diff --git a/parsec/interfaces/dtd/insert_function.c b/parsec/interfaces/dtd/insert_function.c
index af84dba73..abe5d73df 100644
--- a/parsec/interfaces/dtd/insert_function.c
+++ b/parsec/interfaces/dtd/insert_function.c
@@ -2371,6 +2371,7 @@ int parsec_dtd_task_class_add_chore(parsec_taskpool_t *tp,
     }
 
     incarnations[i].type = device_type;
+    incarnations[i].flags = PARSEC_CHORE_FLAG_NONE;
     if(PARSEC_DEV_CUDA == device_type) {
         incarnations[i].hook = parsec_dtd_gpu_task_submit;
         dtd_tc->gpu_func_ptr = (parsec_advance_task_function_t)function;
diff --git a/parsec/interfaces/ptg/ptg-compiler/jdf2c.c b/parsec/interfaces/ptg/ptg-compiler/jdf2c.c
index 37f7e5f10..e1b3dbf65 100644
--- a/parsec/interfaces/ptg/ptg-compiler/jdf2c.c
+++ b/parsec/interfaces/ptg/ptg-compiler/jdf2c.c
@@ -3954,6 +3954,7 @@ jdf_generate_function_incarnation_list( const jdf_t *jdf,
         }
         string_arena_add_string(sa, "#if defined(PARSEC_HAVE_DEV_%s_SUPPORT)\n", dev_upper);
         string_arena_add_string(sa, "    { .type     = PARSEC_DEV_%s,\n", dev_upper);
+        string_arena_add_string(sa, "      .flags    = PARSEC_CHORE_FLAG_NONE,\n");
         if( NULL == dyld_property ) {
             string_arena_add_string(sa, "      .dyld     = NULL,\n");
         } else {
@@ -3986,7 +3987,7 @@ jdf_generate_function_incarnation_list( const jdf_t *jdf,
     } while (NULL != body);
     string_arena_add_string(sa,
                             "    { .type     = PARSEC_DEV_NONE,\n"
-                            "      .evaluate = NULL,\n"
+                            "      .flags    = PARSEC_CHORE_FLAG_NONE,\n"
                             "      .hook     = (parsec_hook_t*)NULL },  /* End marker */\n"
                             "};\n\n");
 }
@@ -4497,6 +4498,7 @@ static void jdf_generate_startup_hook( const jdf_t *jdf )
             "      idx++;\n"
             "    }\n"
             "    chores[idx].type     = PARSEC_DEV_NONE;\n"
+            "    chores[idx].flags    = PARSEC_CHORE_FLAG_NONE;\n"
             "    chores[idx].evaluate = NULL;\n"
             "    chores[idx].hook     = NULL;\n"
             "    /* Create the initialization tasks for each taskclass */\n"
diff --git a/parsec/mca/device/device_gpu.c b/parsec/mca/device/device_gpu.c
index d57ac904e..77533b324 100644
--- a/parsec/mca/device/device_gpu.c
+++ b/parsec/mca/device/device_gpu.c
@@ -4,6 +4,7 @@
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
  * Copyright (c) 2024      NVIDIA Corporation.  All rights reserved.
+ * Copyright (c) 2026      Stony Brook University.  All rights reserved.
  */
 
 #include "parsec/parsec_config.h"
@@ -2498,6 +2499,16 @@ parsec_device_kernel_cleanout( parsec_device_gpu_module_t *gpu_device,
     return 0;
 }
 
+/**
+ * Returns true if the task's completion should be shifted to worker threads.
+ */
+static bool shift_completed_task(parsec_device_gpu_module_t* gpu_device, parsec_gpu_task_t* gpu_task)
+{
+    parsec_task_t* this_task = gpu_task->ec;
+    const __parsec_chore_t *chore = &this_task->task_class->incarnations[this_task->selected_chore];
+    return (bool)(chore->flags & PARSEC_CHORE_FLAG_SHIFT_COMPLETION);
+}
+
 /**
  * This version is based on 4 streams: one for transfers from the memory to
  * the GPU, 2 for kernel executions and one for transfers from the GPU into
@@ -2703,8 +2714,27 @@ parsec_device_kernel_scheduler( parsec_device_module_t *module,
         goto remove_gpu_task;
     }
     parsec_device_kernel_epilog( gpu_device, gpu_task );
-    __parsec_complete_execution( es, gpu_task->ec );
+
+#if defined(PARSEC_DEBUG_PARANOID)
+    /**
+     * Batched submissions should have been split again for completion but
+     * in case this ever changes we will catch that here.
+     */
+    assert(parsec_gpu_task_is_singleton(gpu_task));
+#endif
+    if (shift_completed_task(gpu_device, gpu_task)) {
+        // ship the task to other threads to complete its execution
+        gpu_task->ec->status = PARSEC_TASK_STATUS_COMPLETE;
+        PARSEC_LIST_ITEM_SINGLETON(gpu_task->ec);
+        __parsec_schedule(es, gpu_task->ec, 1);
+        PARSEC_DEBUG_VERBOSE(10, parsec_gpu_output_stream, "GPU[%d:%s]: task %p of gpu_task %p scheduled for completion",
+                             gpu_device->super.device_index, gpu_device->super.name,
+                             gpu_task->ec, gpu_task);
+    } else {
+        __parsec_complete_execution( es, gpu_task->ec );
+    }
     gpu_device->super.executed_tasks++;
+
  remove_gpu_task:
     PARSEC_DEBUG_VERBOSE(10, parsec_gpu_output_stream, "GPU[%d:%s]: gpu_task %p freed",
                          gpu_device->super.device_index, gpu_device->super.name,
diff --git a/parsec/parsec_internal.h b/parsec/parsec_internal.h
index 214d18c75..ab747edce 100644
--- a/parsec/parsec_internal.h
+++ b/parsec/parsec_internal.h
@@ -392,9 +392,12 @@ int parsec_update_deps_with_counter_count_task(parsec_taskpool_t *tp,
                                                const parsec_task_t* PARSEC_RESTRICT origin,
                                                const parsec_flow_t* PARSEC_RESTRICT origin_flow,
                                                const parsec_flow_t* PARSEC_RESTRICT dest_flow);
-    
+
+#define PARSEC_CHORE_FLAG_NONE                  0x00
+#define PARSEC_CHORE_FLAG_SHIFT_COMPLETION      0x01
 typedef struct __parsec_internal_incarnation_s {
     int32_t                     type;
+    int32_t                     flags;
     parsec_evaluate_function_t *evaluate;
     parsec_hook_t              *hook;
     char                       *dyld;