diff --git a/test/FoldsCUDATests/Project.toml b/test/FoldsCUDATests/Project.toml
index 5b3c6c0..e78ac7a 100644
--- a/test/FoldsCUDATests/Project.toml
+++ b/test/FoldsCUDATests/Project.toml
@@ -21,6 +21,7 @@ Referenceables = "42d2dcc6-99eb-4e98-b66c-637b7d73030e"
 Setfield = "efcf1570-3423-57d1-acb7-fd33fddbac46"
 SplittablesBase = "171d559e-b47b-412a-8079-5efa626c420e"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
+TestFunctionRunner = "792026f5-ac9a-4a19-adcb-47b0ce2deb5d"
 Transducers = "28d57a85-8fef-5791-bfe6-a80928e7c999"
 
 [compat]
diff --git a/test/FoldsCUDATests/src/FoldsCUDATests.jl b/test/FoldsCUDATests/src/FoldsCUDATests.jl
index 9b16ad7..a934a95 100644
--- a/test/FoldsCUDATests/src/FoldsCUDATests.jl
+++ b/test/FoldsCUDATests/src/FoldsCUDATests.jl
@@ -1,115 +1,41 @@
 module FoldsCUDATests
 
+import CUDA
 using GPUArrays
 using Test
+using TestFunctionRunner
 
 include("utils.jl")
 
-function include_tests(m = @__MODULE__, dir = @__DIR__)
-    for file in readdir(dir)
-        if match(r"^test_.*\.jl$", file) !== nothing
-            Base.include(m, joinpath(dir, file))
-        end
-    end
-end
-
-include_tests()
-
-module Generic
-using ..FoldsCUDATests: include_tests
+module TestGeneric
+using ..Utils: include_tests
 include_tests(@__MODULE__, joinpath(@__DIR__, "generic"))
 end
 
-should_test_gpu() =
-    lowercase(get(ENV, "JULIA_PKGEVAL", "false")) != "true" &&
-    lowercase(get(ENV, "CUDAFOLDS_JL_TEST_GPU", "true")) == "true"
-
-requires_gpu(m::Module) = parentmodule(m) === @__MODULE__
-
-function collect_modules(root::Module)
-    modules = Module[]
-    for n in names(root, all = true)
-        m = getproperty(root, n)
-        m isa Module || continue
-        m === root && continue
-        startswith(string(nameof(m)), "Test") || continue
-        push!(modules, m)
-    end
-    return modules
-end
-
-function collect_modules()
-    modules = collect_modules(@__MODULE__)
-    append!(modules, collect_modules(Generic))
-    return modules
+module TestGPU
+using ..Utils: include_tests, should_test_gpu
+include_tests(@__MODULE__)
+should_test_module() = should_test_gpu()
 end
 
-this_project() = joinpath(dirname(@__DIR__), "Project.toml")
-
-function is_in_path()
-    project = this_project()
-    paths = Base.load_path()
-    project in paths && return true
-    realproject = realpath(project)
-    realproject in paths && return true
-    matches(path) = path == project || path == realproject
-    return any(paths) do path
-        matches(path) || matches(realpath(path))
-    end
-end
+""" A list of tests to be run from UnionArrays.jl """
+const UNIONARRAYS_TESTS = [TestGPU.TestTypeChangingAccumulators]
 
-function with_project(f)
-    is_in_path() && return f()
-    load_path = copy(LOAD_PATH)
-    push!(LOAD_PATH, this_project())
-    try
-        f()
-    finally
-        append!(empty!(LOAD_PATH), load_path)
-    end
+function runtests_unionarrays(; kwargs...)
+    TestFunctionRunner.run(UNIONARRAYS_TESTS; kwargs...)
 end
 
-function runtests(modules = collect_modules())
-    with_project() do
-        runtests_impl(modules)
+function before_test_module()
+    if !CUDA.functional()
+        @info "CUDA.jl not functional"
+        return
     end
-end
 
