# UE 5.8.2 — Severe GPU Residency Stalls During World Partition Distance-Field Buffer Resizing 

**URL:** <https://forums.unrealengine.com/t/ue-5-8-2-severe-gpu-residency-stalls-during-world-partition-distance-field-buffer-resizing/2841748>\
**Category:** Issues and Bug Reporting (Unreal Engine)\
**Tags:** unreal-engine, question\
**Created:** [October 16, 2026, 5:21pm UTC](https://forums.unrealengine.com/t/ue-5-8-2-severe-gpu-residency-stalls-during-world-partition-distance-field-buffer-resizing/2841748 "2026-10-16T17:21:19Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Tarbasch](https://d1ap1mz92jnks1.cloudfront.net/user_avatar/forums.unrealengine.com/tarbasch/32/1212834_2.png) [@Tarbasch](https://forums.unrealengine.com/u/Tarbasch)\
**Post date:** [October 16, 2026, 5:21pm UTC](https://forums.unrealengine.com/t/ue-5-8-2-severe-gpu-residency-stalls-during-world-partition-distance-field-buffer-resizing/2841748/1 "2026-10-16T17:21:20Z")

</div>

### Summary

In Unreal Engine 5.8.2, streaming dense World Partition foliage causes severe frame hitches associated with distance-field object-buffer replacement and GPU residency synchronization.

The distance-field Data/Bounds buffers change between 1,664 MiB and 3,328 MiB when the active object population crosses a power-of-two capacity threshold. Temporary replacement allocations can overlap at approximately 4,992 MiB.

GPU captures reproduced residency-fence waits exceeding one second, correlated with long occlusion-query waits. The problem occurs during repeated forward/return traversal, including standalone testing.

We are using an RTX 5080 (16 GB VRAM) at 5120×1440. Existing measurements show insufficient conservative VRAM headroom for simply retaining larger buffers.

We would appreciate Epic’s guidance on whether this is a known issue, whether a supported correction exists, and whether the UE 5.8.3 D3D12 synchronization change (UE-387332) addresses this particular resource-retirement and residency path.

### What Type of Bug are you experiencing?

Rendering (Graphics / Niagara)

### Steps to Reproduce

1. Open a dense, foliage-heavy World Partition level in Unreal Engine 5.8.2 using DirectX 12 with distance-field lighting enabled.
2. Run the level in standalone mode at 5120×1440 on an RTX 5080 (16 GB VRAM).
3. Traverse between World Partition cells containing large numbers of distance-field-enabled foliage instances.
4. Repeat forward and return traversal across the same cell boundaries.
5. Observe frame times and GPU residency synchronization while distance-field object counts cross the 8,388,608-slot capacity threshold.
6. In the affected project, the combined distance-field object buffers resize between 1,664 MiB and 3,328 MiB, correlating with severe frame hitches.

Note: This is a project-specific reproduction procedure, not a verified minimal reproduction project.

### Expected Result

World Partition streaming should update distance-field resources without causing severe gameplay freezes. Buffer growth, contraction and deferred retirement should avoid excessive GPU residency/synchronization stalls during normal traversal.

### Observed Result

Repeated traversal produces severe frame hitches associated with distance-field object-buffer replacement.

Captured examples include a 1,230 ms occlusion-query wait and a linked 1,154 ms residency-fence wait. These durations overlap and must not be added together.

On a separate return transition, a 231.12 ms frame included a 223.00 ms occlusion-query wait and a linked 171.94 ms residency-fence wait.

The buffer capacity oscillates between 1,664 MiB and 3,328 MiB, with temporary allocation overlap reaching 4,992 MiB.

The underlying Windows/NVIDIA synchronization cause remains unresolved. No VRAM-exhaustion warning was required to reproduce the stalls.

### Affects Versions

5.8

### Platform(s)

Windows

### Additional Notes

Additional technical findings:

- Engine: Unreal Engine 5.8.2, CL 56702186, DirectX 12.
- Hardware: NVIDIA RTX 5080, 16 GB VRAM.
- Distance-field object-buffer allocation uses power-of-two capacity sizing.
- The observed transition crosses the 8,388,608-slot threshold, causing replacement allocations.
- The issue occurs in standalone execution as well as PIE.
- The relevant allocation path appears unchanged between official UE 5.8.2 and 5.8.3 source.

We identified UE-387332 (CL 57782924), a UE 5.8.3 D3D12 synchronization correction involving deletion-queue backpressure and residency-set handling. Its applicability to this issue remains unproven.

We would appreciate guidance on whether Epic has a supported workaround or correction for distance-field object-buffer replacement/retirement stalls, and whether UE-387332 is relevant.

Detailed diagnostic measurements and source references are available if needed.

---

<div class="post-metadata">

**Author:** ![Tarbasch](https://d1ap1mz92jnks1.cloudfront.net/user_avatar/forums.unrealengine.com/tarbasch/32/1212834_2.png) [@Tarbasch](https://forums.unrealengine.com/u/Tarbasch)\
**Post date:** [October 19, 2026, 5:50pm UTC](https://forums.unrealengine.com/t/ue-5-8-2-severe-gpu-residency-stalls-during-world-partition-distance-field-buffer-resizing/2841748/2 "2026-10-19T17:50:11Z")

</div>

Anyone able to help on this?
