CUDA 13.4 samples update - v13.4-public

Release 13.4 of the CUDA Samples supported by CUDA Toolkit 13.4.
See Changelog for more information.

python/1_GettingStarted/blurImageUnifiedMemory/README.md

@@ -112,12 +112,12 @@ When returning a zero-copy view, the caller must close the buffers after use (e.
 
 ### Platform Support:
 
-This sample relies on `ManagedMemoryResource` with **concurrent host access**
-to managed allocations while GPU kernels are in flight. That behavior
-requires the device property `concurrent_managed_access=True`, which is only
-supported on Linux with HMM (Pascal and newer). On Windows (WDDM/MCDM/TCC)
-the property is `False`, so the sample exits early with a waive message and
-exit code `2` instead of attempting a run that would crash the process.
+This sample builds host NumPy views over managed allocations, which requires
+the device property `concurrent_managed_access=True`. Devices that report
+`False` — Windows (WDDM/MCDM/TCC), and older integrated GPUs such as Jetson
+Orin — keep managed allocations GPU-exclusive, so the sample reports the
+missing capability and exits with code `2` instead of attempting a run that
+would crash the process.
 
 ## Installation
 
@@ -174,4 +174,4 @@ Verifying result...
 - [cuda.core Documentation](https://nvidia.github.io/cuda-python/cuda-core/latest/)
 - [cuda.core.Program](https://nvidia.github.io/cuda-python/cuda-core/latest/generated/cuda.core.Program.html)
 - [cuda.core.ManagedMemoryResource](https://nvidia.github.io/cuda-python/cuda-core/latest/generated/cuda.core.ManagedMemoryResource.html)
-- [CUDA Managed Memory](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#um-unified-memory-programming-hd)
+- [CUDA Managed Memory](https://docs.nvidia.com/cuda/cuda-programming-guide/04-special-topics/unified-memory.html#unified-memory)

python/1_GettingStarted/blurImageUnifiedMemory/blurImageUnifiedMemory.py

@@ -197,14 +197,6 @@ def main():
     3. Unified memory with cuda.core.ManagedMemoryResource
     4. Kernel launch with cuda.core.launch and LaunchConfig
     """
-    if sys.platform == "win32":
-        print(
-            "This sample relies on ManagedMemoryResource with concurrent host "
-            "access, which is not supported on Windows "
-            "(concurrent_managed_access=False). Waiving this sample."
-        )
-        sys.exit(2)
-
     print("=" * 60)
     print("Image Blur with Unified Memory (cuda.core)")
     print("=" * 60)
@@ -213,6 +205,13 @@ def main():
     device = Device(0)
     device.set_current()
 
+    # This sample builds host NumPy views over managed memory. Devices without
+    # concurrent managed access keep managed allocations GPU-exclusive while
+    # the GPU is active, so those host views fault.
+    if not device.properties.concurrent_managed_access:
+        print("Concurrent managed memory access is not supported on this platform.")
+        sys.exit(2)
+
     print(f"\nDevice: {device.name}")
     print(f"Compute Capability: sm_{device.arch}")
 

python/1_GettingStarted/copyImageArraytoGPU/copyImageArraytoGPU.py

@@ -189,6 +189,14 @@ def main():
     # Step 1: Set up CUDA device and stream
     dev = Device()  # Get default CUDA device (GPU 0)
     dev.set_current()  # Make this device the active one
+
+    # Pinned memory resources are backed by a host memory pool, which requires
+    # CUDA memory-pool support. This is not available on every platform (for
+    # example, Windows in TCC mode), so check before using it.
+    if not dev.properties.host_memory_pools_supported:
+        print("Host pinned memory pools are not supported on this platform.")
+        return 2
+
     stream = dev.create_stream()  # Create stream for async operations
 
     print(f"Device: {dev.name}")
@@ -233,7 +241,8 @@ def main():
     cp.cuda.Stream.null.use()  # Reset CuPy's stream to default
 
     print("\nDone")
+    return 0
 
 
 if __name__ == "__main__":
-    main()
+    sys.exit(main())

python/1_GettingStarted/deviceQuery/README.md

@@ -186,4 +186,4 @@ For multi-GPU systems, the output will include information for all detected devi
 
 - [CUDA Python Documentation](https://nvidia.github.io/cuda-python/)
 - [cuda.core API Guide](https://nvidia.github.io/cuda-python/cuda-core/latest/)
-- [CUDA Programming Guide - Device Information](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#device-enumeration)
+- [CUDA Programming Guide - Device Information](https://docs.nvidia.com/cuda/cuda-programming-guide/03-advanced/multi-gpu-systems.html#device-enumeration)

python/1_GettingStarted/deviceQuery/requirements.txt

@@ -1,4 +1,4 @@
 # Device Query Sample Requirements
 
 cuda-python>=13.0.0
-cuda-core>=1.0.0
+cuda-core>=1.2.0

python/1_GettingStarted/simplePrint/README.md

@@ -259,5 +259,5 @@ Try modifying:
 - [nvmath-python Device APIs](https://docs.nvidia.com/cuda/nvmath-python/latest/device-apis/index.html) - Optimized math operations for Numba CUDA kernels
 
 ### CUDA References:
-- [CUDA C Programming Guide - Printf](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#formatted-output)
+- [CUDA C Programming Guide - Printf](https://docs.nvidia.com/cuda/cuda-programming-guide/05-appendices/cpp-language-support.html#printf)
 - [C++ simplePrintf Sample](https://github.com/NVIDIA/cuda-samples/tree/master/Samples/0_Introduction/simplePrintf)

python/1_GettingStarted/simplePrint/requirements.txt

@@ -3,5 +3,9 @@
 cuda-python>=13.0.0
 cuda-core>=1.0.0
 # Numba JIT uses nvJitLink from pip; keep in step with cuda-bindings (e.g. 13.2.x).
+# Also pin libnvvm from pip so Numba doesn't fall back to a mismatched
+# system-toolkit libnvvm (nvJitLink ERROR_OUTDATED_LIBRARY).
+nvidia-nvvm>=13.2.0
 nvidia-nvjitlink>=13.2.0
 numba-cuda>=0.29.0
+numpy<2.5