@@ -88,7 +88,8 @@ def print_driver_info() -> None:
             print(f"Driver branch: unavailable ({e})")
     else:
         print("NVML bindings are not compatible with this driver; device info will be limited.")
-    print(f"Current process: {system.get_process_name(os.getpid())}")
+    if CUDA_BINDINGS_NVML_IS_COMPATIBLE:
+        print(f"Current process: {system.get_process_name(os.getpid())}")
 
 
 def print_device_info(device: "system.Device") -> None:
 
@@ -49,6 +49,8 @@
 import sys
 from pathlib import Path
 
+import numpy as np
+
 # Add parent directory to path to import utilities
 sys.path.insert(0, str(Path(__file__).parent.parent / "Utilities"))
 from cuda_samples_utils import verify_array_result
@@ -96,8 +98,8 @@ def _demo_float_cupy(stream, kernel, num_elements, verify):
     print(f"\n[1] vectorAdd<float> on {num_elements} CuPy-allocated elements")
     dtype = cp.float32
 
-    a = cp.random.rand(num_elements).astype(dtype)
-    b = cp.random.rand(num_elements).astype(dtype)
+    a = cp.asarray(np.random.rand(num_elements).astype(np.float32))
+    b = cp.asarray(np.random.rand(num_elements).astype(np.float32))
     c = cp.empty(num_elements, dtype=dtype)
     _launch_add(stream, kernel, num_elements, a.data.ptr, b.data.ptr, c.data.ptr)
 
@@ -118,8 +120,8 @@ def _demo_double_owned_buffer(device, stream, kernel, num_elements, verify):
     print(f"\n[2] vectorAdd<double> on {num_elements} elements with device.allocate() output")
     dtype = cp.float64
 
-    a = cp.random.rand(num_elements).astype(dtype)
-    b = cp.random.rand(num_elements).astype(dtype)
+    a = cp.asarray(np.random.rand(num_elements).astype(np.float64))
+    b = cp.asarray(np.random.rand(num_elements).astype(np.float64))
 
     out_bytes = num_elements * dtype().itemsize
     out_buf = device.allocate(out_bytes, stream=stream)