diff --git a/cuda_pathfinder/cuda/pathfinder/__init__.py b/cuda_pathfinder/cuda/pathfinder/__init__.py
index d931a264cf..143c4b45cc 100644
--- a/cuda_pathfinder/cuda/pathfinder/__init__.py
+++ b/cuda_pathfinder/cuda/pathfinder/__init__.py
@@ -13,7 +13,7 @@
 from cuda.pathfinder._headers.supported_nvidia_headers import SUPPORTED_HEADERS_CTK as _SUPPORTED_HEADERS_CTK
 from cuda.pathfinder._version import __version__ as __version__
 
-# Indirection to help Sphinx find the docstring.
+# Indirections to help Sphinx find the docstrings.
 #: Mapping from short CUDA Toolkit (CTK) library names to their canonical
 #: header basenames (used to validate a discovered include directory).
 #: Example: ``"cublas" → "cublas.h"``. The key set is platform-aware
diff --git a/cuda_pathfinder/cuda/pathfinder/_dynamic_libs/supported_nvidia_libs.py b/cuda_pathfinder/cuda/pathfinder/_dynamic_libs/supported_nvidia_libs.py
index 6030dc5c85..c205b5eda4 100644
--- a/cuda_pathfinder/cuda/pathfinder/_dynamic_libs/supported_nvidia_libs.py
+++ b/cuda_pathfinder/cuda/pathfinder/_dynamic_libs/supported_nvidia_libs.py
@@ -82,8 +82,11 @@
 }
 DIRECT_DEPENDENCIES = DIRECT_DEPENDENCIES_CTK | {
     "mathdx": ("nvrtc",),
+    "cublasmp": ("cublas", "cublasLt", "nvshmem_host"),
     "cufftMp": ("nvshmem_host",),
     "cudss": ("cublas", "cublasLt"),
+    "cutensor": ("cublasLt",),
+    "cutensorMg": ("cutensor", "cublasLt"),
 }
 
 # Based on these released files:
@@ -238,9 +241,11 @@
 }
 SUPPORTED_LINUX_SONAMES_OTHER = {
     "cublasmp": ("libcublasmp.so.0",),
-    "cufftMp": ("libcufftMp.so.11",),
+    "cufftMp": ("libcufftMp.so.12", "libcufftMp.so.11"),
     "mathdx": ("libmathdx.so.0",),
     "cudss": ("libcudss.so.0",),
+    "cutensor": ("libcutensor.so.2",),
+    "cutensorMg": ("libcutensorMg.so.2",),
     "nccl": ("libnccl.so.2",),
     "nvpl_fftw": ("libnvpl_fftw.so.0",),
     "nvshmem_host": ("libnvshmem_host.so.3",),
@@ -402,6 +407,8 @@
 SUPPORTED_WINDOWS_DLLS_OTHER = {
     "mathdx": ("mathdx64_0.dll",),
     "cudss": ("cudss64_0.dll",),
+    "cutensor": ("cutensor.dll",),
+    "cutensorMg": ("cutensorMg.dll",),
 }
 SUPPORTED_WINDOWS_DLLS = SUPPORTED_WINDOWS_DLLS_CTK | SUPPORTED_WINDOWS_DLLS_OTHER
 
