diff --git a/.github/workflows/cd.yml b/.github/workflows/cd.yml
index e1099f87..d2a43138 100644
--- a/.github/workflows/cd.yml
+++ b/.github/workflows/cd.yml
@@ -184,6 +184,7 @@ jobs:
             PY_OMP=1
             PY_CXXFLAGS_OMP="-Xpreprocessor -fopenmp"
             PY_LDFLAGS_OMP="-Xpreprocessor -fopenmp"
+            PY_BUILD_PARALLEL='-j'
             MACOSX_DEPLOYMENT_TARGET=${{ env.MACOSX_VERS_MIN }}
             SYSTEM_VERSION_COMPAT=0
           CIBW_BEFORE_ALL_MACOS:
diff --git a/.github/workflows/cd_cuda.yml b/.github/workflows/cd_cuda.yml
index ae4c3c37..11fafa58 100644
--- a/.github/workflows/cd_cuda.yml
+++ b/.github/workflows/cd_cuda.yml
@@ -81,6 +81,7 @@ jobs:
             PY_CXX=nvcc
             PY_OMP=1
             PY_CUDA=1
+            PY_BUILD_PARALLEL='-j'
 
       - name: Verify built distribution
         run: python -m twine check --strict dist/*
@@ -128,7 +129,8 @@ jobs:
         with:
           log-file-suffix: '${{matrix.os}}_cudatoolkit_log.txt'
           method: 'network'
-          sub-packages: '["nvcc", "cufft"]'
+          sub-packages: '["nvcc"]'
+          non-cuda-sub-packages: '["libcufft"]'
 
       - name: Set up (Mini)conda
         uses: conda-incubator/setup-miniconda@v3
@@ -149,7 +151,7 @@ jobs:
         if: github.event_name == 'workflow_dispatch'
         uses: knicknic/os-specific-run@v1.0.4
         env:
-          recipe_file: deploy/pkg/conda_recipe/meta.yaml
+          recipe_file: deploy/pkg/conda_recipe_cuda/meta.yaml
         with:
           linux: |
             vers_tag=${{ github.event.inputs.version_tag }}
@@ -161,7 +163,7 @@ jobs:
 
       - name: Build Conda package
         env:
-          recipe_dir: deploy/pkg/conda_recipe
+          recipe_dir: deploy/pkg/conda_recipe_cuda
           output_dir: dist/
           variants: "{'python': ['3.10', '3.11', '3.12']}"
         run: |
diff --git a/.github/workflows/cd_xp.yml b/.github/workflows/cd_xp.yml
index 51fa6e1a..66787944 100644
--- a/.github/workflows/cd_xp.yml
+++ b/.github/workflows/cd_xp.yml
@@ -191,6 +191,7 @@ jobs:
             PY_OMP=1
             PY_CXXFLAGS_OMP="-Xpreprocessor -fopenmp"
             PY_LDFLAGS_OMP="-Xpreprocessor -fopenmp"
+            PY_BUILD_PARALLEL='-j'
             MACOSX_DEPLOYMENT_TARGET=${{ env.MACOSX_VERS_MIN }}
             SYSTEM_VERSION_COMPAT=0
           CIBW_BEFORE_ALL_MACOS:
diff --git a/.pyproject_cuda.toml b/.pyproject_cuda.toml
index a3ed2986..752e7bb9 100644
--- a/.pyproject_cuda.toml
+++ b/.pyproject_cuda.toml
@@ -96,9 +96,9 @@ skip = [
     '*_ppc64le',
     '*_s390x',
 ]
-environment = { PY_OMP='1', PY_CUDA='1', PY_BUILD_PARALLEL='-j', SYSTEM_VERSION_COMPAT='0' }
-test-requires = "pytest>=7.0"
-test-command = "pytest {project}/tests"
+environment = { PY_OMP='1', PY_CUDA='1', PY_BUILD_PARALLEL='-j' }
+# test-requires = "pytest>=7.0"
+# test-command = "pytest {project}/tests"
 
 manylinux-x86_64-image = 'manylinux_2_28'
 manylinux-aarch64-image = 'manylinux_2_28'