diff --git a/test/gpuarrays_tests.jl b/test/gpuarrays_tests.jl
index 0a0be3106..b4318ad5a 100644
--- a/test/gpuarrays_tests.jl
+++ b/test/gpuarrays_tests.jl
@@ -1,89 +1,22 @@
-@testsetup module TSGPUArrays
-    export gpuarrays_test
+@testitem "gpuarrays" begin
 
     using AMDGPU
     import GPUArrays
     include(joinpath(pkgdir(GPUArrays), "test", "testsuite.jl"))
 
-    gpuarrays_test(test_name::String) = TestSuite.tests[test_name](ROCArray)
-end
-# TODO define `@testitem` dynamically
+    const gpuarrays_tests = collect(keys(TestSuite.tests))
 
-@testitem "gpuarrays - base" setup=[TSGPUArrays] begin
-    gpuarrays_test("base")
-end
-@testitem "gpuarrays - broadcasting" setup=[TSGPUArrays] begin
-    gpuarrays_test("broadcasting")
-end
-@testitem "gpuarrays - constructors" setup=[TSGPUArrays] begin
-    gpuarrays_test("constructors")
-end
-@testitem "gpuarrays - indexing find" setup=[TSGPUArrays] begin
-    gpuarrays_test("indexing find")
-end
-@testitem "gpuarrays - indexing multidimensional" setup=[TSGPUArrays] begin
-    if Sys.islinux() # TODO Windows does not support hostcalls.
-        gpuarrays_test("indexing multidimensional")
-        AMDGPU.synchronize(; stop_hostcalls=true)
+    run_gpuarrays_test(test_name::String) =
+        TestSuite.tests[test_name](ROCArray)
+
+    for test_name in gpuarrays_tests
+        if test_name == "indexing multidimensional" && !Sys.islinux()
+            # Skipping indexing multidimensional test on non-Linux
+        else
+            run_gpuarrays_test(test_name)
+            if test_name == "indexing multidimensional"
+                AMDGPU.synchronize(; stop_hostcalls=true)
+            end
+        end
     end
 end
-@testitem "gpuarrays - indexing scalar" setup=[TSGPUArrays] begin
-    gpuarrays_test("indexing scalar")
-end
-@testitem "gpuarrays - linalg/core" setup=[TSGPUArrays] begin
-    gpuarrays_test("linalg/core")
-end
-@testitem "gpuarrays - linalg/mul!/matrix-matrix" setup=[TSGPUArrays] begin
-    gpuarrays_test("linalg/mul!/matrix-matrix")
-end
-@testitem "gpuarrays - linalg/mul!/vector-matrix" setup=[TSGPUArrays] begin
-    gpuarrays_test("linalg/mul!/vector-matrix")
-end
-@testitem "gpuarrays - linalg/norm" setup=[TSGPUArrays] begin
-    gpuarrays_test("linalg/norm")
-end
-@testitem "gpuarrays - math/intrinsics" setup=[TSGPUArrays] begin
-    gpuarrays_test("math/intrinsics")
-end
-@testitem "gpuarrays - math/power" setup=[TSGPUArrays] begin
-    gpuarrays_test("math/power")
-end
-@testitem "gpuarrays - random" setup=[TSGPUArrays] begin
-    gpuarrays_test("random")
-end
-@testitem "gpuarrays - reductions/== isequal" setup=[TSGPUArrays] begin
-    gpuarrays_test("reductions/== isequal")
-end
-@testitem "gpuarrays - reductions/any all count" setup=[TSGPUArrays] begin
-    gpuarrays_test("reductions/any all count")
-end
-@testitem "gpuarrays - reductions/mapreduce" setup=[TSGPUArrays] begin
-    gpuarrays_test("reductions/mapreduce")
-end
-@testitem "gpuarrays - reductions/mapreducedim!" setup=[TSGPUArrays] begin
-    gpuarrays_test("reductions/mapreducedim!")
-end
-@testitem "gpuarrays - reductions/mapreducedim!_large" setup=[TSGPUArrays] begin
-    gpuarrays_test("reductions/mapreducedim!_large")
-end
-@testitem "gpuarrays - reductions/minimum maximum extrema" setup=[TSGPUArrays] begin
-    gpuarrays_test("reductions/minimum maximum extrema")
-end
-@testitem "gpuarrays - reductions/reduce" setup=[TSGPUArrays] begin
-    gpuarrays_test("reductions/reduce")
-end
-@testitem "gpuarrays - reductions/reducedim!" setup=[TSGPUArrays] begin
-    gpuarrays_test("reductions/reducedim!")
-end
-@testitem "gpuarrays - reductions/sum prod" setup=[TSGPUArrays] begin
-    gpuarrays_test("reductions/sum prod")
-end
-@testitem "gpuarrays - statistics" setup=[TSGPUArrays] begin
-    gpuarrays_test("statistics")
-end
-@testitem "gpuarrays - uniformscaling" setup=[TSGPUArrays] begin
-    gpuarrays_test("uniformscaling")
-end
-@testitem "gpuarrays - alloc cache" setup=[TSGPUArrays] begin
-    gpuarrays_test("alloc cache")
-end