-function runtests_impl(modules)
     GPUArrays.allowscalar(false)
-    test_gpu = should_test_gpu()
-    @testset "$(nameof(m))" for m in modules
-        if !test_gpu && requires_gpu(m)
-            continue  # branch inside `for` loop for printing skipped tests
-        end
-        if m in (
-            Generic.TestReducePartitionBy,
-            TestCoalescedCUDAEx,
-            TestTypeChangingAccumulators,
-        )
-            VERSION < v"1.6-" && continue
-        end
-        tests = map(names(m, all = true)) do n
-            n == :test || startswith(string(n), "test_") || return nothing
-            f = getproperty(m, n)
-            f !== m || return nothing
-            parentmodule(f) === m || return nothing
-            applicable(f) || return nothing  # removed by Revise?
-            return f
-        end
-        filter!(!isnothing, tests)
-        @testset "$f" for f in tests
-            f()
-        end
-    end
-end
-
-""" A list of tests to be run from UnionArrays.jl """
-const UNIONARRAYS_TESTS = [TestTypeChangingAccumulators]
 
-function runtests_unionarrays()
-    with_project() do
-        runtests_impl(UNIONARRAYS_TESTS)
+    if lowercase(get(ENV, "CI", "false")) == "true"
+        CUDA.versioninfo()
+        println()
     end
 end
 
diff --git a/test/FoldsCUDATests/src/utils.jl b/test/FoldsCUDATests/src/utils.jl
index a9dfdc1..c1280e8 100644
--- a/test/FoldsCUDATests/src/utils.jl
+++ b/test/FoldsCUDATests/src/utils.jl
@@ -2,6 +2,18 @@ module Utils
 
 using CUDA
 
+function include_tests(m, dir = @__DIR__)
+    for file in readdir(dir)
+        if match(r"^test_.*\.jl$", file) !== nothing
+            Base.include(m, joinpath(dir, file))
+        end
+    end
+end
+
+should_test_gpu() =
+    lowercase(get(ENV, "JULIA_PKGEVAL", "false")) != "true" &&
+    lowercase(get(ENV, "CUDAFOLDS_JL_TEST_GPU", "true")) == "true"
+
 """
     unsafe_free_all!(namespace::Module)
 
diff --git a/test/Project.toml b/test/Project.toml
index 6e2a597..173997f 100644
--- a/test/Project.toml
+++ b/test/Project.toml
@@ -16,8 +16,10 @@ Referenceables = "42d2dcc6-99eb-4e98-b66c-637b7d73030e"
 Setfield = "efcf1570-3423-57d1-acb7-fd33fddbac46"
 SplittablesBase = "171d559e-b47b-412a-8079-5efa626c420e"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
+TestFunctionRunner = "792026f5-ac9a-4a19-adcb-47b0ce2deb5d"
 Transducers = "28d57a85-8fef-5791-bfe6-a80928e7c999"
 
 [compat]
 Aqua = "0.5"
 Random123 = "~1.3"
+TestFunctionRunner = "0.1.3"
diff --git a/test/environments/jl16/Manifest.toml b/test/environments/jl16/Manifest.toml
index 8d31fd9..e7a5dc0 100644
--- a/test/environments/jl16/Manifest.toml
+++ b/test/environments/jl16/Manifest.toml
@@ -195,7 +195,7 @@ uuid = "1457febb-a09b-4652-98c9-46b8ccd8ff53"
 version = "0.1.0"
 
 [[FoldsCUDATests]]
-deps = ["Adapt", "Aqua", "CUDA", "Documenter", "FLoops", "Folds", "FoldsCUDA", "FoldsCUDABenchmarks", "GPUArrays", "InitialValues", "LiterateTest", "Random", "Random123", "Referenceables", "Setfield", "SplittablesBase", "Test", "Transducers"]
+deps = ["Adapt", "Aqua", "CUDA", "Documenter", "FLoops", "Folds", "FoldsCUDA", "FoldsCUDABenchmarks", "GPUArrays", "InitialValues", "LiterateTest", "Random", "Random123", "Referenceables", "Setfield", "SplittablesBase", "Test", "TestFunctionRunner", "Transducers"]
 path = "../../FoldsCUDATests"
 uuid = "d11caea5-3c98-4cd5-8a56-9589fe6662ee"
 version = "0.1.0"
@@ -310,6 +310,12 @@ git-tree-sha1 = "f10e6f4d46da26a33e9cea7a5aed3476b592bbfb"
 uuid = "d77d25b0-90d3-4a16-b10a-412a9d48f625"
 version = "0.1.4"
 
