diff --git a/CMakeLists.txt b/CMakeLists.txt
index 0031296..173f599 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -39,6 +39,7 @@ if(CMAKE_CURRENT_SOURCE_DIR STREQUAL CMAKE_SOURCE_DIR)
     GIT_TAG        main
     CMAKE_ARGS     -DCMAKE_BUILD_TYPE=Release
   )
+  set(BENCHMARK_ENABLE_TESTING OFF CACHE BOOL "" FORCE)
   FetchContent_MakeAvailable(google_benchmark)
 endif()
 
diff --git a/examples/pi_cuda.cu b/examples/pi_cuda.cu
index dc9e427..5936f0c 100644
--- a/examples/pi_cuda.cu
+++ b/examples/pi_cuda.cu
@@ -32,7 +32,10 @@
  */
 
 #include <curand_kernel.h>
+#include <openrand/phillox.h>
 #include <openrand/tyche.h>
+#include <openrand/threefry.h>
+#include <openrand/squares.h>
 
 #include <cmath>
 #include <iostream>
@@ -42,7 +45,7 @@ const int SAMPLES_PER_THREAD = 1000;          // Number of samples per thread
 const int NTHREADS = N / SAMPLES_PER_THREAD;  // Number of threads
 const int THREADS_PER_BLOCK = 256;            // Number of threads per block
 
-typedef openrand::Tyche RNG;
+typedef openrand::Phillox RNG;
 
 __global__ void monteCarloPi(int *d_sum) {
   int idx = blockIdx.x * blockDim.x + threadIdx.x;
@@ -77,9 +80,11 @@ int main() {
   int h_sum;
   cudaMemcpy(&h_sum, d_sum, sizeof(int), cudaMemcpyDeviceToHost);
 
-  float pi = 4.0 * (float)h_sum / N;
+  double pi_estimate = 4.0 * (float)h_sum / N;
 
-  std::cout << "Approximated value of Pi: " << pi << std::endl;
+  constexpr double pi = 3.14159265358979323846;
+  std::cout << "pi_estimate: " << pi_estimate << std::endl;
+  std::cout << "log10(|pi - pi_estimate|): " << std::log10(std::abs(pi - pi_estimate)) << std::endl;
 
   cudaFree(d_sum);
 
diff --git a/examples/pi_openmp.cpp b/examples/pi_openmp.cpp
index 656213c..74eac6b 100644
--- a/examples/pi_openmp.cpp
+++ b/examples/pi_openmp.cpp
@@ -70,6 +70,8 @@ int main() {
   double pi_estimate = compute_pi();
 
   constexpr double pi = 3.14159265358979323846;
+
+  std::cout << "pi_estimate: " << pi_estimate << std::endl;
   std::cout << "log10(|pi - pi_estimate|): " << std::log10(std::abs(pi - pi_estimate)) << std::endl;
 
   return 0;
diff --git a/include/openrand/phillox.h b/include/openrand/phillox.h
index 8e4b114..cd33f8d 100644
--- a/include/openrand/phillox.h
+++ b/include/openrand/phillox.h
@@ -35,30 +35,10 @@
 #include <iostream>
 #include <limits>
 
-namespace {
-
-constexpr uint32_t PHILOX_M4x32_0 = 0xD2511F53;
-constexpr uint32_t PHILOX_M4x32_1 = 0xCD9E8D57;
-constexpr uint32_t PHILOX_W32_0 = 0x9E3779B9;
-constexpr uint32_t PHILOX_W32_1 = 0xBB67AE85;
-
-inline DEVICE uint32_t mulhilo(uint32_t L, uint32_t R, uint32_t *hip) {
-  uint64_t product = static_cast<uint64_t>(L) * static_cast<uint64_t>(R);
-  *hip = product >> 32;
-  return static_cast<uint32_t>(product);
-}
-
-inline DEVICE void round(const uint32_t (&key)[2], uint32_t (&ctr)[4]) {
-  uint32_t hi0;
-  uint32_t hi1;
-  uint32_t lo0 = mulhilo(PHILOX_M4x32_0, ctr[0], &hi0);
-  uint32_t lo1 = mulhilo(PHILOX_M4x32_1, ctr[2], &hi1);
-  ctr[0] = hi1 ^ ctr[1] ^ key[0];
-  ctr[1] = lo1;
-  ctr[2] = hi0 ^ ctr[3] ^ key[1];
-  ctr[3] = lo0;
-}
-}  // namespace
+#define PHILOX_W0 0x9E3779B9
+#define PHILOX_W1 0xBB67AE85
+#define PHILOX_M0 0xD2511F53
+#define PHILOX_M1 0xCD9E8D57
 
 namespace openrand {
 
@@ -66,6 +46,7 @@ namespace openrand {
  * @class Phillox
  * @brief Phillox generator
  * @note This is a modified version of Phillox generator from Random123 library.
+ * This uses 4x 32-bit counter, 2x 32-bit key along with 10 rounds.
  */
 class Phillox : public BaseRNG<Phillox> {
  public:
@@ -118,14 +99,31 @@ class Phillox : public BaseRNG<Phillox> {
 
     for (int r = 0; r < 10; r++) {
       if (r > 0) {
-        key[0] += PHILOX_W32_0;
-        key[1] += PHILOX_W32_1;
+        key[0] += PHILOX_W0;
+        key[1] += PHILOX_W1;
       }
       round(key, _out);
     }
     _ctr++;
   }
 
+  inline DEVICE uint32_t mulhilo(uint32_t L, uint32_t R, uint32_t *hip) {
+    uint64_t product = static_cast<uint64_t>(L) * static_cast<uint64_t>(R);
+    *hip = product >> 32;
+    return static_cast<uint32_t>(product);
+  }
+
+  inline DEVICE void round(const uint32_t (&key)[2], uint32_t (&ctr)[4]) {
+    uint32_t hi0;
+    uint32_t hi1;
+    uint32_t lo0 = mulhilo(PHILOX_M0, ctr[0], &hi0);
+    uint32_t lo1 = mulhilo(PHILOX_M1, ctr[2], &hi1);
+    ctr[0] = hi1 ^ ctr[1] ^ key[0];
+    ctr[1] = lo1;
+    ctr[2] = hi0 ^ ctr[3] ^ key[1];
+    ctr[3] = lo0;
+  }
+
   // User provided seed and counter broken up, constant throughout
   // the lifetime of the object
   const uint32_t seed_hi, seed_lo;
diff --git a/tests/run_stat_tests.py b/tests/run_stat_tests.py
index 321cc82..8c2a0de 100644
--- a/tests/run_stat_tests.py
+++ b/tests/run_stat_tests.py
@@ -22,7 +22,7 @@
 PRACT_RND_EXEC = os.path.join(BUILD_DIR, "Practrand", "RNG_test")
 
 for gen in ["philox", "tyche", "threefry", "squares"]:
-    command  = f"{BUILD_DIR}/tests/pract_rand_multi {gen} | {PRACT_RND_EXEC} stdin32 -multithreaded -tlmax 4GB > {RES_DIR}/practrandm_{gen}.txt"
+    command  = f"{BUILD_DIR}/tests/pract_rand_multi {gen} | {PRACT_RND_EXEC} stdin32 -multithreaded -tlmax 8GB > {RES_DIR}/practrandm_{gen}.txt"
 
     p = subprocess.Popen(command, shell=True)
     p.name = f"practrand_{gen}"