HRESULT 0x887A0034 (DXGI_ERROR_CACHE_FULL): meaning and diagnosis

Status / error code:

Category:

Constant name:

Source:

Facility:

Operating System:

Description:

This entry would cause the cache to exceed its quota. On a load operation, this may indicate exceeding the maximum in-memory size.

Previous	Next
DXGI_ERROR_CACHE_CORRUPT	DXGI_ERROR_CACHE_HASH_COLLISION

DXGI_ERROR_CACHE_FULL

Why this is not a generic GPU failure: cache full

The useful interpretation of DXGI_ERROR_CACHE_FULL (0x887A0034) starts at the dxgi error boundary: storing or loading the requested entry would exceed the cache session maximum size or memory bound. In practical terms, inspect configured shader-cache quota before changing application-wide graphics settings.

DXGI_ERROR_CACHE_FULL: Do not erase all shader caches or recreate every graphics object before preserving the first failing call. For DXGI_ERROR_CACHE_FULL, Cache, policy, remote-session and device-lifetime failures have different recovery rules. For DXGI_ERROR_CACHE_FULL, the built-in one-line message identifies the immediate result; diagnosis also needs the producing API, object ownership and the display generation current at the time.

The smallest evidence set

Capture item	Why it matters for `DXGI_ERROR_CACHE_FULL`
Object identity	configured shader-cache quota
Rejected boundary	storing or loading the requested entry would exceed the cache session maximum size or memory bound
Decisive capture	cache-session limits, current entry sizes, key, value length, eviction behavior and process memory budget
Current graph state	DXGI factory, adapter, device, resource or cache session
Supporting trace	DXGI debug messages, adapter LUID, driver version, device-removal reason where applicable and the exact API parameters

For DXGI_ERROR_CACHE_FULL, preserve the unsigned HRESULT, symbolic name, first failing API and timestamp in one record. If cleanup later fails too, keep that secondary result separately from DXGI_ERROR_CACHE_FULL.

A/B test for the exact condition

Freeze the DXGI_ERROR_CACHE_FULL generation. Record cache-session limits, current entry sizes, key, value length, eviction behavior and process memory budget. Before diagnosing DXGI_ERROR_CACHE_FULL, do not resize, hot-plug, recreate or release the object under examination.
Run the narrow DXGI_ERROR_CACHE_FULL comparison. store progressively larger disposable entries until the configured quota is reached. During the DXGI_ERROR_CACHE_FULL comparison, keep every driver, monitor, resource format and unrelated policy unchanged.
Observe the layer after DXGI_ERROR_CACHE_FULL. For DXGI_ERROR_CACHE_FULL, if the exact check passes, record the next HRESULT or visible outcome. A different downstream result shows that the DXGI_ERROR_CACHE_FULL boundary was crossed.
Repeat DXGI_ERROR_CACHE_FULL through a lifecycle transition. For DXGI_ERROR_CACHE_FULL, exercise one relevant resize, mode switch, device recreation, hot-plug or session change and confirm that this result does not reuse stale handles.

The comparison with DXGI_ERROR_HW_PROTECTION_OUTOFMEMORY is especially useful: DXGI_ERROR_HW_PROTECTION_OUTOFMEMORY concerns protected GPU memory rather than cache quota. For DXGI_ERROR_CACHE_FULL, recording both names prevents a broad “graphics error” label from merging distinct ownership, capability and lifetime problems.

Safe remediation

Bound entry size, evict application-owned data or create a cache session with an appropriate documented maximum. Repair of DXGI_ERROR_CACHE_FULL is complete only when the original call succeeds or returns its documented nonfatal status and the same lifetime test remains correct after a second display transition.

The DXGI_ERROR_CACHE_FULL trace identifies one producing API and one current graphics object, rather than only the final UI symptom.
The passing run changes exactly the condition described for DXGI_ERROR_CACHE_FULL; unrelated adapter, monitor and application state remains unchanged.
For DXGI_ERROR_CACHE_FULL, debug-layer, ETW or driver diagnostics no longer report the rejected configured shader-cache quota contract during the same scenario.
The application handles recurrence of DXGI_ERROR_CACHE_FULL without an unbounded retry loop, leaked resource, duplicate composition target or stale topology handle.

Implementation notes for `DXGI_ERROR_CACHE_FULL`

For DXGI_ERROR_CACHE_FULL, log DXGI debug messages, adapter LUID, driver version, device-removal reason where applicable and the exact API parameters. When the DXGI_ERROR_CACHE_FULL path returns a count, size, status flag or replacement object, retain it even on a nonfatal result because it can direct the next call.

A production fallback for DXGI_ERROR_CACHE_FULL should be explicit: pause rendering, re-enumerate, rebuild one cache entry, recreate a device, or decline a protected path only when this layer calls for that action. Reinstalling every display component, deleting all caches or forcing a resolution change is not an evidence-based fix for DXGI_ERROR_CACHE_FULL.

Technical references

Microsoft: DXGI error codes — background for the configured shader-cache quota boundary involved in DXGI_ERROR_CACHE_FULL.
Microsoft: DXGI overview — background for the configured shader-cache quota boundary involved in DXGI_ERROR_CACHE_FULL.
Microsoft: Direct3D 12 shader cache StoreValue — background for the configured shader-cache quota boundary involved in DXGI_ERROR_CACHE_FULL.
Microsoft: Handle device-removed scenarios — background for the configured shader-cache quota boundary involved in DXGI_ERROR_CACHE_FULL.

Looking for a different code? Search another status or error code.