RTX SPARK BIG PROBLEMO

Age of Empires IV IV - Report a Bug
1

  • Build Number: 16.3.11308.0

  • Platform: Xbox

  • Operating System: Windows 11 (ARM!!!)

Issue: ON RTX SPARK ALL MY OTHER (ADMITTEDLY 3 OTHER) GAMES WORK BRILLIANTLY

BUT AOE4 HORRENDOUSLY CHOPPY FRAMES, UNPLAYABLE

TRIED EVERY GRAPHICS SETTING

AND EVERY ARM COMPATIBILITY SETTING

AND EVERY RESTART, DRIVER UPDATE TYPE THING

PLS FIX!!!

THANKS

MASSIVE DETAILS:
Hardware:

  • ASUS ProArt P16 H7607BA.

  • NVIDIA RTX Spark N1X: 20-core ARM64 CPU and 6144-core Blackwell RTX GPU.

  • 64 GB installed unified memory.

  • Native display: 2880×1800; Windows scaling: 200%. Different game resolutions have been tested.

  • NVIDIA driver 621.01, confirmed in the latest game log.

Measured frame-time evidence

Intel PresentMon 2.6.0 recorded 2,084 frames over approximately 29.8 seconds on 24 October 2026.

In the first 23 seconds, before opening the pause menu:

Measurement Result
Average presented frame rate 66.3 FPS
Median gap between frame submissions 11.25 ms
99th-percentile gap 77.97 ms
Frame gaps exceeding 50 ms 41
Frame gaps exceeding 100 ms 8
Worst gap between submissions 735.59 ms
Displayed-frame gap at that hitch 760.83 ms

The worst hitch occurred approximately 15.62 seconds into the recording. That frame reports:

  • 14.60 ms of active GPU work.

  • 706.19 ms of GPU idle time within its GPU interval.

  • 735.35 ms in the application-side interval before Present.

Across the first 23 seconds, active GPU work never exceeded 15.64 ms per frame. This suggests delayed application-side frame delivery rather than simply excessive graphics workload, but does not identify the responsible component.

Core 19 averaged approximately 87% CPU time, while aggregate CPU time averaged 17%. The three busiest cores reported approximately 4.0 GHz around the worst hitch. These are one-second Windows counters and cannot exclude brief throttling.

Separate CPU-trace findings

An earlier Windows Performance Recorder capture contains main/render thread stacks through:

xtajit64se.dll → RtlAcquireSRWLockExclusive → RtlpAcquireSRWLockExclusiveContended → RtlBackoff

It also contains emulator-originated NtFlushInstructionCache calls, with kernel work in KeSweepIcacheRange and HvcallpInitiateHypercall.

The main thread had 3,203 of 27,541 sampled instruction pointers at RtlBackoff. This is a sample share, not a measured percentage of frame delay.

Limitations: this older CPU trace predates driver 621.01 and was not simultaneous with the new frame capture. Emulator contention is a specific lead; anti-cheat/anti-tamper involvement and the cause of individual hitches are not established.

The game log initializes threading for 20/20 cores and registers main, simulation and render threads with worker-thread support.

Trigger Event:

Normal gameplay after loading a match. No reliable single triggering action has been isolated.

The game logged opening the pause menu approximately eight seconds after the worst hitch. A separate display-mode transition near the recording’s end produced another spike; that transition is excluded from the gameplay figures above.

Repro Steps:

  1. Launch the installed Microsoft Store/Xbox PC build on the Windows ARM64/N1X system described above.

  2. Load a match and play normally.

  3. Observe repeated uneven motion and hitches despite a reasonable average frame rate.

  4. Lower graphics/resolution or change window mode and retest. The choppiness persists on this machine.

  5. Record frame times. The supplied capture quantifies the problem; reproduction on other machines is not yet established.

Troubleshooting already attempted:

  • Lower graphics and resolution.

  • Window-mode changes.

  • Lower physics quality, followed by restarting.

  • Updating NVIDIA to 621.01.

  • Temporary performance-core affinity.

  • Safe emulation.

  • An isolated “Disable RWX page performance optimisation” test.

None produced a noticeable improvement. These were troubleshooting attempts, not controlled before/after benchmarks. Temporary affinity was restored, and no RelicCardinal.exe compatibility-layer overrides remained at the latest registry check.

Requested investigation:

Please investigate the x64 emulation lock/cache-maintenance paths on N1X and advise whether this is a known compatibility issue. What additional simultaneous CPU/frame trace would your team need to identify the stalls?

The attached evidence ZIP contains frame-time CSV, CPU counters, selected game-log lines, symbolized evidence from the older CPU trace, and analysis notes. The original full CPU ETL is available for a private support upload; please provide a suitable upload route.

Timestamp note: PresentMon’s date labels are one hour ahead of the recorder metadata, game log and CPU counters. The analysis accounts for that offset; original frame timestamps are preserved.