diff --git a/pkgs/development/libraries/science/math/nccl/default.nix b/pkgs/development/libraries/science/math/nccl/default.nix
index d3c20258425..badd08291de 100644
--- a/pkgs/development/libraries/science/math/nccl/default.nix
+++ b/pkgs/development/libraries/science/math/nccl/default.nix
@@ -1,24 +1,44 @@
-{ callPackage, cudatoolkit_8, cudatoolkit_9 }:
+{ stdenv, fetchFromGitHub, which, cudatoolkit }:
 
-let
-  generic = args: callPackage (import ./generic.nix (removeAttrs args ["cudatoolkit"])) {
-    inherit (args) cudatoolkit;
+stdenv.mkDerivation rec {
+  name = "nccl-${version}-cuda-${cudatoolkit.majorVersion}";
+  version = "2.4.2-1";
+
+  src = fetchFromGitHub {
+    owner = "NVIDIA";
+    repo = "nccl";
+    rev = "v${version}";
+    sha256 = "0aa4gv51nbmmdhx6vp40l249m4arp30sijrn6kwxdfi1k9kajiq5";
   };
 
-in
+  outputs = [ "out" "dev" ];
 
-{
-  nccl_cudatoolkit_8 = generic rec {
-    version = "2.1.4";
-    cudatoolkit = cudatoolkit_8;
-    srcName = "nccl_${version}-1+cuda${cudatoolkit.majorVersion}_x86_64.txz";
-    sha256 = "1lwwm8kdhna5m318yg304kl2gsz1jwhv4zv4gn8av2m57zh848zi";
-  };
+  nativeBuildInputs = [ which ];
 
-  nccl_cudatoolkit_9 = generic rec {
-    version = "2.1.4";
-    cudatoolkit = cudatoolkit_9;
-    srcName = "nccl_${version}-1+cuda${cudatoolkit.majorVersion}_x86_64.txz";
-    sha256 = "0pajmqzkacpszs63jh2hw2qqc49kj75kcf7r0ky8hdh560q8xn0p";
+  buildInputs = [ cudatoolkit ];
+
+  preConfigure = ''
+    patchShebangs src/collectives/device/gen_rules.sh
+  '';
+
+  makeFlags = [
+    "CUDA_HOME=${cudatoolkit}"
+    "PREFIX=$(out)"
+  ];
+
+  postFixup = ''
+    moveToOutput lib/libnccl_static.a $dev
+  '';
+
+  NIX_CFLAGS_COMPILE = [ "-Wno-unused-function" ];
+
+  enableParallelBuilding = true;
+
+  meta = with stdenv.lib; {
+    description = "Multi-GPU and multi-node collective communication primitives for NVIDIA GPUs";
+    homepage = https://developer.nvidia.com/nccl;
+    license = licenses.bsd3;
+    platforms = [ "x86_64-linux" ];
+    maintainers = with maintainers; [ mdaiter orivej ];
   };
 }
diff --git a/pkgs/development/libraries/science/math/nccl/generic.nix b/pkgs/development/libraries/science/math/nccl/generic.nix
deleted file mode 100644
index 609a7df51e5..00000000000
--- a/pkgs/development/libraries/science/math/nccl/generic.nix
+++ /dev/null
@@ -1,57 +0,0 @@
-{ version
-, srcName
-, sha256
-}:
-
-{ stdenv
-, lib
-, requireFile
-, cudatoolkit
-}:
-
-stdenv.mkDerivation rec {
-  name = "cudatoolkit-${cudatoolkit.majorVersion}-nccl-${version}";
-
-  inherit version;
-
-  src = requireFile rec {
-    name = srcName;
-    inherit sha256;
-    message = ''
-      This nix expression requires that ${name} is already part of the store.
-      Register yourself to NVIDIA Accelerated Computing Developer Program, retrieve the NCCL library
-      at https://developer.nvidia.com/nccl, and run the following command in the download directory:
-      nix-prefetch-url file://\$PWD/${name}
-    '';
-  };
-
-  unpackCmd = "tar xJf $src";
-
-  installPhase = ''
-    function fixRunPath {
-      p=$(patchelf --print-rpath $1)
-      patchelf --set-rpath "$p:${lib.makeLibraryPath [ stdenv.cc.cc ]}" $1
-    }
-    fixRunPath lib/libnccl.so
-
-    mkdir -p $out
-    cp -a include $out/include
-    cp -a lib $out/lib
-  '';
-
-  propagatedBuildInputs = [
-    cudatoolkit
-  ];
-
-  passthru = {
-    inherit cudatoolkit;
-  };
-
-  meta = with stdenv.lib; {
-    description = "Multi-GPU and multi-node collective communication primitives that are performance optimized for NVIDIA GPUs";
-    homepage = https://developer.nvidia.com/nccl;
-    license = licenses.unfree;
-    platforms = [ "x86_64-linux" ];
-    maintainers = with maintainers; [ mdaiter ];
-  };
-}
diff --git a/pkgs/top-level/all-packages.nix b/pkgs/top-level/all-packages.nix
index b33dfd77d7a..87b27bcd226 100644
--- a/pkgs/top-level/all-packages.nix
+++ b/pkgs/top-level/all-packages.nix
@@ -4311,11 +4311,10 @@ in
   nbd = callPackage ../tools/networking/nbd { };
   xnbd = callPackage ../tools/networking/xnbd { };
 
-  inherit (callPackages ../development/libraries/science/math/nccl { })
-    nccl_cudatoolkit_8
-    nccl_cudatoolkit_9;
-
-  nccl = nccl_cudatoolkit_9;
+  nccl = callPackage ../development/libraries/science/math/nccl { };
+  nccl_cudatoolkit_9_0 = nccl.override { cudatoolkit = cudatoolkit_9_0; };
+  nccl_cudatoolkit_9 = nccl.override { cudatoolkit = cudatoolkit_9; };
+  nccl_cudatoolkit_10 = nccl.override { cudatoolkit = cudatoolkit_10; };
 
   ndjbdns = callPackage ../tools/networking/ndjbdns { };