diff --git a/pkgs/development/libraries/science/math/tiny-cuda-nn/default.nix b/pkgs/development/libraries/science/math/tiny-cuda-nn/default.nix
new file mode 100644
index 0000000000000..ef501dfd09a04
--- /dev/null
+++ b/pkgs/development/libraries/science/math/tiny-cuda-nn/default.nix
@@ -0,0 +1,152 @@
+{ cmake
+, cudaPackages
+, fetchFromGitHub
+, lib
+, ninja
+, pkgs
+, python3Packages ? { }
+, pythonSupport ? false
+, stdenv
+, symlinkJoin
+, which
+}:
+let
+  inherit (lib) lists strings;
+  inherit (cudaPackages) backendStdenv cudaFlags;
+
+  cuda-common-redist = with cudaPackages; [
+    libcublas # cublas_v2.h
+    libcusolver # cusolverDn.h
+    libcusparse # cusparse.h
+  ];
+
+  cuda-native-redist = symlinkJoin {
+    name = "cuda-redist";
+    paths = with cudaPackages; [
+      cuda_cudart # cuda_runtime.h
+      cuda_nvcc
+    ] ++ cuda-common-redist;
+  };
+
+  cuda-redist = symlinkJoin {
+    name = "cuda-redist";
+    paths = cuda-common-redist;
+  };
+in
+stdenv.mkDerivation (finalAttrs: {
+  name = "tiny-cuda-nn";
+  version = "1.6";
+
+  format = strings.optionalString pythonSupport "setuptools";
+
+  src = fetchFromGitHub {
+    owner = "NVlabs";
+    repo = finalAttrs.name;
+    rev = "v${finalAttrs.version}";
+    fetchSubmodules = true;
+    hash = "sha256-qW6Fk2GB71fvZSsfu+mykabSxEKvaikZ/pQQZUycOy0=";
+  };
+
+  nativeBuildInputs = [
+    cmake
+    cuda-native-redist
+    ninja
+    which
+  ] ++ lists.optionals pythonSupport (with python3Packages; [
+    pip
+    setuptools
+    wheel
+  ]);
+
+  buildInputs = [
+    cuda-redist
+  ] ++ lib.optionals pythonSupport (
+    with python3Packages; [
+      pybind11
+      python
+    ]
+  );
+
+  propagatedBuildInputs = lib.optionals pythonSupport (
+    with python3Packages; [
+      torch
+    ]
+  );
+
+  # NOTE: We cannot use pythonImportsCheck for this module because it uses torch to immediately
+  #   initailize CUDA and GPU access is not allowed in the nix build environment.
+  # NOTE: There are no tests for the C++ library or the python bindings, so we just skip the check
+  #   phase -- we're not missing anything.
+  doCheck = false;
+
+  preConfigure = ''
+    export TCNN_CUDA_ARCHITECTURES=${
+      strings.concatStringsSep ";" (lists.map cudaFlags.dropDot cudaFlags.cudaCapabilities)
+    }
+    export CUDA_HOME=${cuda-native-redist}
+    export LIBRARY_PATH=${cuda-native-redist}/lib/stubs:$LIBRARY_PATH
+    export CC=${backendStdenv.cc}/bin/cc
+    export CXX=${backendStdenv.cc}/bin/c++
+  '';
+
+  # When building the python bindings, we cannot re-use the artifacts from the C++ build so we
+  # skip the CMake confurePhase and the buildPhase.
+  dontUseCmakeConfigure = pythonSupport;
+
+  # The configurePhase usually puts you in the build directory, so for the python bindings we
+  # need to change directories to the source directory.
+  configurePhase = strings.optionalString pythonSupport ''
+    runHook preConfigure
+    mkdir -p $NIX_BUILD_TOP/build
+    cd $NIX_BUILD_TOP/build
+    runHook postConfigure
+  '';
+
+  buildPhase = strings.optionalString pythonSupport ''
+    runHook preBuild
+    python -m pip wheel \
+      --no-build-isolation \
+      --no-clean \
+      --no-deps \
+      --no-index \
+      --verbose \
+      --wheel-dir $NIX_BUILD_TOP/build \
+      $NIX_BUILD_TOP/source/bindings/torch
+    runHook postBuild
+  '';
+
+  installPhase = ''
+    runHook preInstall
+    mkdir -p $out/lib
+  ''
+  # Installing the C++ library just requires copying the static library to the output directory
+  + strings.optionalString (!pythonSupport) ''
+    cp libtiny-cuda-nn.a $out/lib/
+  ''
+  # Installing the python bindings requires building the wheel and installing it
+  + strings.optionalString pythonSupport ''
+    python -m pip install \
+      --no-build-isolation \
+      --no-cache-dir \
+      --no-deps \
+      --no-index \
+      --no-warn-script-location \
+      --prefix="$out" \
+      --verbose \
+      ./*.whl
+  '' + ''
+    runHook postInstall
+  '';
+
+  passthru = {
+    inherit cudaPackages;
+  };
+
+  meta = with lib; {
+    description = "Lightning fast C++/CUDA neural network framework";
+    homepage = "https://github.com/NVlabs/tiny-cuda-nn";
+    license = licenses.bsd3;
+    maintainers = with maintainers; [ connorbaker ];
+    platforms = platforms.linux;
+  };
+})
diff --git a/pkgs/top-level/all-packages.nix b/pkgs/top-level/all-packages.nix
index bfecf795b2092..366c7316d4bf6 100644
--- a/pkgs/top-level/all-packages.nix
+++ b/pkgs/top-level/all-packages.nix
@@ -3812,6 +3812,8 @@ with pkgs;
 
   tensorflow-lite = callPackage ../development/libraries/science/math/tensorflow-lite { };
 
+  tiny-cuda-nn = callPackage ../development/libraries/science/math/tiny-cuda-nn { };
+
   tezos-rust-libs = callPackage ../development/libraries/tezos-rust-libs { };
 
   behave = with python3Packages; toPythonApplication behave;
diff --git a/pkgs/top-level/python-packages.nix b/pkgs/top-level/python-packages.nix
index de3f2ba61f0fe..def6dd5472404 100644
--- a/pkgs/top-level/python-packages.nix
+++ b/pkgs/top-level/python-packages.nix
@@ -11694,6 +11694,12 @@ self: super: with self; {
 
   timm = callPackage ../development/python-modules/timm { };
 
+  tiny-cuda-nn = toPythonModule (pkgs.tiny-cuda-nn.override {
+    cudaPackages = self.torch.cudaPackages;
+    python3Packages = self;
+    pythonSupport = true;
+  });
+
   tinycss2 = callPackage ../development/python-modules/tinycss2 { };
 
   tinycss = callPackage ../development/python-modules/tinycss { };