diff --git a/CHANGELOG.md b/CHANGELOG.md
index 243f35b767..485945fa39 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -45,6 +45,8 @@ Bottom level categories:
 
 ### New features
 
+#### General
+
 - Many numeric built-ins have had a constant evaluation implementation added for them, which allows them to be used in a `const` context:
     - [#4879](https://github.com/gfx-rs/wgpu/pull/4879) by @ErichDonGubler:
         - `abs`
@@ -64,6 +66,8 @@ Bottom level categories:
         - `step`
         - `tan`
         - `tanh`
+- Eager release of GPU resources comes from device.trackers. By @bradwerth in [#5075](https://github.com/gfx-rs/wgpu/pull/5075)
+
 
 ### Bug Fixes
 
diff --git a/wgpu-core/src/device/life.rs b/wgpu-core/src/device/life.rs
index 22699d1068..b454fcaa8f 100644
--- a/wgpu-core/src/device/life.rs
+++ b/wgpu-core/src/device/life.rs
@@ -837,47 +837,4 @@ impl<A: HalApi> LifetimeTracker<A> {
         }
         pending_callbacks
     }
-
-    pub(crate) fn release_gpu_resources(&mut self) {
-        // This is called when the device is lost, which makes every associated
-        // resource invalid and unusable. This is an opportunity to release all of
-        // the underlying gpu resources, even though the objects remain visible to
-        // the user agent. We purge this memory naturally when resources have been
-        // moved into the appropriate buckets, so this function just needs to
-        // initiate movement into those buckets, and it can do that by calling
-        // "destroy" on all the resources we know about which aren't already marked
-        // for cleanup.
-
-        // During these iterations, we discard all errors. We don't care!
-
-        // Destroy all the mapped buffers.
-        for buffer in &self.mapped {
-            let _ = buffer.destroy();
-        }
-
-        // Destroy all the unmapped buffers.
-        for buffer in &self.ready_to_map {
-            let _ = buffer.destroy();
-        }
-
-        // Destroy all the future_suspected_buffers.
-        for buffer in &self.future_suspected_buffers {
-            let _ = buffer.destroy();
-        }
-
-        // Destroy the buffers in all active submissions.
-        for submission in &self.active {
-            for buffer in &submission.mapped {
-                let _ = buffer.destroy();
-            }
-        }
-
-        // Destroy all the future_suspected_textures.
-        // TODO: texture.destroy is not implemented
-        /*
-        for texture in &self.future_suspected_textures {
-            let _ = texture.destroy();
-        }
-        */
-    }
 }
diff --git a/wgpu-core/src/device/resource.rs b/wgpu-core/src/device/resource.rs
index 547774ec63..3523981562 100644
--- a/wgpu-core/src/device/resource.rs
+++ b/wgpu-core/src/device/resource.rs
@@ -380,7 +380,7 @@ impl<A: HalApi> Device<A> {
         let mut device_lost_invocations = SmallVec::new();
         if !self.is_valid() && life_tracker.queue_empty() {
             // We can release gpu resources associated with this device.
-            life_tracker.release_gpu_resources();
+            self.release_gpu_resources();
 
             // If we have a DeviceLostClosure, build an invocation with the
             // reason DeviceLostReason::Destroyed and no message.
@@ -3331,7 +3331,6 @@ impl<A: HalApi> Device<A> {
             // It's important to not hold the lock while calling the closure.
             drop(life_lock);
             device_lost_closure.call(DeviceLostReason::Unknown, message.to_string());
-            life_lock = self.lock_life();
         }
 
         // 2) Complete any outstanding mapAsync() steps.
@@ -3343,7 +3342,26 @@ impl<A: HalApi> Device<A> {
         // until they are cleared, and then drop the device.
 
         // Eagerly release GPU resources.
-        life_lock.release_gpu_resources();
+        self.release_gpu_resources();
+    }
+
+    pub(crate) fn release_gpu_resources(&self) {
+        // This is called when the device is lost, which makes every associated
+        // resource invalid and unusable. This is an opportunity to release all of
+        // the underlying gpu resources, even though the objects remain visible to
+        // the user agent. We purge this memory naturally when resources have been
+        // moved into the appropriate buckets, so this function just needs to
+        // initiate movement into those buckets, and it can do that by calling
+        // "destroy" on all the resources we know about.
+
+        // During these iterations, we discard all errors. We don't care!
+        let trackers = self.trackers.lock();
+        for buffer in trackers.buffers.used_resources() {
+            let _ = buffer.destroy();
+        }
+        for texture in trackers.textures.used_resources() {
+            let _ = texture.destroy();
+        }
     }
 }