diff --git a/src/common.cu b/src/common.cu
index 07f3890..411477c 100644
--- a/src/common.cu
+++ b/src/common.cu
@@ -540,7 +540,7 @@ testResult_t startColl(struct threadArgs* args, ncclDataType_t type, ncclRedOp_t
     // Complete op before returning
     TESTCHECK(testStreamSynchronize(args->nGpus, args->streams, args->comms));
   }
-  if (blocking_coll) Barrier(args);
+  if (blocking_coll == 1) Barrier(args);
   return testSuccess;
 }
 
@@ -1167,7 +1167,7 @@ int main(int argc, char* argv[], char **envp) {
 #endif
             "[-d,--datatype <nccltype/all>] \n\t"
             "[-r,--root <root>] \n\t"
-            "[-z,--blocking <0/1>] \n\t"
+            "[-z,--blocking <0/1/2> 1=wait for completion and barrier (default behavior), 2=wait without barrier] \n\t"
             "[-y,--stream_null <0/1>] \n\t"
             "[-T,--timeout <time in seconds>] \n\t"
             "[-G,--cudagraph <num graph launches>] \n\t"
diff --git a/src/util.cu b/src/util.cu
index 2772841..d2a604b 100644
--- a/src/util.cu
+++ b/src/util.cu
@@ -518,7 +518,8 @@ testResult_t writeDeviceReport(size_t *maxMem, int localRank, int proc, int tota
         nThreads, nGpus, minBytes, maxBytes,
         (stepFactor > 1)?stepFactor:stepBytes, (stepFactor > 1)?"factor":"bytes",
         warmup_iters, iters, agg_iters, datacheck, cudaGraphLaunches);
-  if (blocking_coll) PRINT("# Blocking Enabled: wait for completion and barrier after each collective \n");
+  if (blocking_coll == 1) PRINT("# Blocking Enabled: wait for completion and barrier after each collective \n");
+  if (blocking_coll > 1)  PRINT("# Blocking Enabled: wait for completion after each collective (no barrier) \n");
   if (parallel_init) PRINT("# Parallel Init Enabled: threads call into NcclInitRank concurrently \n");
   PRINT("#\n");