diff --git a/Cargo.lock b/Cargo.lock
index 2eb5cdfe71..c417ac4aa8 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -5015,6 +5015,7 @@ dependencies = [
  "ordered-float 5.0.0",
  "parking_lot",
  "portable-atomic",
+ "portable-atomic-util",
  "profiling",
  "range-alloc",
  "raw-window-handle 0.5.2",
diff --git a/Cargo.toml b/Cargo.toml
index e9f0254aed..52ef517577 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -159,7 +159,8 @@ pico-args = { version = "0.5", features = [
 ] }
 png = "0.17.6"
 pollster = "0.4"
-portable-atomic = "1.2"
+portable-atomic = "1.8"
+portable-atomic-util = "0.2.4"
 pp-rs = "0.2.1"
 profiling = { version = "1", default-features = false }
 quote = "1.0.38"
diff --git a/wgpu-hal/Cargo.toml b/wgpu-hal/Cargo.toml
index 07c52ddadb..94dc0e07e0 100644
--- a/wgpu-hal/Cargo.toml
+++ b/wgpu-hal/Cargo.toml
@@ -107,7 +107,6 @@ gles = [
     "naga/glsl-out",
     "dep:arrayvec",
     "dep:bytemuck",
-    "dep:cfg-if",
     "dep:glow",
     "dep:glutin_wgl_sys",
     "dep:hashbrown",
@@ -165,7 +164,7 @@ renderdoc = ["dep:libloading", "dep:renderdoc-sys", "dep:log"]
 fragile-send-sync-non-atomic-wasm = [
     "wgpu-types/fragile-send-sync-non-atomic-wasm",
 ]
-portable-atomic = ["dep:portable-atomic"]
+portable-atomic = ["dep:portable-atomic", "dep:portable-atomic-util"]
 
 ###################################
 ### Internal Debugging Features ###
@@ -207,6 +206,7 @@ wgpu-types = { workspace = true, default-features = false }
 
 # Dependencies in the lib and empty backend
 bitflags.workspace = true
+cfg-if.workspace = true
 raw-window-handle.workspace = true
 parking_lot = { workspace = true, optional = true }
 thiserror.workspace = true
@@ -216,14 +216,12 @@ arrayvec = { workspace = true, optional = true }
 bytemuck = { workspace = true, optional = true, features = ["derive"] }
 hashbrown = { workspace = true, optional = true }
 log = { workspace = true, optional = true }
-once_cell = { workspace = true, optional = true }
 ordered-float = { workspace = true, optional = true }
 profiling = { workspace = true, optional = true, default-features = false }
 rustc-hash = { workspace = true, optional = true }
 
 # Backend: GLES
 glow = { workspace = true, optional = true }
-cfg-if = { workspace = true, optional = true }
 
 ########################
 ### Platform: Native ###
@@ -318,14 +316,18 @@ khronos-egl = { workspace = true, optional = true, features = [
 # Note: it's unused by emscripten, but we keep it to have single code base in egl.rs
 libloading = { workspace = true, optional = true }
 
-[target.'cfg(not(target_has_atomic = "64"))'.dependencies]
+[target.'cfg(any(not(target_has_atomic = "64"), not(target_has_atomic = "ptr")))'.dependencies]
 portable-atomic = { workspace = true, optional = true }
 
+[target.'cfg(not(target_has_atomic = "ptr"))'.dependencies]
+portable-atomic-util = { workspace = true, features = [
+    "alloc",
+], optional = true }
+
 [build-dependencies]
 cfg_aliases.workspace = true
 
 [dev-dependencies]
-cfg-if.workspace = true
 env_logger.workspace = true
 glam.workspace = true                                            # for ray-traced-triangle example
 naga = { workspace = true, features = ["wgsl-in", "termcolor"] }
diff --git a/wgpu-hal/build.rs b/wgpu-hal/build.rs
index 960f645300..a89db73d53 100644
--- a/wgpu-hal/build.rs
+++ b/wgpu-hal/build.rs
@@ -25,6 +25,7 @@ fn main() {
         vulkan: { all(not(target_arch = "wasm32"), feature = "vulkan") },
         // ⚠️ Keep in sync with target.cfg() definition in Cargo.toml and cfg_alias in `wgpu` crate ⚠️
         static_dxc: { all(target_os = "windows", feature = "static-dxc", not(target_arch = "aarch64")) },
-        supports_64bit_atomics: { target_has_atomic = "64" }
+        supports_64bit_atomics: { target_has_atomic = "64" },
+        supports_ptr_atomics: { target_has_atomic = "ptr" }
     }
 }
diff --git a/wgpu-hal/src/lib.rs b/wgpu-hal/src/lib.rs
index 82a5af8d42..c5d81b2860 100644
--- a/wgpu-hal/src/lib.rs
+++ b/wgpu-hal/src/lib.rs
@@ -292,7 +292,7 @@ pub use dynamic::{
 
 #[allow(unused)]
 use alloc::boxed::Box;
-use alloc::{borrow::Cow, string::String, sync::Arc, vec::Vec};
+use alloc::{borrow::Cow, string::String, vec::Vec};
 use core::{
     borrow::Borrow,
     error::Error,
@@ -306,6 +306,14 @@ use bitflags::bitflags;
 use thiserror::Error;
 use wgt::WasmNotSendSync;
 
+cfg_if::cfg_if! {
+    if #[cfg(supports_ptr_atomics)] {
+        use alloc::sync::Arc;
+    } else if #[cfg(feature = "portable-atomic")] {
+        use portable_atomic_util::Arc;
+    }
+}
+
 // - Vertex + Fragment
 // - Compute
 // Task + Mesh + Fragment
@@ -450,9 +458,18 @@ impl InstanceError {
     }
     #[allow(dead_code)] // may be unused on some platforms
     pub(crate) fn with_source(message: String, source: impl Error + Send + Sync + 'static) -> Self {
+        cfg_if::cfg_if! {
+            if #[cfg(supports_ptr_atomics)] {
+                let source = Arc::new(source);
+            } else {
+                // TODO(https://github.com/rust-lang/rust/issues/18598): avoid indirection via Box once arbitrary types support unsized coercion
+                let source: Box<dyn Error + Send + Sync + 'static> = Box::new(source);
+                let source = Arc::from(source);
+            }
+        }
         Self {
             message,
-            source: Some(Arc::new(source)),
+            source: Some(source),
         }
     }
 }
diff --git a/wgpu-hal/src/noop/buffer.rs b/wgpu-hal/src/noop/buffer.rs
index 6e9572ef8f..9b8ab17a1f 100644
--- a/wgpu-hal/src/noop/buffer.rs
+++ b/wgpu-hal/src/noop/buffer.rs
@@ -1,6 +1,14 @@
-use alloc::{sync::Arc, vec::Vec};
+use alloc::vec::Vec;
 use core::{cell::UnsafeCell, ops::Range, ptr};
 
+cfg_if::cfg_if! {
+    if #[cfg(supports_ptr_atomics)] {
+        use alloc::sync::Arc;
+    } else if #[cfg(feature = "portable-atomic")] {
+        use portable_atomic_util::Arc;
+    }
+}
+
 #[derive(Clone, Debug)]
 pub struct Buffer {
     /// This data is potentially accessed mutably in arbitrary non-overlapping slices,