+[[LoadAllPackages]]
+deps = ["Pkg"]
+git-tree-sha1 = "de9eababca8fdc9b6b207ed05f99693faa608230"
+uuid = "b37bcd2d-1570-475d-a8c6-9b4fae6d0ba9"
+version = "0.1.0"
+
 [[LogExpFunctions]]
 deps = ["ChainRulesCore", "ChangesOfVariables", "DocStringExtensions", "InverseFunctions", "IrrationalConstants", "LinearAlgebra"]
 git-tree-sha1 = "e5718a00af0ab9756305a0392832c8952c7426c1"
@@ -494,10 +500,21 @@ version = "1.6.1"
 deps = ["ArgTools", "SHA"]
 uuid = "a4e569a6-e804-4fa4-b0f3-eef7a1d5b13e"
 
+[[Terminators]]
+git-tree-sha1 = "e72e157c08773595052733e8b39944a49b0109d9"
+uuid = "b0858c3f-a9bf-4cba-a0e3-6aaeea5477ac"
+version = "0.1.1"
+
 [[Test]]
 deps = ["InteractiveUtils", "Logging", "Random", "Serialization"]
 uuid = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
+[[TestFunctionRunner]]
+deps = ["Distributed", "LoadAllPackages", "Pkg", "Terminators", "Test"]
+git-tree-sha1 = "5e58b00d31869b4972aad04e6b34558653b2831f"
+uuid = "792026f5-ac9a-4a19-adcb-47b0ce2deb5d"
+version = "0.1.3"
+
 [[TimerOutputs]]
 deps = ["ExprTools", "Printf"]
 git-tree-sha1 = "97e999be94a7147d0609d0b9fc9feca4bf24d76b"
diff --git a/test/environments/jl16/Project.toml b/test/environments/jl16/Project.toml
index 34e50bf..8cca63a 100644
--- a/test/environments/jl16/Project.toml
+++ b/test/environments/jl16/Project.toml
@@ -2,3 +2,4 @@
 FoldsCUDA = "6cd66ae4-5932-4b96-926d-e73e578e42cc"
 FoldsCUDABenchmarks = "1457febb-a09b-4652-98c9-46b8ccd8ff53"
 FoldsCUDATests = "d11caea5-3c98-4cd5-8a56-9589fe6662ee"
+TestFunctionRunner = "792026f5-ac9a-4a19-adcb-47b0ce2deb5d"
diff --git a/test/environments/jl17/Manifest.toml b/test/environments/jl17/Manifest.toml
index d250a28..a9b86de 100644
--- a/test/environments/jl17/Manifest.toml
+++ b/test/environments/jl17/Manifest.toml
@@ -198,7 +198,7 @@ uuid = "1457febb-a09b-4652-98c9-46b8ccd8ff53"
 version = "0.1.0"
 
 [[deps.FoldsCUDATests]]
-deps = ["Adapt", "Aqua", "CUDA", "Documenter", "FLoops", "Folds", "FoldsCUDA", "FoldsCUDABenchmarks", "GPUArrays", "InitialValues", "LiterateTest", "Random", "Random123", "Referenceables", "Setfield", "SplittablesBase", "Test", "Transducers"]
+deps = ["Adapt", "Aqua", "CUDA", "Documenter", "FLoops", "Folds", "FoldsCUDA", "FoldsCUDABenchmarks", "GPUArrays", "InitialValues", "LiterateTest", "Random", "Random123", "Referenceables", "Setfield", "SplittablesBase", "Test", "TestFunctionRunner", "Transducers"]
 path = "../../FoldsCUDATests"
 uuid = "d11caea5-3c98-4cd5-8a56-9589fe6662ee"
 version = "0.1.0"
@@ -313,6 +313,12 @@ git-tree-sha1 = "f10e6f4d46da26a33e9cea7a5aed3476b592bbfb"
 uuid = "d77d25b0-90d3-4a16-b10a-412a9d48f625"
 version = "0.1.4"
 