@@ -446,7 +453,9 @@
 SITE_PACKAGES_LIBDIRS_LINUX_OTHER = {
     "cublasmp": ("nvidia/cublasmp/cu13/lib", "nvidia/cublasmp/cu12/lib"),
     "cudss": ("nvidia/cu13/lib", "nvidia/cu12/lib"),
-    "cufftMp": ("nvidia/cufftmp/cu12/lib",),
+    "cufftMp": ("nvidia/cufftmp/cu13/lib", "nvidia/cufftmp/cu12/lib"),
+    "cutensor": ("cutensor/lib",),
+    "cutensorMg": ("cutensor/lib",),
     "mathdx": ("nvidia/cu13/lib", "nvidia/cu12/lib"),
     "nccl": ("nvidia/nccl/lib",),
     "nvpl_fftw": ("nvpl/lib",),
@@ -484,7 +493,10 @@
     "nvvm": ("nvidia/cu13/bin/x86_64", "nvidia/cuda_nvcc/nvvm/bin"),
 }
 SITE_PACKAGES_LIBDIRS_WINDOWS_OTHER = {
+    "cudss": ("nvidia/cu13/bin", "nvidia/cu12/bin"),
     "mathdx": ("nvidia/cu13/bin/x86_64", "nvidia/cu12/bin"),
+    "cutensor": ("cutensor/bin",),
+    "cutensorMg": ("cutensor/bin",),
 }
 SITE_PACKAGES_LIBDIRS_WINDOWS = SITE_PACKAGES_LIBDIRS_WINDOWS_CTK | SITE_PACKAGES_LIBDIRS_WINDOWS_OTHER
 
diff --git a/cuda_pathfinder/cuda/pathfinder/_headers/find_nvidia_headers.py b/cuda_pathfinder/cuda/pathfinder/_headers/find_nvidia_headers.py
index 535d4b8003..b141700ab7 100644
--- a/cuda_pathfinder/cuda/pathfinder/_headers/find_nvidia_headers.py
+++ b/cuda_pathfinder/cuda/pathfinder/_headers/find_nvidia_headers.py
@@ -22,28 +22,12 @@ def _joined_isfile(dirpath: str, basename: str) -> bool:
     return os.path.isfile(os.path.join(dirpath, basename))
 
 
-def _find_nvshmem_header_directory() -> Optional[str]:
-    if IS_WINDOWS:
-        # nvshmem has no Windows support.
-        return None
-
+def _find_under_site_packages(sub_dir: str, h_basename: str) -> Optional[str]:
     # Installed from a wheel
-    nvidia_sub_dirs = ("nvidia", "nvshmem", "include")
     hdr_dir: str  # help mypy
-    for hdr_dir in find_sub_dirs_all_sitepackages(nvidia_sub_dirs):
-        if _joined_isfile(hdr_dir, "nvshmem.h"):
-            return hdr_dir
-
-    conda_prefix = os.environ.get("CONDA_PREFIX")
-    if conda_prefix and os.path.isdir(conda_prefix):
-        hdr_dir = os.path.join(conda_prefix, "include")
-        if _joined_isfile(hdr_dir, "nvshmem.h"):
-            return hdr_dir
-
-    for hdr_dir in sorted(glob.glob("/usr/include/nvshmem_*"), reverse=True):
-        if _joined_isfile(hdr_dir, "nvshmem.h"):
+    for hdr_dir in find_sub_dirs_all_sitepackages(tuple(sub_dir.split("/"))):
+        if _joined_isfile(hdr_dir, h_basename):
             return hdr_dir
-
     return None
 
 
@@ -54,6 +38,13 @@ def _find_based_on_ctk_layout(libname: str, h_basename: str, anchor_point: str)
     parts.append("include")
     idir = os.path.join(*parts)
     if libname == "cccl":
+        if IS_WINDOWS:
+            cdir_ctk12 = os.path.join(idir, "targets", "x64")  # conda has this anomaly
+            cdir_ctk13 = os.path.join(cdir_ctk12, "cccl")
+            if _joined_isfile(cdir_ctk13, h_basename):
+                return cdir_ctk13
+            if _joined_isfile(cdir_ctk12, h_basename):
+                return cdir_ctk12
         cdir = os.path.join(idir, "cccl")  # CTK 13
         if _joined_isfile(cdir, h_basename):
             return cdir
@@ -62,20 +53,27 @@ def _find_based_on_ctk_layout(libname: str, h_basename: str, anchor_point: str)
     return None
 
