Release v13.3 of the CUDA samples with CUDA 13.3 Toolkit (#435)

This is the release of the CUDA 13.3 samples, which include additions for CUDA Tile C++, and updated CCCL and Python samples.

python/2_CoreConcepts/pageRank/README.md

@@ -1,5 +1,23 @@
 # Sample: PageRank Algorithm (Python)
 
+> **Known issue — version-pinned sample.** Unlike the other samples in this
+> repository, this sample is pinned to `cuda-core<1.0.0`. The reason is that
+> `cudf-cu13` transitively requires `numba-cuda<0.29.0`, and every
+> `numba-cuda` release in that range pins `cuda-core<1.0.0`. Installing this
+> sample's `requirements.txt` into a shared environment will downgrade
+> `cuda-core` and break the other samples (which use the 1.0 API).
+>
+> The recommended workflow is one of:
+>
+> - Install this sample's requirements in a **dedicated virtual
+>   environment**, or
+> - Re-run the other samples' `pip install -r requirements.txt` afterwards
+>   to upgrade `cuda-core` back to 1.0.
+>
+> This sample will be re-aligned with the rest of the repository
+> (`cuda-core>=1.0.0`) once `cudf-cu13` ships a release that lifts its
+> `numba-cuda` upper bound.
+
 ## Description
 
 Demonstrates GPU-accelerated PageRank computation for graph analysis using RAPIDS cuGraph, with cuda.core for device, stream, and GPU timing. This sample focuses on cuda.core integration with high-level libraries (cuGraph/cuDF); for custom kernel programming (Program, LaunchConfig, launch), see the blockwiseSum sample.
@@ -16,7 +34,7 @@ Demonstrates GPU-accelerated PageRank computation for graph analysis using RAPID
 - `cugraph` - RAPIDS GPU-accelerated graph analytics
 - `cudf` - RAPIDS GPU DataFrame library
 - `cuda.core` - Device, stream, and event APIs for GPU timing
-- `cupy` - GPU array library (ExternalStream for cuDF/cuGraph)
+- `cupy` - GPU array library (Stream.from_external for cuDF/cuGraph)
 - `numpy` - CPU reference implementation
 
 ## Key APIs
@@ -25,7 +43,7 @@ Demonstrates GPU-accelerated PageRank computation for graph analysis using RAPID
 
 - `Device(0)` - Create device, `device.set_current()`, `device.create_stream()`
 - `EventOptions(enable_timing=True)` - GPU timing via `stream.record()`
-- `cp.cuda.ExternalStream(stream.handle).use()` - Make cuDF/cuGraph use cuda.core stream
+- `cp.cuda.Stream.from_external(stream).use()` - Make cuDF/cuGraph use cuda.core stream
 
 ### From cuGraph:
 
@@ -50,6 +68,14 @@ Demonstrates GPU-accelerated PageRank computation for graph analysis using RAPID
 - Python 3.10 or newer
 - See requirements.txt for package dependencies
 
+### Platform Support:
+
+This sample depends on RAPIDS (`cugraph-cu13`, `cudf-cu13`, `dask-cuda`),
+which is currently published only as **Linux (manylinux) wheels** on
+`pypi.nvidia.com` — no Windows wheels exist. On Windows the sample exits
+early with a waive message and exit code `2` instead of attempting an
+install that cannot succeed.
+
 ## Installation
 
 ```bash

python/2_CoreConcepts/pageRank/pageRank.py

@@ -46,6 +46,14 @@ from pathlib import Path
 sys.path.insert(0, str(Path(__file__).parent.parent.parent / "Utilities"))
 from cuda_samples_utils import print_gpu_info, verify_array_result  # noqa: E402
 
+if sys.platform == "win32":
+    print(
+        "This sample depends on RAPIDS (cugraph-cu13 / cudf-cu13), which is "
+        "currently published only as Linux (manylinux) wheels on "
+        "pypi.nvidia.com. Waiving this sample on Windows."
+    )
+    sys.exit(2)
+
 try:
     import cudf
     import cugraph
@@ -210,8 +218,22 @@ def run_pagerank_benchmark(
     print()
     print_gpu_info(device)
 
+    # RAPIDS cuGraph wheels currently don't ship kernel binaries for
+    # every CUDA architecture. Skip cleanly on architectures known to
+    # be unsupported instead of failing deep inside cuGraph with a
+    # cryptic cudaErrorNoKernelImageForDevice. Remove an arch from this
+    # set once the matching cuGraph release ships kernels for it.
+    _CUGRAPH_UNSUPPORTED_ARCHES = {"110"}  # sm_110 = Thor / Tegra
+    if device.arch in _CUGRAPH_UNSUPPORTED_ARCHES:
+        print(
+            f"RAPIDS cuGraph does not yet ship kernels for sm_{device.arch}, "
+            "waiving this sample."
+        )
+        stream.close()
+        sys.exit(2)
+
     # Make CuPy/cuDF use our cuda.core stream
-    cp.cuda.ExternalStream(int(stream.handle)).use()
+    cp.cuda.Stream.from_external(stream).use()
 
     # Generate random graph
     print("\nGraph Parameters:")

python/2_CoreConcepts/pageRank/requirements.txt

@@ -1,8 +1,11 @@
 # PageRank Requirements (RAPIDS cuGraph)
 
 cuda-python>=13.0.0
-cuda-core>=0.6.0
-cugraph-cu13>=25.0.0
-cudf-cu13>=25.0.0
-cupy-cuda13x>=13.0.0
+# cudf-cu13 transitively pins numba-cuda<0.29.0 which requires cuda-core<1.0.0
+cuda-core<1.0.0
+cugraph-cu13>=26.0.0
+cudf-cu13>=26.0.0
+# dask-cuda <26.4 incorrectly pins cuda-core==0.3.*; require the fixed release
+dask-cuda>=26.4.0
+cupy-cuda13x>=14.0.0
 numpy>=2.3.2