+[[deps.LoadAllPackages]]
+deps = ["Pkg"]
+git-tree-sha1 = "de9eababca8fdc9b6b207ed05f99693faa608230"
+uuid = "b37bcd2d-1570-475d-a8c6-9b4fae6d0ba9"
+version = "0.1.0"
+
 [[deps.LogExpFunctions]]
 deps = ["ChainRulesCore", "ChangesOfVariables", "DocStringExtensions", "InverseFunctions", "IrrationalConstants", "LinearAlgebra"]
 git-tree-sha1 = "e5718a00af0ab9756305a0392832c8952c7426c1"
@@ -501,10 +507,21 @@ version = "1.6.1"
 deps = ["ArgTools", "SHA"]
 uuid = "a4e569a6-e804-4fa4-b0f3-eef7a1d5b13e"
 
+[[deps.Terminators]]
+git-tree-sha1 = "e72e157c08773595052733e8b39944a49b0109d9"
+uuid = "b0858c3f-a9bf-4cba-a0e3-6aaeea5477ac"
+version = "0.1.1"
+
 [[deps.Test]]
 deps = ["InteractiveUtils", "Logging", "Random", "Serialization"]
 uuid = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
+[[deps.TestFunctionRunner]]
+deps = ["Distributed", "LoadAllPackages", "Pkg", "Terminators", "Test"]
+git-tree-sha1 = "9b4dc94c31f85a386e7e7793d78dd4c237f8cfdc"
+uuid = "792026f5-ac9a-4a19-adcb-47b0ce2deb5d"
+version = "0.1.4"
+
 [[deps.TimerOutputs]]
 deps = ["ExprTools", "Printf"]
 git-tree-sha1 = "97e999be94a7147d0609d0b9fc9feca4bf24d76b"
diff --git a/test/environments/jl17/Project.toml b/test/environments/jl17/Project.toml
index 34e50bf..8cca63a 100644
--- a/test/environments/jl17/Project.toml
+++ b/test/environments/jl17/Project.toml
@@ -2,3 +2,4 @@
 FoldsCUDA = "6cd66ae4-5932-4b96-926d-e73e578e42cc"
 FoldsCUDABenchmarks = "1457febb-a09b-4652-98c9-46b8ccd8ff53"
 FoldsCUDATests = "d11caea5-3c98-4cd5-8a56-9589fe6662ee"
+TestFunctionRunner = "792026f5-ac9a-4a19-adcb-47b0ce2deb5d"
diff --git a/test/load.jl b/test/load.jl
deleted file mode 100644
index 409a626..0000000
--- a/test/load.jl
+++ /dev/null
@@ -1,14 +0,0 @@
-try
-    using FoldsCUDATests
-    true
-catch
-    false
-end || begin
-    let path = joinpath(@__DIR__, "FoldsCUDATests")
-        path in LOAD_PATH || push!(LOAD_PATH, path)
-    end
-    let path = joinpath(@__DIR__, "../benchmark/FoldsCUDABenchmarks")
-        path in LOAD_PATH || push!(LOAD_PATH, path)
-    end
-    using FoldsCUDATests
-end
diff --git a/test/runtests.jl b/test/runtests.jl
index 00b6ffd..1f4ebec 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -7,5 +7,6 @@ if VERSION ≥ v"1.6-"
         push!(LOAD_PATH, "@stdlib")
     end
 end
-include("load.jl")
-FoldsCUDATests.runtests()
+
+using TestFunctionRunner
+TestFunctionRunner.@run(paths = ["../benchmark/FoldsCUDABenchmarks"])