 
-def _find_based_on_conda_layout(libname: str, h_basename: str, conda_prefix: str) -> Optional[str]:
+def _find_based_on_conda_layout(libname: str, h_basename: str, ctk_layout: bool) -> Optional[str]:
+    conda_prefix = os.environ.get("CONDA_PREFIX")
+    if not conda_prefix:
+        return None
     if IS_WINDOWS:
         anchor_point = os.path.join(conda_prefix, "Library")
         if not os.path.isdir(anchor_point):
             return None
     else:
-        targets_include_path = glob.glob(os.path.join(conda_prefix, "targets", "*", "include"))
-        if not targets_include_path:
-            return None
-        if len(targets_include_path) != 1:
-            # Conda does not support multiple architectures.
-            # QUESTION(PR#956): Do we want to issue a warning?
-            return None
-        anchor_point = os.path.dirname(targets_include_path[0])
+        if ctk_layout:
+            targets_include_path = glob.glob(os.path.join(conda_prefix, "targets", "*", "include"))
+            if not targets_include_path:
+                return None
+            if len(targets_include_path) != 1:
+                # Conda does not support multiple architectures.
+                # QUESTION(PR#956): Do we want to issue a warning?
+                return None
+            include_path = targets_include_path[0]
+        else:
+            include_path = os.path.join(conda_prefix, "include")
+        anchor_point = os.path.dirname(include_path)
     return _find_based_on_ctk_layout(libname, h_basename, anchor_point)
 
 
@@ -83,17 +81,12 @@ def _find_ctk_header_directory(libname: str) -> Optional[str]:
     h_basename = supported_nvidia_headers.SUPPORTED_HEADERS_CTK[libname]
     candidate_dirs = supported_nvidia_headers.SUPPORTED_SITE_PACKAGE_HEADER_DIRS_CTK[libname]
 
-    # Installed from a wheel
     for cdir in candidate_dirs:
-        hdr_dir: str  # help mypy
-        for hdr_dir in find_sub_dirs_all_sitepackages(tuple(cdir.split("/"))):
-            if _joined_isfile(hdr_dir, h_basename):
-                return hdr_dir
+        if hdr_dir := _find_under_site_packages(cdir, h_basename):
+            return hdr_dir
 
-    conda_prefix = os.environ.get("CONDA_PREFIX")
-    if conda_prefix:  # noqa: SIM102
-        if result := _find_based_on_conda_layout(libname, h_basename, conda_prefix):
-            return result
+    if hdr_dir := _find_based_on_conda_layout(libname, h_basename, True):
+        return hdr_dir
 
     cuda_home = get_cuda_home_or_path()
     if cuda_home:  # noqa: SIM102
@@ -132,19 +125,28 @@ def find_nvidia_header_directory(libname: str) -> Optional[str]:
         3. **CUDA Toolkit environment variables**
 
            - Use ``CUDA_HOME`` or ``CUDA_PATH`` (in that order).
-
-    Notes:
-        - The ``SUPPORTED_HEADERS_CTK`` dictionary maps each supported CUDA Toolkit
-          (CTK) library to the name of its canonical header (e.g., ``"cublas" →
-          "cublas.h"``). This is used to verify that the located directory is valid.
-
-        - The only supported non-CTK library at present is ``nvshmem``.
     """
 
-    if libname == "nvshmem":
-        return _abs_norm(_find_nvshmem_header_directory())
-
     if libname in supported_nvidia_headers.SUPPORTED_HEADERS_CTK:
         return _abs_norm(_find_ctk_header_directory(libname))
 
