Hello everyone,

I spent some time investigating a VR performance issue in Le Mans Ultimate where enabling shadows causes a significant FPS drop.

This is not a request for basic VR settings advice. I tried to isolate the exact point where the performance loss happens.

The issue appears to be related to a CPU rendering workload triggered by specific illuminated scene elements in VR. Dynamic shadows significantly amplify the effect but do not appear to be the sole cause.

When shadows are enabled:

• FPS drops from stable 90 FPS to around 64 FPS
• CPU frame time increases by approximately +7 ms
• GPU frame time remains unchanged
• The GPU is not saturated. The bottleneck happens before the GPU rendering stage.

Test system

• CPU: AMD Ryzen 7 7700 (overclocked)
• GPU: NVIDIA RTX 5080 (overclocked)
• RAM: DDR5 6000 CL30
• VR headset: Pimax Crystal Super / EVO
• Resolution: 3888 × 3696 pixels per eye
• API: DirectX 11
• VR driver tested: SBoys3 Native SteamVR Driver
• OpenXR Toolkit: tested ON, tested OFF, completely removed (No difference).

Main observation

Shadows OFF

• Result: 90 FPS stable
• CPU Busy: ~8 ms
• GPU Busy: 5–6 ms

Shadows ON

• Result: ~64 FPS
• CPU Busy: 14–16 ms
• GPU Busy: 5–6 ms

The GPU workload is almost identical. The additional cost appears entirely on the CPU side.

A 90 Hz VR headset requires 1000 / 90 = 11.1 ms per frame.

With shadows enabled, the CPU frame time hits 14–16 ms. Therefore, the engine cannot maintain 90 Hz.

Tests performed to exclude other causes

Resolution / FOV

Tested: native Pimax resolution, reduced OpenXR resolution, and Crop2FOV (Up 85, Down 80, Left 95, Right 100).

• Result: No improvement. If the problem was caused by pixel rendering load, GPU Busy should have decreased. It did not.

OpenXR Toolkit

Test matrix:

Shadows	OpenXR Toolkit	CPU Busy	GPU Busy	FPS
OFF	OFF	~8 ms	5-6 ms	90
ON	OFF	14-16 ms	5-6 ms	~64
ON	ON	14-16 ms	5-6 ms	~64

• Result: OpenXR Toolkit is not responsible.

Graphics settings

Tested: Track Detail = 2, Special FX = 1, Road Reflections reduced, Environment Reflections reduced, SSAO disabled, frame synchronization options modified.

• Result: No measurable improvement.

CPU scheduling / threads

Tested: Process Lasso, CPU affinity changes, physical cores only, SMT disabled, CPU parking disabled.

• Result: No improvement. CPU global usage remains below 50%. However, the main CPU thread sits at 80–90% while other cores are below 50%. This looks like a main-thread limitation rather than insufficient overall CPU power.

Different circuits

The issue was reproduced on Bahrain WEC night, Spa-Francorchamps, and other environments.

• Result: It does not appear to be related to one specific track, object, or texture. The common factor is strictly VR + dynamic shadows.

Current hypothesis

The behavior strongly suggests an excessive CPU cost during the preparation of shadow rendering in VR.

Possible areas to investigate:

• Shadow pass generation
• Shadow caster culling
• Draw call generation
• DirectX 11 command submission
• Stereo VR shadow rendering

A possible problematic scenario could be:

• Main rendering: optimized stereo rendering
• Shadow rendering: separate processing per eye (which would massively increase CPU preparation cost).

Information for Studio 397

The most useful profiling points would probably be:

• CPU time spent preparing shadow passes in VR
• Comparison of draw calls: monitor rendering vs VR rendering
• Number of shadow casters processed per eye
• Whether stereo instancing/multi-view techniques are applied to shadow passes
• CPU cost of dynamic lights and headlights during night sessions

The goal is not to remove shadows or reduce visual quality. The objective is to identify why a system capable of 300+ FPS on screen, 90 FPS VR without shadows, and boasting an RTX 5080 GPU with significant headroom loses performance only when VR shadows are enabled.

Hopefully, this can help reproduce and investigate the issue internally.

(Note: I have already submitted a formal ticket to S397 with a full 34-section technical breakdown. For those interested in the raw data and deeper methodology, you can read my full PDF below.

Are any other VR users noticing this exact same behavior (GPU chilling while CPU frame times explode) during night sessions? Let me know what PresentMon is telling you!

📄 Read the full report here:
https://drive.google.com/file/d/1BWxXKSjGq93_DtZTQ36B438ZZ2odkQT5/view?usp=sharing