HOW-TO · POWER · CHECKED ON NPI 3.0.2.3
Power management, CUDA P2 state and threaded optimization
Power Management - Mode (0x1057EB71) only changes how fast the GPU clocks up. In the one published test we found, Prefer maximum performance gave no consistent FPS gain and raised whole-system idle power by 33 to 42 W. Set it per game and only if clocks visibly drop, never in Global. Turn CUDA - Force P2 State off only if your memory clock drops under CUDA apps. Leave Threaded Optimization on Auto. We measured none of this ourselves.
You need
- NPI
- 3.0.2.1 or newer
- Profile
- the game's own, not Global
- Time
- 3 minutes
Power Management Mode
↗ decides how quickly the GPU moves to a high clock state. NPI lists three values: Adaptive (0), Prefer maximum performance (1) and Optimal performance (5, the default). NVIDIA’s header also defines driver-controlled and consistent-performance values, which NPI’s common list does not show.
CODE · NPI value list
| Source | Finding |
|---|---|
| The FPS Review (2019, RTX 2080 SUPER and 2060 SUPER, 5 games) | No consistent FPS gain over Optimal or Adaptive. The fastest mode changed per game. Whole-system idle power rose from 61 to 103 W and from 58 to 91 W. |
| NVIDIA release notes 610.47 and 616.64 | An open issue says Prefer maximum performance may not be applied correctly. |
The one dataset is from 2019 hardware, and we could not add our own. REPORTED · The FPS Review OFFICIAL · NVIDIA release notes
Open the game's profile
Add the game if it has no profile.
Search for power
Type
power. Pick Power Management - Mode.Choose the value and apply
Use Prefer maximum performance only if you see clocks drop in that game. Click Apply Changes and restart the game.
CUDA - Force P2 State
↗ is on by default. Compute users report that with it on, CUDA work runs in the P2 state with memory clocked about 500 MHz lower, and that turning it off restores full memory clock. COMMUNITY · mining and compute forums BabelTechReviews found no significant average FPS change in 15 games on an RTX 3080 when it was off, with mixed frame-time stability. REPORTED · BabelTechReviews NVIDIA has no documentation we could find. Treat Off as a compute tweak, not a gaming boost, and do not use it as advice for overclocking or mining.
CUDA - Sysmem Fallback Policy
↗ has Default, Prefer no sysmem fallback and Prefer sysmem fallback. It decides whether a CUDA app may spill GPU memory into system memory. It matters for local AI use. CODE · NPI value list We have not tested it.
Threaded Optimization
↗ has Auto, On and Off. NVIDIA’s name for it, OGL_THREAD_CONTROL_ID, shows it applies to OpenGL games.
CODE · NVAPI header The only test we found is Phoronix on Linux from 2012 to 2014, with mixed results including a large drop in one game.
REPORTED · Phoronix There is no modern Windows test, so leave it on Auto.
Laptops and Optimus
| Setting | ID | What it holds |
|---|---|---|
| Enable Application For Optimus | ↗ | Raw number values, default 0x10. |
| Optimus - Compatibility Flags | ↗ | Integrated GPU 0, Enable discrete GPU 1, User editable 2, Auto-select 0x10, Override 0x80000000. |
| Shim Rendering Mode Options | ↗ | Flags such as disabling async present. |
| Multi-Display / Mixed-GPU Acceleration | ↗ | Single, Compatibility or Multi display performance mode. |
None of these are in NPI’s default list except the last. We have no laptop, so we do not claim which one forces the NVIDIA GPU on current drivers. Windows Graphics settings is the supported route, and the requirements page mentions laptops. For battery, the background frame limit helps when the game is not in the foreground.
Undo
Set the row back to its default, or use the reset action on the row. Undo and reset has the details.
Questions people ask
Should I set power management mode to Prefer maximum performance?
Only per game and only if the clocks visibly drop. The published test found no consistent gain and about 33 to 42 W more whole-system idle power, and NVIDIA lists an open bug that the mode may not be applied correctly.
Does it increase FPS?
Not reliably. In the test we found, the fastest mode changed from game to game.
What does CUDA Force P2 State do?
With it on, which is the default, CUDA work runs in a lower performance state, and users report memory clocked about 500 MHz lower. Turning it off restores full memory clock. A 15-game test found no significant FPS change in games.
Should Threaded Optimization be on or off?
Leave it on Auto. It applies to OpenGL only, and we found no modern Windows test.
How do I force a laptop to use the NVIDIA GPU?
Windows Graphics settings is the usual way. NPI also has Optimus settings, which we could not test without a laptop.
Sources
- NVIDIA 616.64 release notes nvidia.com · accessed 8 Nov 2026 · official
- NVIDIA NVAPI header github.com · accessed 8 Nov 2026 · official
- The FPS Review thefpsreview.com · accessed 8 Nov 2026 · reported
- BabelTechReviews babeltechreviews.com · accessed 8 Nov 2026 · reported