-    raise RuntimeError(f"UNKNOWN {libname=}")
+    h_basename = supported_nvidia_headers.SUPPORTED_HEADERS_NON_CTK.get(libname)
+    if h_basename is None:
+        raise RuntimeError(f"UNKNOWN {libname=}")
+
+    candidate_dirs = supported_nvidia_headers.SUPPORTED_SITE_PACKAGE_HEADER_DIRS_NON_CTK.get(libname, [])
+    hdr_dir: Optional[str]  # help mypy
+    for cdir in candidate_dirs:
+        if hdr_dir := _find_under_site_packages(cdir, h_basename):
+            return _abs_norm(hdr_dir)
+
+    if hdr_dir := _find_based_on_conda_layout(libname, h_basename, False):
+        return _abs_norm(hdr_dir)
+
+    candidate_dirs = supported_nvidia_headers.SUPPORTED_INSTALL_DIRS_NON_CTK.get(libname, [])
+    for cdir in candidate_dirs:
+        for hdr_dir in sorted(glob.glob(cdir), reverse=True):
+            if _joined_isfile(hdr_dir, h_basename):
+                return _abs_norm(hdr_dir)
+
+    return None
diff --git a/cuda_pathfinder/cuda/pathfinder/_headers/supported_nvidia_headers.py b/cuda_pathfinder/cuda/pathfinder/_headers/supported_nvidia_headers.py
index 1b2855c0c9..7fdf4159ff 100644
--- a/cuda_pathfinder/cuda/pathfinder/_headers/supported_nvidia_headers.py
+++ b/cuda_pathfinder/cuda/pathfinder/_headers/supported_nvidia_headers.py
@@ -57,3 +57,28 @@
     "nvrtc": ("nvidia/cu13/include", "nvidia/cuda_nvrtc/include"),
     "nvvm": ("nvidia/cu13/include", "nvidia/cuda_nvcc/nvvm/include"),
 }
+
+SUPPORTED_HEADERS_NON_CTK_COMMON = {
+    "cutensor": "cutensor.h",
+}
+SUPPORTED_HEADERS_NON_CTK_LINUX_ONLY = {
+    "nvshmem": "nvshmem.h",
+}
+SUPPORTED_HEADERS_NON_CTK_WINDOWS_ONLY: Final[dict[str, str]] = {}
+SUPPORTED_HEADERS_NON_CTK_LINUX = SUPPORTED_HEADERS_NON_CTK_COMMON | SUPPORTED_HEADERS_NON_CTK_LINUX_ONLY
+SUPPORTED_HEADERS_NON_CTK_WINDOWS = SUPPORTED_HEADERS_NON_CTK_COMMON | SUPPORTED_HEADERS_NON_CTK_WINDOWS_ONLY
+SUPPORTED_HEADERS_NON_CTK_ALL = (
+    SUPPORTED_HEADERS_NON_CTK_COMMON | SUPPORTED_HEADERS_NON_CTK_LINUX_ONLY | SUPPORTED_HEADERS_NON_CTK_WINDOWS_ONLY
+)
+SUPPORTED_HEADERS_NON_CTK: Final[dict[str, str]] = (
+    SUPPORTED_HEADERS_NON_CTK_WINDOWS if IS_WINDOWS else SUPPORTED_HEADERS_NON_CTK_LINUX
+)
+
+SUPPORTED_SITE_PACKAGE_HEADER_DIRS_NON_CTK = {
+    "cutensor": ("cutensor/include",),
+    "nvshmem": ("nvidia/nvshmem/include",),
+}
+
+SUPPORTED_INSTALL_DIRS_NON_CTK = {
+    "nvshmem": ("/usr/include/nvshmem_*",),
+}
diff --git a/cuda_pathfinder/cuda/pathfinder/_version.py b/cuda_pathfinder/cuda/pathfinder/_version.py
index cb42a4a3c5..b5fa5ff85c 100644
--- a/cuda_pathfinder/cuda/pathfinder/_version.py
+++ b/cuda_pathfinder/cuda/pathfinder/_version.py
@@ -1,4 +1,4 @@
 # SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
-__version__ = "1.3.1"
+__version__ = "1.3.2"
diff --git a/cuda_pathfinder/docs/nv-versions.json b/cuda_pathfinder/docs/nv-versions.json
index f25c50bc6d..2bec723d4b 100644
--- a/cuda_pathfinder/docs/nv-versions.json
+++ b/cuda_pathfinder/docs/nv-versions.json
@@ -3,6 +3,10 @@
         "version": "latest",
         "url": "https://nvidia.github.io/cuda-python/cuda-pathfinder/latest/"
     },
+    {
+        "version": "1.3.2",
+        "url": "https://nvidia.github.io/cuda-python/cuda-pathfinder/1.3.2/"
+    },
     {
         "version": "1.3.1",
         "url": "https://nvidia.github.io/cuda-python/cuda-pathfinder/1.3.1/"
diff --git a/cuda_pathfinder/docs/source/api.rst b/cuda_pathfinder/docs/source/api.rst
index 3cae4b6f70..72e5e40724 100644
--- a/cuda_pathfinder/docs/source/api.rst
+++ b/cuda_pathfinder/docs/source/api.rst
@@ -18,4 +18,5 @@ and experimental APIs for locating NVIDIA C/C++ header directories.
    DynamicLibNotFoundError
 
