[BUG REPORT] DX12 PSO Cache Corruption / Severe FPS drop facing map center

System Specifications:

OS: Windows 11 Home

CPU: Intel Core i5-12600KF

GPU: NVIDIA GeForce RTX 5060 Ti (Tested on both the latest driver and downgraded stable 596.49)

RAM: 32 GB

Storage: High-speed NVMe M.2 SSD (both OS and Game)

The Issue:

Following the Season 29 Split 2 update, I am experiencing a crippling, deterministic framerate drop (from a stable 150 FPS cap down to ~50–55 FPS) strictly when the camera is oriented toward the center of any map (tested extensively in the Firing Range and Ranked BR). Orienting the camera toward the sky or the map perimeter instantly restores a rock-solid 150 FPS.

Through rigorous testing, I have isolated the exact trigger. This is not a hardware bottleneck; it is an engine-level Pipeline State Object (PSO) cache serialization bug.

The 100% Reproducible Trigger

The game runs flawlessly, but only once per GPU environment reset.

The "Cold Boot" (Working State): If I perform a clean, fresh installation of the NVIDIA Graphics Driver OR immediately after Steam downloads a game patch -> I launch Apex. The engine compiles shaders into RAM/VRAM live. I enter the Firing Range. The game runs perfectly. Facing the center of the map yields an unwavering 150 FPS. No stutters.

The "Warm Boot" (Broken State): I launch the game for the second time. The engine bypasses live RAM compilation and reads the saved cache from the disk. I enter the Firing Range. The FPS drop to 55 facing the center returns permanently.

From this point on, restarting the PC, restarting the game, manually deleting %localappdata%\NVIDIA\DXCache, deleting pso_cache.bin, or disabling Steam Pre-Caching does nothing. The engine reads a permanently corrupted cache file.

If I reinstall the NVIDIA driver right now (forcing a complete clearing of the locked kernel cache) and launch the game, it runs at 150 FPS in the central area again—but seemingly only for one single match.

It appears that the Split 2 DX12 compiler generates healthy shader code during real-time runtime (RAM). However, when serializing and writing the PSO cache to the disk, it generates malformed instructions specifically for high-entity-density geometry (map center) on this specific GPU architecture/bus width. When the engine attempts to read this specific serialized package from the storage on subsequent launches, it triggers a silent pipeline sync stall (a thread deadlock). The main rendering thread gets put into a wait-state trying to resolve the malformed PSO, which artificially throttles the frame output down to ~55 FPS without generating any corresponding spike in CPU or GPU utilization (both components sit underutilized waiting for the execution instruction to clear.

Tested extensively on the latest available NVIDIA driver version initially, as well as rolled-back stable versions (e.g., 596.49) via clean DDU installations. The bug behaves completely identically across all driver versions.

Confirmed the game resides on an NVMe M.2 SSD

hermals are pristine (GPU sits at ~55°C during the 50 FPS drops; no thermal throttling).

Tested unlocked FPS, 150 cap, and 99 cap – the math scales, but the proportional drop facing the center remains identical.

Please keep in mind that I am just a player and not a rendering engineer, so my technical hypothesis regarding the exact internal trigger might be completely off the mark. My only goal was to provide your team with the cleanest, most deterministic reproduction steps possible to help narrow this down.

I would be immensely grateful for a brief acknowledgment that this bug report has reached the appropriate DX12 team.