diff --git a/src/alltoallv.cu b/src/alltoallv.cu
index 34b6462619..bb335eda87 100644
--- a/src/alltoallv.cu
+++ b/src/alltoallv.cu
@@ -181,7 +181,7 @@ testResult_t AlltoAllvRunTest(struct threadArgs* args, int root, ncclDataType_t
   }
 
   for (int i=0; i<type_count; i++) {
-      TESTCHECK(TimeTest(args, run_types[i], run_typenames[i], (ncclRedOp_t)0, "", -1));
+      TESTCHECK(TimeTest(args, run_types[i], run_typenames[i], (ncclRedOp_t)0, "none", -1));
   }
   return testSuccess;
 }
diff --git a/src/common.cu b/src/common.cu
index edcb1d389c..89d9a2bbbc 100644
--- a/src/common.cu
+++ b/src/common.cu
@@ -133,16 +133,6 @@ Reporter::Reporter(std::string fileName, std::string outputFormat) : _outputForm
     if (isMainThread()) {
       _out = std::ofstream(fileName, std::ios_base::out);
       _outputValid = true;
-      if (_outputFormat == "csv") {
-        _out << "numCycle, ";
-        _out << "collective, ";
-#ifdef MPI_SUPPORT
-        _out << "ranks, rankspernode, gpusperrank, ";
-#else
-        _out << "gpus, ";
-#endif
-        _out << "size, type, redop, inplace, time, algbw, busbw, #wrong\n";
-      }
     }
   }
 }
@@ -184,27 +174,54 @@ void Reporter::addResult(int gpusPerRank, int ranksPerNode, int totalRanks, size
   outputValuesKeys.push_back(makeValueKeyPair(busBw, "busBw"));
   outputValuesKeys.push_back(makeValueKeyPair(wrongEltsStr, "wrong"));
 
-  for (auto iter = outputValuesKeys.begin(); iter != outputValuesKeys.end(); ++iter) {
-    if (_outputFormat == "csv") {
-      _out << iter->first;
-      if (std::next(iter) != outputValuesKeys.end()) {
-        _out << ", ";
+  _outputData.push_back(outputValuesKeys);
+}
+
+void Reporter::writeFile() {
+  if (!isMainThread() || !_outputValid)
+    return;
+
+  if (_outputFormat == "csv") {
+    _out << "numCycle,";
+    _out << "collective,";
+#ifdef MPI_SUPPORT
+    _out << "ranks,rankspernode,gpusperrank,";
+#else
+    _out << "gpus,";
+#endif
+    _out << "size,type,redop,inplace,time,algbw,busbw,#wrong\n";
+    for (auto iterEntries = _outputData.begin(); iterEntries != _outputData.end(); ++iterEntries) {
+      for (auto iterVals = (*iterEntries).begin(); iterVals != (*iterEntries).end(); ++iterVals) {
+	_out << iterVals->first;
+	if (std::next(iterVals) != (*iterEntries).end()) {
+	  _out << ",";
+	}
       }
-    } else { //json
-      if (iter == outputValuesKeys.begin()) {
-        _out << "{";
+      _out << std::endl;
+    }
+  } else { //json
+    _out << "[" << std::endl;
+    for (auto iterEntries = _outputData.begin(); iterEntries != _outputData.end(); ++iterEntries) {
+      for (auto iterVals = (*iterEntries).begin(); iterVals != (*iterEntries).end(); ++iterVals) {
+        if (iterVals == (*iterEntries).begin()) {
+          _out << "{";
+        }
+        _out << "\"" << iterVals->second << "\":" << iterVals->first;
+        if (std::next(iterVals) != (*iterEntries).end()) {
+          _out << ", ";
+	}
       }
-      _out << "\"" << iter->second << "\":" << iter->first;
-      if (std::next(iter) != outputValuesKeys.end()) {
-        _out << ", ";
+      if (std::next(iterEntries) != _outputData.end()) {
+        _out << "}," << std::endl;
       } else {
-        _out << "}";
+	_out << "}" << std::endl;
       }
     }
+    _out << "]" << std::endl;
   }
-  _out << std::endl;
 }
 
+
 bool Reporter::isMainThread() { return is_main_thread == 1; }
 
 #define NUM_BLOCKS 32
@@ -1711,6 +1728,8 @@ testResult_t run() {
   MPI_Finalize();
 #endif
 
+  reporter.writeFile();
+
   // 'cuda-memcheck --leak-check full' requires this
   PRINT("%s\n", ncclGetLastError(NULL));
   cudaDeviceReset();
diff --git a/src/common.h b/src/common.h
index 7a961c15a5..97ea1829f5 100644
--- a/src/common.h
+++ b/src/common.h
@@ -22,6 +22,8 @@
 #include <string>
 #include <fstream>
 #include <iostream>
+#include <utility>
+#include <vector>
 
 // Ensures backward compatibility for FP8 datatypes
 #if NCCL_VERSION_CODE < NCCL_VERSION(2,24,3)
@@ -119,6 +121,7 @@ class Reporter {
     ~Reporter() { if (_outputValid) { _out.close(); } };
     void setParameters(const size_t numCycle, const char* name, const char* typeName, const char* opName);
     void addResult(int gpusPerRank, int ranksPerNode, int totalRanks, size_t numBytes, int inPlace, double timeUsec, double algBw, double busBw, int64_t wrongElts = -1);
+    void writeFile();
 
   private:
     bool isMainThread();
@@ -132,6 +135,7 @@ class Reporter {
     std::string _collectiveName;
     std::string _typeName;
     std::string _opName;
+    std::vector<std::vector<std::pair<std::string, std::string>>> _outputData;
 };
 
 struct testEngine {
diff --git a/src/hypercube.cu b/src/hypercube.cu
index ff9201045f..f5d94f026d 100644
--- a/src/hypercube.cu
+++ b/src/hypercube.cu
@@ -101,7 +101,7 @@ testResult_t HyperCubeRunTest(struct threadArgs* args, int root, ncclDataType_t
   int nRanks = args->nProcs*args->nThreads*args->nGpus;
   if (nRanks && !(nRanks & (nRanks - 1))) {
     for (int i=0; i<type_count; i++) {
-      TESTCHECK(TimeTest(args, run_types[i], run_typenames[i], (ncclRedOp_t)0, "", -1));
+      TESTCHECK(TimeTest(args, run_types[i], run_typenames[i], (ncclRedOp_t)0, "none", -1));
     }
   } else {
     printf("nRanks %d is not a power of 2, skipping\n", nRanks);