    SUPPORTED_HEADERS_CTK
+   SUPPORTED_HEADERS_NON_CTK
    find_nvidia_header_directory
diff --git a/cuda_pathfinder/docs/source/release/1.3.2-notes.rst b/cuda_pathfinder/docs/source/release/1.3.2-notes.rst
new file mode 100644
index 0000000000..a0b3efa247
--- /dev/null
+++ b/cuda_pathfinder/docs/source/release/1.3.2-notes.rst
@@ -0,0 +1,15 @@
+.. SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+.. SPDX-License-Identifier: Apache-2.0
+
+.. py:currentmodule:: cuda.pathfinder
+
+``cuda-pathfinder`` 1.3.2 Release notes
+=======================================
+
+Released on Oct 29, 2025
+
+Highlights
+----------
+
+* Add cuTENSOR support & bug fixes discovered while working on conda testing
+  (`PR #1194 <https://github.com/NVIDIA/cuda-python/pull/1194>`_)
diff --git a/cuda_pathfinder/pyproject.toml b/cuda_pathfinder/pyproject.toml
index 80aeb1a4b6..3db1aecbc8 100644
--- a/cuda_pathfinder/pyproject.toml
+++ b/cuda_pathfinder/pyproject.toml
@@ -18,6 +18,7 @@ test = [
 cu12 = [
     "cuda-toolkit[nvcc,cublas,nvrtc,cudart,cufft,curand,cusolver,cusparse,npp,nvfatbin,nvjitlink,nvjpeg,cccl]==12.*",
     "cuda-toolkit[cufile]==12.*; sys_platform != 'win32'",
+    "cutensor-cu12",
     "nvidia-cublasmp-cu12; sys_platform != 'win32'",
     "nvidia-cudss-cu12",
     "nvidia-cufftmp-cu12; sys_platform != 'win32'",
@@ -28,6 +29,7 @@ cu12 = [
 cu13 = [
     "cuda-toolkit[nvcc,cublas,nvrtc,cudart,cufft,curand,cusolver,cusparse,npp,nvfatbin,nvjitlink,nvjpeg,cccl,nvvm]==13.*",
     "cuda-toolkit[cufile]==13.*; sys_platform != 'win32'",
+    "cutensor-cu13",
     "nvidia-cublasmp-cu13; sys_platform != 'win32'",
     "nvidia-cudss-cu13",
     "nvidia-nccl-cu13; sys_platform != 'win32'",
diff --git a/cuda_pathfinder/tests/local_helpers.py b/cuda_pathfinder/tests/local_helpers.py
new file mode 100644
index 0000000000..7893ba8229
--- /dev/null
+++ b/cuda_pathfinder/tests/local_helpers.py
@@ -0,0 +1,16 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+import functools
+import importlib.metadata
+import re
+
+
+@functools.cache
+def have_distribution(name_pattern: str) -> bool:
+    re_name_pattern = re.compile(name_pattern)
+    return any(
+        re_name_pattern.match(dist.metadata["Name"])
+        for dist in importlib.metadata.distributions()
+        if "Name" in dist.metadata
+    )
diff --git a/cuda_pathfinder/tests/test_find_nvidia_headers.py b/cuda_pathfinder/tests/test_find_nvidia_headers.py
index cdea0cd28f..bbc920d765 100644
--- a/cuda_pathfinder/tests/test_find_nvidia_headers.py
+++ b/cuda_pathfinder/tests/test_find_nvidia_headers.py
@@ -3,9 +3,7 @@
 
