diff --git a/tests/regression/idle/kernel.cpp b/tests/regression/idle/kernel.cpp
index 3d3f0c6f..ccd9bcc5 100644
--- a/tests/regression/idle/kernel.cpp
+++ b/tests/regression/idle/kernel.cpp
@@ -12,23 +12,28 @@
 #define HW_TID() ({uint32_t gtid; asm volatile ("csrr %0, mhartid" : "=r" (gtid)); gtid;})
 
 void kernel_body(int task_id, kernel_arg_t *__UNIFORM__ arg) {
-  constexpr uint32_t timer = 50000;
-  uint32_t counter = 0;
-  while ((counter++) < timer) {
-    asm("");
-  }
+  // constexpr uint32_t timer = 50000;
+  // uint32_t counter = 0;
+  // while ((counter++) < timer) {
+  //   asm("");
+  // }
+  //
   // to prevent optimize-out
   // reinterpret_cast<uint32_t *>(arg->addr_c)[0] = counter;
 
+  // call barrier in a divergent branch, which will hang the core
+  if ((vx_thread_id() % NUM_THREADS) == 0) {
+    vx_barrier(0, NUM_WARPS);
+  }
+
   vx_tmc(0);
 }
 
 int main() {
   kernel_arg_t *arg = (kernel_arg_t *)KERNEL_ARG_DEV_MEM_ADDR;
 
-  // const uint32_t num_threads_in_cluster = NUM_THREADS_IN_CLUSTER;
-  // const uint32_t grid_size = num_threads_in_cluster * NUM_CLUSTERS;
-  const uint32_t grid_size = 1;
+  // spawn a single warp in every core
+  const uint32_t grid_size = NUM_THREADS * NUM_CORES;
 #ifdef RADIANCE
   vx_spawn_tasks_cluster(grid_size, (vx_spawn_tasks_cb)kernel_body, arg);
 #else