[RTX 50] DLSS Frame Generation rapid toggle causes GPU power state lock (~170W) — workaround found via disabling Force P2 State

Summary

On RTX 5070 Ti with driver 596.49, rapidly toggling DLSS 4.5 Multi Frame Generation (MFG 6x with Preset B frame generation model) causes the GPU to get stuck at a reduced power draw of approximately 170W, well below the expected ~280W under load. The rapid toggling happens automatically when opening/closing in-game submenus in certain titles, which pause and resume MFG. Notably, GPU core clocks remain at normal boost frequencies — only the power budget is capped. The state persists until the game is restarted or the MFG state is manually cycled.

System Specifications

  • GPU: NVIDIA GeForce RTX 5070 Ti
  • CPU: AMD Ryzen 9 9900X
  • Motherboard: Gigabyte (AM5)
  • RAM: 32GB DDR5-6000
  • PSU: Seasonic Focus GX-850 (ATX 3.0, 850W)
  • OS: Windows 11 25H2
  • Driver: 596.49 WHQL (Game Ready)
  • Display: 3840x2160 @ 170Hz

DLSS Configuration

  • DLSS Version: 4.5
  • Multi Frame Generation: 6x mode
  • Frame Generation Model: Preset B (new enhanced FG model)
  • Enabled via NVIDIA App DLSS Override

Affected Games (Confirmed)

  • Wuthering Waves (Steam version)
  • Stellar Blade (Steam version)

Both titles, when overridden to use DLSS 4.5 MFG 6x with Preset B, pause MFG when opening certain in-game menus and resume MFG when closing them. This automatic pause/resume cycle is what triggers the bug.

Reproduction Steps

  1. Enable DLSS 4.5 MFG 6x with Preset B for Wuthering Waves via NVIDIA App DLSS Override.
  2. Launch the game and begin gameplay. Confirm GPU is running at expected power draw (~280W on RTX 5070 Ti).
  3. Open the secondary menu (which auto-disables MFG).
  4. Immediately close the secondary menu (which auto-re-enables MFG).
  5. Repeat steps 3-4 several times in quick succession.
  6. Observe GPU power draw via HWiNFO64, MSI Afterburner, or NVIDIA overlay.

Expected Behavior

GPU should return to full power budget (~280W) after MFG re-enables.

Actual Behavior

GPU power draw locks at approximately 170W. GPU core clocks remain at normal boost frequencies (no clock throttling observed). The reduced power cap persists indefinitely until the game is restarted. Temperature is normal (well below thermal limits), confirming this is not thermal throttling. Power supply has ample headroom (850W ATX 3.0 PSU for a ~300W GPU + ~170W CPU system).

This is particularly impactful because MFG 6x with Preset B is the highest-performance DLSS 4.5 configuration available, and the resulting ~110W power deficit translates to a significant performance loss precisely when maximum frame generation throughput is desired.

Workaround

Setting “CUDA - Force P2 State” to Off via NVIDIA Profile Inspector (latest version) appears to resolve the issue. After applying this change, the rapid MFG toggle no longer triggers the power state lock during testing.

This suggests the root cause involves a race condition triggered by DLSS 4.5 MFG workloads being forced into P2 state and back. Since DLSS MFG (especially at 6x with Preset B) is a heavy CUDA workload utilizing Tensor Cores via CUDA, the forced P2 transition appears to conflict with normal power management when MFG toggles rapidly.

Request

Please investigate the power management / TGP arbitration logic for DLSS 4.5 Multi Frame Generation workloads on RTX 50 series GPUs, particularly at higher multipliers (6x) using the Preset B frame generation model. Consider whether the “Force P2 State” default behavior should be revised for the Blackwell architecture and modern DLSS 4.5 workloads.

Thank you.

1 Like