 # Currently these installations are only manually tested:
 
-# conda create -y -n nvshmem python=3.12
-# conda activate nvshmem
-# conda install -y conda-forge::libnvshmem3 conda-forge::libnvshmem-dev
+# ../toolshed/conda_create_for_pathfinder_testing.*
 
 # wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64/cuda-keyring_1.1-1_all.deb
 # sudo dpkg -i cuda-keyring_1.1-1_all.deb
@@ -14,6 +12,7 @@
 # sudo apt install libnvshmem3-cuda-13 libnvshmem3-dev-cuda-13
 
 import functools
+import glob
 import importlib.metadata
 import os
 import re
@@ -24,20 +23,19 @@
 from cuda.pathfinder._headers.supported_nvidia_headers import (
     SUPPORTED_HEADERS_CTK,
     SUPPORTED_HEADERS_CTK_ALL,
+    SUPPORTED_HEADERS_NON_CTK,
+    SUPPORTED_HEADERS_NON_CTK_ALL,
+    SUPPORTED_INSTALL_DIRS_NON_CTK,
     SUPPORTED_SITE_PACKAGE_HEADER_DIRS_CTK,
 )
-from cuda.pathfinder._utils.platform_aware import IS_WINDOWS
 
 STRICTNESS = os.environ.get("CUDA_PATHFINDER_TEST_FIND_NVIDIA_HEADERS_STRICTNESS", "see_what_works")
 assert STRICTNESS in ("see_what_works", "all_must_work")
 
-
-@functools.cache
-def have_nvidia_nvshmem_package() -> bool:
-    pattern = re.compile(r"^nvidia-nvshmem-.*$")
-    return any(
-        pattern.match(dist.metadata["Name"]) for dist in importlib.metadata.distributions() if "Name" in dist.metadata
-    )
+NON_CTK_IMPORTLIB_METADATA_DISTRIBUTIONS_NAMES = {
+    "cutensor": r"^cutensor-.*$",
+    "nvshmem": r"^nvidia-nvshmem-.*$",
+}
 
 
 def test_unknown_libname():
@@ -45,24 +43,43 @@ def test_unknown_libname():
         find_nvidia_header_directory("unknown-libname")
 
 
-def test_find_libname_nvshmem(info_summary_append):
-    hdr_dir = find_nvidia_header_directory("nvshmem")
+def test_non_ctk_importlib_metadata_distributions_names():
+    # Ensure the dict keys above stay in sync with supported_nvidia_headers
+    assert sorted(NON_CTK_IMPORTLIB_METADATA_DISTRIBUTIONS_NAMES) == sorted(SUPPORTED_HEADERS_NON_CTK_ALL)
+
+
+@functools.cache
+def have_distribution_for(libname: str) -> bool:
+    pattern = re.compile(NON_CTK_IMPORTLIB_METADATA_DISTRIBUTIONS_NAMES[libname])
+    return any(
+        pattern.match(dist.metadata["Name"]) for dist in importlib.metadata.distributions() if "Name" in dist.metadata
+    )
+
+
+@pytest.mark.parametrize("libname", SUPPORTED_HEADERS_NON_CTK.keys())
+def test_find_non_ctk_headers(info_summary_append, libname):
+    hdr_dir = find_nvidia_header_directory(libname)
     info_summary_append(f"{hdr_dir=!r}")
-    if IS_WINDOWS:
-        assert hdr_dir is None
-        pytest.skip("nvshmem has no Windows support.")
     if hdr_dir:
         assert os.path.isdir(hdr_dir)
-        assert os.path.isfile(os.path.join(hdr_dir, "nvshmem.h"))
-    if STRICTNESS == "all_must_work" or have_nvidia_nvshmem_package():
+        assert os.path.isfile(os.path.join(hdr_dir, SUPPORTED_HEADERS_NON_CTK[libname]))
+    if have_distribution_for(libname):
+        assert hdr_dir is not None
+        hdr_dir_parts = hdr_dir.split(os.path.sep)
+        assert "site-packages" in hdr_dir_parts
+    elif STRICTNESS == "all_must_work":
         assert hdr_dir is not None
-        if have_nvidia_nvshmem_package():
-            hdr_dir_parts = hdr_dir.split(os.path.sep)
-            assert "site-packages" in hdr_dir_parts
-        elif conda_prefix := os.environ.get("CONDA_PREFIX"):
+        if conda_prefix := os.environ.get("CONDA_PREFIX"):
             assert hdr_dir.startswith(conda_prefix)
         else:
-            assert hdr_dir.startswith("/usr/include/nvshmem_")
+            inst_dirs = SUPPORTED_INSTALL_DIRS_NON_CTK.get(libname)
+            if inst_dirs is not None:
+                for inst_dir in inst_dirs:
+                    globbed = glob.glob(inst_dir)
+                    if hdr_dir in globbed:
+                        break
+                else:
+                    raise RuntimeError(f"{hdr_dir=} does not match any {inst_dirs=}")
 
