diff --git a/examples_cuda/sort/sort.cpp b/examples_cuda/sort/sort.cpp
index 1221013f..c439fdd0 100644
--- a/examples_cuda/sort/sort.cpp
+++ b/examples_cuda/sort/sort.cpp
@@ -108,7 +108,7 @@ int main (int argc, char *argv[])
         progressbar (i, m);
   }
 
-  printf("[sort ispc + task]:\t[%.3f] million cycles\n", tISPC2);
+  printf("[sort ispc + task]:\t[%.3f] million cycles :: rate= %g Mel/sec\n", tISPC2, 1.0e-6*n*m/tISPC2);
 
   for (int i = 0; i < n-1; i++)
   {
@@ -133,7 +133,7 @@ int main (int argc, char *argv[])
         progressbar (i, m);
   }
 
-  printf("[sort ispc]:\t[%.3f] million cycles\n", tISPC1);
+  printf("[sort ispc]:\t[%.3f] million cycles :: rate= %g Mel/sec\n", tISPC1, 1.0e-6*n*m/tISPC1);
 
 
   srand (0);
@@ -153,7 +153,7 @@ int main (int argc, char *argv[])
         progressbar (i, m);
   }
 
-  printf("[sort serial]:\t\t[%.3f] million cycles\n", tSerial);
+  printf("[sort serial]:\t[%.3f] million cycles :: rate= %g Mel/sec\n", tSerial, 1.0e-6*n*m/tSerial);
 
   printf("\t\t\t\t(%.2fx speedup from ISPC, %.2fx speedup from ISPC + tasks)\n", tSerial/tISPC1, tSerial/tISPC2);
 
diff --git a/examples_cuda/sort/sort1.ispc b/examples_cuda/sort/sort1.ispc
index 855aa194..608d851a 100644
--- a/examples_cuda/sort/sort1.ispc
+++ b/examples_cuda/sort/sort1.ispc
@@ -219,12 +219,16 @@ export void sort_ispc (uniform int n,
 {
   uniform int num = ntasks;
   uniform int span = n / num;
-#if 0
+#if 1 
+  /* if fails, change to 0. some issues with stack size/heap inside CUDA context
+   * grep for CuCtxSetLimit in  sort_cu.cpp 
+   */ 
   uniform int hsize = 256*programCount*num;
   uniform int * uniform hist = uniform new uniform int [hsize];
   uniform int64 * uniform pair = uniform new uniform int64 [n];
   uniform int64 * uniform temp = uniform new uniform int64 [n];
   uniform int * uniform g = uniform new uniform int [num+1];
+#define ALLOCATED
 #else
    uniform int * uniform hist = _hist;
    uniform int64 * uniform pair = _pair;
@@ -253,7 +257,7 @@ export void sort_ispc (uniform int n,
   launch[num] unpack (span, n, pair, code, order);
   sync;
 
-#if ALLOCATED
+#ifdef ALLOCATED
   delete g;
   delete hist;
   delete pair;
diff --git a/examples_cuda/sort/sort_cu.cpp b/examples_cuda/sort/sort_cu.cpp
index ed206499..d5d77b6e 100644
--- a/examples_cuda/sort/sort_cu.cpp
+++ b/examples_cuda/sort/sort_cu.cpp
@@ -99,7 +99,13 @@ void createContext(const int deviceId = 0)
 
   // Create driver context
   checkCudaErrors(cuCtxCreate(&context, 0, device));
+  size_t limit;
+  checkCudaErrors(cuCtxGetLimit(&limit, CU_LIMIT_STACK_SIZE));
+  fprintf(stderr, " stack_limit= %llu KB\n", limit/1024);
+  checkCudaErrors(cuCtxGetLimit(&limit, CU_LIMIT_MALLOC_HEAP_SIZE));
+  fprintf(stderr, " heap= %llu KB\n", limit/1024);
   checkCudaErrors(cuCtxSetLimit(CU_LIMIT_MALLOC_HEAP_SIZE,1024*1024*1024));
+  checkCudaErrors(cuCtxSetLimit(CU_LIMIT_STACK_SIZE,1024*4));
 }
 void destroyContext()
 {
@@ -386,7 +392,7 @@ int main (int argc, char *argv[])
         progressbar (i, m);
   }
 
-  printf("[sort cuda]:\t[%.3f] million cycles\n", tISPC2);
+  printf("[sort cuda]:\t[%.3f] million cycles :: rate= %g Mel/sec\n", tISPC2, 1.0e-6*n*m/tISPC2);
   memcpyD2H(code,  d_code,  n*sizeof(int));
   memcpyD2H(order, d_order, n*sizeof(int));
   for (int i = 0; i < n-1; i++)