c b/src/modules/module_03200.c
index b07af2bc9..73a8d3a76 100644
--- a/src/modules/module_03200.c
+++ b/src/modules/module_03200.c
@@ -94,25 +94,32 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
   }
   else
   {
+    u32 overhead = 0;
+
+    if (device_param->device_vendor_id == VENDOR_ID_NV)
+    {
+      // note we need to use device_param->device_local_mem_size - 4 because opencl jit returns with:
+      // Entry function '...' uses too much shared data (0xc004 bytes, 0xc000 max)
+      // on my development system. no clue where the 4 bytes are spent.
+      // I did some research on this and it seems to be related with the datatype.
+      // For example, if i used u8 instead, there's only 1 byte wasted.
+
+      overhead = 4;
+    }
+
     if (user_options->kernel_threads_chgd == true)
     {
       fixed_local_size = user_options->kernel_threads;
+
+      // otherwise out-of-bound reads
+
+      if ((fixed_local_size * 4096) > (device_param->device_local_mem_size - overhead))
+      {
+        fixed_local_size = (device_param->device_local_mem_size - overhead) / 4096;
+      }
     }
     else
     {
-      u32 overhead = 0;
-
-      if (device_param->device_vendor_id == VENDOR_ID_NV)
-      {
-        // note we need to use device_param->device_local_mem_size - 4 because opencl jit returns with:
-        // Entry function '...' uses too much shared data (0xc004 bytes, 0xc000 max)
-        // on my development system. no clue where the 4 bytes are spent.
-        // I did some research on this and it seems to be related with the datatype.
-        // For example, if i used u8 instead, there's only 1 byte wasted.
-
-        overhead = 4;
-      }
-
       fixed_local_size = (device_param->device_local_mem_size - overhead) / 4096;
     }
   }