 
 def test_supported_headers_site_packages_ctk_consistency():
diff --git a/cuda_pathfinder/tests/test_load_nvidia_dynamic_lib.py b/cuda_pathfinder/tests/test_load_nvidia_dynamic_lib.py
index b50c644a61..cff5b74290 100644
--- a/cuda_pathfinder/tests/test_load_nvidia_dynamic_lib.py
+++ b/cuda_pathfinder/tests/test_load_nvidia_dynamic_lib.py
@@ -1,19 +1,18 @@
 # SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
-import functools
 import json
 import os
+import platform
 from unittest.mock import patch
 
 import pytest
 import spawned_process_runner
 from child_load_nvidia_dynamic_lib_helper import build_child_process_failed_for_libname_message, child_process_func
+from local_helpers import have_distribution
 
-from cuda.pathfinder import SUPPORTED_NVIDIA_LIBNAMES, load_nvidia_dynamic_lib
+from cuda.pathfinder import load_nvidia_dynamic_lib
 from cuda.pathfinder._dynamic_libs import supported_nvidia_libs
-from cuda.pathfinder._utils.find_site_packages_dll import find_all_dll_files_via_metadata
-from cuda.pathfinder._utils.find_site_packages_so import find_all_so_files_via_metadata
 from cuda.pathfinder._utils.platform_aware import IS_WINDOWS, quote_for_shell
 
 STRICTNESS = os.environ.get("CUDA_PATHFINDER_TEST_LOAD_NVIDIA_DYNAMIC_LIB_STRICTNESS", "see_what_works")
@@ -71,28 +70,25 @@ def test_runtime_error_on_non_64bit_python():
         load_nvidia_dynamic_lib("not_used")
 
 
-@functools.cache
-def _get_libnames_for_test_load_nvidia_dynamic_lib():
-    result = list(SUPPORTED_NVIDIA_LIBNAMES)
-    if IS_WINDOWS:
-        spld_other = supported_nvidia_libs.SITE_PACKAGES_LIBDIRS_WINDOWS_OTHER
-        all_dyn_libs = find_all_dll_files_via_metadata()
-        for libname in spld_other:
-            for dll_name in all_dyn_libs:
-                if dll_name.startswith(libname):
-                    result.append(libname)
-    else:
-        spld_other = supported_nvidia_libs.SITE_PACKAGES_LIBDIRS_LINUX_OTHER
-        all_dyn_libs = find_all_so_files_via_metadata()
-        for libname in spld_other:
-            so_basename = f"lib{libname}.so"
-            if so_basename in all_dyn_libs:
-                result.append(libname)
+IMPORTLIB_METADATA_DISTRIBUTIONS_NAMES = {
+    "cufftMp": r"^nvidia-cufftmp-.*$",
+    "mathdx": r"^nvidia-libmathdx-.*$",
+}
+
 
