• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

DLSS 4 Multi Frame Generation 3x/4x on RTX 4000 – Software Lock, Not Hardware

FirstEver

n00b
Joined
Mar 25, 2026
Messages
2
I've been investigating whether DLSS 4 Multi Frame Generation 3x/4x is genuinely a hardware limitation on RTX 4000 series
— or simply a software-enforced restriction. The data points strongly in one direction.



Benchmark Results RTX 4070 Ti

Cyberpunk 2077 – Ray Tracing: Overdrive, 4K

Code:
| Mode  | Avg FPS | Min FPS | Max FPS |
|-------|---------|---------|---------|
| FG 2x |   75.83 |   69.78 |   82.18 |
| FG 3x |  124.49 |  113.24 |  136.96 |
| FG 4x |  164.68 |  149.75 |  180.54 |

3DMark NVIDIA DLSS Feature Test

Code:
| DLSS FG Mode        | DLSS Off   | DLSS On    | Multiplier |
|---------------------|------------|------------|------------|
| Frame Generation 2x | 32.27 FPS  | 125.16 FPS |    3.9x    |
| Frame Generation 3x | 32.16 FPS  | 224.02 FPS |    7.0x    |
| Frame Generation 4x | 32.20 FPS  | 291.56 FPS |    9.1x    |

Official 3DMark results (verified, public):
https://www.3dmark.com/compare/nd/580395/nd/580396/nd/580397



The Black Overlay Problem

MFG 3x/4x can be enabled on RTX 4000 via NVIDIA Profile Inspector, bypassing the lock in NVIDIA App.
When enabled, something unexpected happens: a full-screen black overlay is rendered on top of the output.

But here's the key — everything underneath is working correctly:
  • NVIDIA App reports the expected FPS scaling
  • The game remains fully interactive
  • Frame pacing is consistent
  • Audio continues normally
  • 3DMark records every single generated frame
    — including frame-time graphs that are identical in shape to FG 2x, just scaled proportionally
1774461358829.png


NVIDIA didn't throw an error.
They didn't disable the feature.

They added a curtain.

MFG-RTX4000.png




Ghidra Analysis – nvlddmkm.sys

I performed static analysis of nvlddmkm.sys (driver 32.0.15.9579) using Ghidra. Key finding:
  • Identified string reference: ShowEvalModeOverlay
  • Associated code path appears tied to runtime hardware/feature validation
  • A single instruction change at offset 0x14C8A58: JNZ → JMP (0x75 → 0xEB) appears to influence whether the overlay is triggered
This is not yet definitive proof of the exact mechanism, but observed behavior is fully consistent with this hypothesis.



The Blocker – Kernel Mode Code Integrity

Since nvlddmkm.sys is a kernel-mode driver, any modification hits Windows 11 KMCI enforcement:
  • Test signing mode is insufficient
  • Self-signed certificates don't work
  • Secure Boot + modern Windows 11 = Code 52 signing error
This remains the current blocker for end-to-end validation of the patch.



Full write-up, screenshots, Ghidra analysis and .3dmark-result files (openable locally in 3DMark to verify frame-time graphs yourself):

NVIDIA GeForce RTX 4000 Series - DLSS 4 Frame Generation 3x/4x Unlock

Looking forward to discussion — especially from anyone who's done similar driver analysis or has access to other RTX 4000 GPUs for cross-validation.
 
Last edited:
I’ve now completed an Nsight Systems analysis of the Cyberpunk 2077 built-in benchmark using three captures taken under matching conditions:
  • FG 2x
  • FG 3x
  • FG 4x
The captures were analyzed in NVIDIA Nsight Systems 2026.2.1.

Preliminary result: the data does not look like simple FPS counter inflation or trivial frame replay.
The present pattern scales in structured groups consistent with the selected mode (2 / 3 / 4),
which suggests that additional frames are actually being presented at the pipeline level.

This still does not automatically prove full image-quality equivalence with native RTX 5000-series MFG,
but it is much stronger evidence than benchmark FPS numbers alone. I’m attaching a few screenshots here,
and a full write-up will be published soon in my GitHub repository together with the supporting materials.



Present events in a stable benchmark segment

kwMGM5NjExOWE1JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCJ9.png

Inter-present interval distribution in the 20-40 s analysis window

A3MjM2M2FhNGU4JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCJ9.png

Output present rate vs inferred base render cadence

YzNmE0MzdlMjk1JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCJ9.png

Packet size distribution inferred from >12 ms inter-packet gaps

BmM2U1NjA1MzZjJlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCJ9.png

GitHub update:
  • a full technical summary
  • figures from the Nsight analysis
  • and the original capture files used for the investigation
https://github.com/FirstEverTech/RTX4000-MFG-Unlock/releases/tag/v1.0-nsight-notes
 
  • Like
Reactions: MaZa
like this
Back
Top