diff --git a/core/teca_thread_util.cxx b/core/teca_thread_util.cxx
index 5c12dbf18..40b87930e 100644
--- a/core/teca_thread_util.cxx
+++ b/core/teca_thread_util.cxx
@@ -528,16 +528,22 @@ int thread_parameters(MPI_Comm comm, int base_core_id, int n_requested,
 
     // thread pool size is based on core and process count
     int nlg = 0;
+    // map threads to physical cores
+    nlg = cores_per_node % n_procs;
+    n_threads = cores_per_node / n_procs + (proc_id < nlg ? 1 : 0);
     if (n_requested > 0)
     {
-        // user specified override
+        // use exactly this many
         n_threads = n_requested;
     }
+    else if (n_requested < -1)
+    {
+        // use at most this many
+        n_threads = std::min(-n_requested, n_threads);
+        n_requested = n_threads;
+    }
     else
     {
-        // map threads to physical cores
-        nlg = cores_per_node % n_procs;
-        n_threads = cores_per_node/n_procs + (proc_id < nlg ? 1 : 0);
     }
 
     // if the user runs more MPI ranks than cores some of the ranks
diff --git a/core/teca_thread_util.h b/core/teca_thread_util.h
index 33f0db276..c420a8431 100644
--- a/core/teca_thread_util.h
+++ b/core/teca_thread_util.h
@@ -37,8 +37,10 @@ namespace teca_thread_util
  *                        be bound to to acheive this. Passing n_requested >= 1
  *                        specifies a run time override. This indicates that
  *                        caller wants to use a specific number of threads,
- *                        rather than one per physical core. In this case the
- *                        affinity map is also constructed.
+ *                        rather than one per physical core. Passing
+ *                        n_requested < -1 specifies a maximum to use if
+ *                        sufficient cores are available.  In all cases the
+ *                        affinity map is constructed.
  *
  * @param[in] n_threads_per_device the number of threads that should service
  *                                 GPUs. If 0 the run will be CPU only.  If -1
@@ -60,10 +62,11 @@ namespace teca_thread_util
  *                          of threads one can use such that there is one
  *                          thread per phycial core taking into account all
  *                          ranks running on the node. if n_requested is >= 1
- *                          n_threads will be set to n_requested. This allows a
- *                          run time override for cases when the caller knows
- *                          how she wants to schedule things. if an error
- *                          occurs and n_requested is -1 this will be set to 1.
+ *                          n_threads will explicitly be set to n_requested.  If
+ *                          n_requested < -1 at most -n_requested threads will
+ *                          be used. Fewer threads will be used if there are
+ *                          insufficient cores available.  if an error occurs
+ *                          and n_requested is -1 this will be set to 1.
  *
  * @param[out] affinity an affinity map, describing for each of n_threads,
  *                      a core id that the thread can be bound to. if