-    return tuple(result)
+def _is_expected_load_nvidia_dynamic_lib_failure(libname):
+    if libname == "nvpl_fftw" and platform.machine().lower() != "aarch64":
+        return True
+    dist_name_pattern = IMPORTLIB_METADATA_DISTRIBUTIONS_NAMES.get(libname)
+    if dist_name_pattern is not None:
+        return not have_distribution(dist_name_pattern)
+    return False
 
 
-@pytest.mark.parametrize("libname", _get_libnames_for_test_load_nvidia_dynamic_lib())
+@pytest.mark.parametrize(
+    "libname",
+    supported_nvidia_libs.SUPPORTED_WINDOWS_DLLS if IS_WINDOWS else supported_nvidia_libs.SUPPORTED_LINUX_SONAMES,
+)
 def test_load_nvidia_dynamic_lib(info_summary_append, libname):
     # We intentionally run each dynamic library operation in a child process
     # to ensure isolation of global dynamic linking state (e.g., dlopen handles).
@@ -108,7 +104,7 @@ def raise_child_process_failed():
         raise_child_process_failed()
     assert not result.stderr
     if result.stdout.startswith("CHILD_LOAD_NVIDIA_DYNAMIC_LIB_HELPER_DYNAMIC_LIB_NOT_FOUND_ERROR:"):
-        if STRICTNESS == "all_must_work":
+        if STRICTNESS == "all_must_work" and not _is_expected_load_nvidia_dynamic_lib_failure(libname):
             raise_child_process_failed()
         info_summary_append(f"Not found: {libname=!r}")
     else:
diff --git a/toolshed/conda_create_for_pathfinder_testing.ps1 b/toolshed/conda_create_for_pathfinder_testing.ps1
new file mode 100644
index 0000000000..a2c6c31ef0
--- /dev/null
+++ b/toolshed/conda_create_for_pathfinder_testing.ps1
@@ -0,0 +1,30 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION
+# SPDX-License-Identifier: Apache-2.0
+
+param(
+    [Parameter(Mandatory = $true)]
+    [string]$CudaVersion
+)
+
+$ErrorActionPreference = "Stop"
+
+& "$env:CONDA_EXE" "shell.powershell" "hook" | Out-String | Invoke-Expression
+
+conda create --yes -n "pathfinder_testing_cu$CudaVersion" python=3.13 "cuda-toolkit=$CudaVersion"
+conda activate "pathfinder_testing_cu$CudaVersion"
+
+$cpkgs = @(
+    "cutensor",
+    "libcublasmp-dev",
+    "libcudss-dev",
+    "libcufftmp-dev",
+    "libmathdx-dev",
+    "libnvshmem3",
+    "libnvshmem-dev",
+    "libnvpl-fft-dev"
+)
+
+foreach ($cpkg in $cpkgs) {
+    Write-Host "CONDA INSTALL: $cpkg"
+    conda install -y -c conda-forge $cpkg
+}
diff --git a/toolshed/conda_create_for_pathfinder_testing.sh b/toolshed/conda_create_for_pathfinder_testing.sh
new file mode 100755
index 0000000000..2433787581
--- /dev/null
+++ b/toolshed/conda_create_for_pathfinder_testing.sh
@@ -0,0 +1,27 @@
+#!/bin/bash
+
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+if [[ $# -ne 1 ]]; then
+    echo "Usage: $(basename "$0") ctk-major-minor-patch" 1>&2
+    exit 1
+fi
+
+eval "$(conda shell.bash hook)"
+
+conda create --yes -n "pathfinder_testing_cu$1" python=3.13 cuda-toolkit="$1"
+conda activate "pathfinder_testing_cu$1"
+
+for cpkg in \
+    cutensor \
+    libcublasmp-dev \
+    libcudss-dev \
+    libcufftmp-dev \
+    libmathdx-dev \
+    libnvshmem3 \
+    libnvshmem-dev \
+    libnvpl-fft-dev; do
+    echo "CONDA INSTALL: $cpkg"
+    conda install -y -c conda-forge "$cpkg"
+done