diff --git a/.buildkite/pipeline.yml b/.buildkite/pipeline.yml
index 43cfba3e..12135f33 100644
--- a/.buildkite/pipeline.yml
+++ b/.buildkite/pipeline.yml
@@ -15,8 +15,7 @@ steps:
                 println("+++ :julia: Running tests")
                 Pkg.test("AcceleratedKernels", test_args=["--CUDA"])'
     agents:
-      queue: "juliagpu"
-      cuda: "*"
+      queue: "cuda"
     if: build.message !~ /\[skip tests\]/
     timeout_in_minutes: 60
 
@@ -34,8 +33,7 @@ steps:
                 println("+++ :julia: Running tests")
                 Pkg.test("AcceleratedKernels", test_args=["--CUDA"])'
     agents:
-      queue: "juliagpu"
-      cuda: "*"
+      queue: "cuda"
     if: build.message !~ /\[skip tests\]/
     timeout_in_minutes: 60
 
@@ -55,8 +53,7 @@ steps:
                 println("+++ :julia: Running tests")
                 Pkg.test("AcceleratedKernels", test_args=["--AMDGPU"])'
     agents:
-      queue: "juliagpu"
-      rocm: "*"
+      queue: "rocm"
       rocmgpu: "*"
     if: build.message !~ /\[skip tests\]/
     timeout_in_minutes: 60
@@ -75,8 +72,7 @@ steps:
                 println("+++ :julia: Running tests")
                 Pkg.test("AcceleratedKernels", test_args=["--AMDGPU"])'
     agents:
-      queue: "juliagpu"
-      rocm: "*"
+      queue: "rocm"
       rocmgpu: "*"
     if: build.message !~ /\[skip tests\]/
     timeout_in_minutes: 60
@@ -97,8 +93,7 @@ steps:
                 println("+++ :julia: Running tests")
                 Pkg.test("AcceleratedKernels", test_args=["--oneAPI"])'
     agents:
-      queue: "juliagpu"
-      intel: "*"
+      queue: "oneapi"
     if: build.message !~ /\[skip tests\]/
     timeout_in_minutes: 60
 
@@ -116,8 +111,7 @@ steps:
                 println("+++ :julia: Running tests")
                 Pkg.test("AcceleratedKernels", test_args=["--oneAPI"])'
     agents:
-      queue: "juliagpu"
-      intel: "*"
+      queue: "oneapi"
     if: build.message !~ /\[skip tests\]/
     timeout_in_minutes: 60
 
@@ -137,9 +131,7 @@ steps:
                 println("+++ :julia: Running tests")
                 Pkg.test("AcceleratedKernels", test_args=["--Metal"])'
     agents:
-      queue: "juliaecosystem"
-      os: "macos"
-      arch: "aarch64"
+      queue: "metal"
     if: build.message !~ /\[skip tests\]/
     timeout_in_minutes: 60
 
@@ -157,8 +149,6 @@ steps:
                 println("+++ :julia: Running tests")
                 Pkg.test("AcceleratedKernels", test_args=["--Metal"])'
     agents:
-      queue: "juliaecosystem"
-      os: "macos"
-      arch: "aarch64"
+      queue: "metal"
     if: build.message !~ /\[skip tests\]/
     timeout_in_minutes: 60
diff --git a/.github/workflows/CI-CPU.yml b/.github/workflows/CI-CPU.yml
index 298d02ce..2944bf2b 100644
--- a/.github/workflows/CI-CPU.yml
+++ b/.github/workflows/CI-CPU.yml
@@ -31,7 +31,7 @@ jobs:
           - ubuntu-latest
           - macOS-latest
           - windows-latest
-          - macOS-13
+          - macOS-15-intel
         arch:
           - x64
           - x86
@@ -40,9 +40,9 @@ jobs:
           - JULIA_NUM_THREADS: 1
           - JULIA_NUM_THREADS: 2
         exclude:
-          - os: macOS-13
+          - os: macOS-15-intel
             arch: x86
-          - os: macOS-13
+          - os: macOS-15-intel
             arch: aarch64
           - os: macOS-latest
             arch: x86
diff --git a/Project.toml b/Project.toml
index 2fccea8b..3b60f510 100644
--- a/Project.toml
+++ b/Project.toml
@@ -1,7 +1,7 @@
 name = "AcceleratedKernels"
 uuid = "6a4ca0a5-0e36-4168-a932-d9be78d558f1"
 authors = ["Andrei-Leonard Nicusan <leonard@evophase.co.uk> and contributors"]
-version = "0.4.3"
+version = "0.4.4"
 
 [deps]
 ArgCheck = "dce04be8-c92d-5529-be00-80e4d2c0e197"
@@ -21,6 +21,6 @@ ArgCheck = "2"
 GPUArraysCore = "0.2.0"
 KernelAbstractions = "0.9.34"
 Markdown = "1"
-UnsafeAtomics = "0.3.0"
+UnsafeAtomics = "0.3, 0.4"
 julia = "1.10"
 oneAPI = "1, 2"
diff --git a/test/runtests.jl b/test/runtests.jl
index 716fd8e0..e4587649 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -1,8 +1,16 @@
+import Pkg
+
+# Install the back-end before loading anything, as it may need different versions of
+# packages that AcceleratedKernels depends on (e.g. UnsafeAtomics).
+for backend in ("CUDA", "oneAPI", "AMDGPU", "Metal")
+    "--$backend" in ARGS && Pkg.add(backend)
+end
+"--OpenCL" in ARGS && Pkg.add(["OpenCL", "pocl_jll"])
+
 import AcceleratedKernels as AK
 using KernelAbstractions
 using Test
 using Random
-import Pkg
 
 # Set to true when testing backends that support this
 const TEST_DL = Ref{Bool}(false)
@@ -11,13 +19,11 @@ const TEST_DL = Ref{Bool}(false)
 #   julia> import Pkg
 #   julia> Pkg.test(test_args=["--oneAPI"])
 if "--CUDA" in ARGS
-    Pkg.add("CUDA")
     using CUDA
     CUDA.versioninfo()
     const BACKEND = CUDABackend()
     TEST_DL[] = true
 elseif "--oneAPI" in ARGS
-    Pkg.add("oneAPI")
     using oneAPI
     oneAPI.versioninfo()
     const BACKEND = oneAPIBackend()
@@ -25,20 +31,15 @@ elseif "--oneAPI" in ARGS
     # FIXME: need atomic orderings for `DecoupledLookback` in oneAPI
     # TEST_DL[] = true
 elseif "--AMDGPU" in ARGS
-    Pkg.add("AMDGPU")
     using AMDGPU
     AMDGPU.versioninfo()
     const BACKEND = ROCBackend()
     TEST_DL[] = true
 elseif "--Metal" in ARGS
-    Pkg.add("Metal")
     using Metal
     Metal.versioninfo()
     const BACKEND = MetalBackend()
 elseif "--OpenCL" in ARGS
-    Pkg.add(name="OpenCL", rev="master")
-    Pkg.add(name="SPIRVIntrinsics", rev="master")
-    Pkg.add("pocl_jll")
     using pocl_jll
     using OpenCL
     OpenCL.versioninfo()