A patched version of Nvidia’s DLSS 5 neural rendering engine ran on RTX 4000 “Ada Lovelace” graphics cards before Nvidia had even shipped the feature officially. Tom’s Hardware reported on August 28, 2026 that a modder re-encoded a leaked DLSS 5 library, nvngx_dlssnr.dll, and rewrote its CUDA instructions so RTX 4080 and RTX 4090 owners could run it in Remedy’s Control. German outlet IT BOLTWISE and Russian gaming site Shazoo independently confirmed the mod works, though at a steep cost: an RTX 4090 running the patched neural renderer dropped from roughly 135 FPS to about 82 FPS in the same test scene, a 39% hit. Nvidia then launched DLSS 5 for real on September 3, 2026, as a Blackwell-only feature, before reversing course a day later and confirming RTX 40-series cards would get official support too.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What actually happened with the DLSS 5 RTX 4000 patch
The story starts with a leak. A copy of Nvidia’s DLSS 5 neural rendering DLL, tied to Remedy’s Control and its early Blackwell-exclusive build, surfaced online and quickly made its way to modding communities. Tom’s Hardware verified that within roughly a day of that leak, modders had gone further than simply swapping a file into a game folder, which is how most DLSS mods have worked historically. Instead, they opened the compiled DLL and patched the CUDA instructions inside it that check for a Blackwell device ID before the neural renderer will initialize.
That distinction matters. Dropping a newer DLL into an older game’s install folder, the classic “DLSS swap,” only works when the underlying code already runs on the installed hardware and the game is simply pointed at a newer file. DLSS 5’s neural rendering path is different: Nvidia built it around instructions that assume Blackwell’s tensor cores are present. To get it running on Ada Lovelace silicon, the modder had to rewrite those checks and adapt the instruction set so a GPU that was never meant to execute this code could still process it, according to IT BOLTWISE’s account of the mod.
Shazoo’s report adds a concrete data point: the patched build was tested and confirmed running on a GeForce RTX 4080, a GPU that, in Nvidia’s own marketing, sits firmly outside DLSS 5’s supported hardware list. The outlet noted that the file, in its original leaked form, refused to initialize on anything but Blackwell chips before the patch was applied.
The performance cost: a 39% FPS drop on RTX 4090
Getting DLSS 5 to boot on an RTX 4000 card is one thing. Getting it to run well is another matter entirely, and the early benchmark numbers make that gap obvious. IT BOLTWISE’s test in Control found an RTX 4090 delivering approximately 135 FPS at baseline, without the patched neural renderer engaged. Switching on the modded DLSS 5 path dropped that figure to roughly 82 FPS, a decline of close to 40%. Russian outlet 3DNews independently reported the same 135-to-82 FPS numbers and put the decline at roughly 39%, corroborating IT BOLTWISE’s benchmark rather than contradicting it.
For context, that is a far larger performance tax than gamers have historically paid to unlock restricted DLSS features on older hardware. When modders enabled DLSS 3 Frame Generation on RTX 20 and RTX 30 series cards through community tools, the overhead was real but usually secondary, mostly extra CPU and latency cost layered on top of frame generation’s inherent limitations. What’s happening with the DLSS 5 patch looks structurally different: Ada Lovelace’s fourth-generation tensor cores appear to be straining to execute FP8-heavy neural rendering instructions that Blackwell’s fifth-generation tensor cores were purpose-built to handle.
That distinction is the core of why this story matters beyond a curiosity for tinkerers. It is not simply that Nvidia locked a software toggle behind a driver check. The performance data suggests Ada Lovelace genuinely lacks the throughput to run DLSS 5’s full neural model at a usable frame rate, even once the artificial hardware gate is removed.
Why Nvidia restricts DLSS 5 to RTX 50 Blackwell cards
Nvidia’s official position shifted within 24 hours of DLSS 5’s actual launch, though the groundwork was laid months earlier: when Nvidia previewed DLSS 5 in March 2026, it said the feature would arrive in a launch wave of more than 15 games, including AION 2 and Starfield. At its September 3, 2026 debut, DLSS 5 neural rendering was confirmed exclusively for the RTX 5060 through RTX 5090, spanning the full Blackwell stack, with compatibility trackers dlss5.org and dlss5.app both listing RTX 40-series cards as unsupported. On September 4, 2026, Nvidia told French outlet Frandroid that DLSS 5 will officially come to RTX 40-series cards as well, after the company finishes further optimization work on RTX 50 hardware. As of September 2026, Nvidia has still given no official date for that RTX 40 rollout, nor said whether an official implementation would avoid the roughly 39% FPS penalty seen in the community-patched version.
The rationale Nvidia has given, echoed in coverage from outlets tracking the DLSS 5 rollout, is that the neural renderer’s pretrained model runs on FP8 precision, and Blackwell’s tensor cores were engineered with substantially higher FP8 throughput than Ada Lovelace’s. RTX 40-series GPUs can still run DLSS 4.5, which Nvidia announced in January 2026 with support across more than 250 DLSS-enabled games and a new 6X Dynamic Multi Frame Generation mode the company says can push 4K path-traced gaming past 240 FPS; that Super Resolution and Ray Reconstruction feature set remains the only official neural-upscaling tier available on Ada Lovelace until Nvidia’s promised RTX 40 version of DLSS 5 actually ships.
This isn’t the company’s first time drawing a hardware line around a DLSS generation. Nvidia limited DLSS 3 Frame Generation to RTX 40-series cards at launch, citing the Optical Flow Accelerator hardware unique to Ada Lovelace, only for modders to later prove that frame generation could technically function on RTX 20 and RTX 30 cards through software workarounds, with tradeoffs in latency and consistency. The DLSS 5 situation on RTX 4000 cards looks like a repeat of that pattern, except this time the performance penalty for bypassing the gate is far steeper.
RTX 4000 vs RTX 50: the tensor core gap driving this story
| Spec / Feature | RTX 4000 “Ada Lovelace” | RTX 50 “Blackwell” |
|---|---|---|
| Tensor core generation | 4th-generation | 5th-generation |
| Precision focus for neural rendering | FP16 / TF32, limited FP8 | High-throughput FP8 |
| Official DLSS support ceiling | DLSS 4.5 today; DLSS 5 confirmed to follow after RTX 50 optimization | DLSS 5 (neural rendering, Multi-Frame Generation, SR), live since Sept 3, 2026 |
| DLSS 5 neural rendering status | Unofficial mod only (~39% FPS penalty); official support confirmed, no ship date yet | Officially confirmed and shipping on RTX 5060 through RTX 5090 |
| Frame generation ceiling | Up to 4x (DLSS 3/4) | Up to 6x Multi-Frame Generation on select models |
| RTX 4090 baseline vs modded DLSS 5 (Control, per IT BOLTWISE) | ~135 FPS baseline / ~82 FPS with patched DLSS 5 | Not directly comparable; Blackwell runs DLSS 5 natively |
How the DLL patch actually works
The technical method behind this mod is worth breaking down for readers who work with graphics APIs or reverse engineering. DLSS integrations typically expose themselves to games through an Nvidia-provided library, and modders have swapped these DLLs for years to force newer DLSS versions into older games or unlock features tied to specific hardware generations. What sets the DLSS 5 case apart is that a straightforward DLL swap wasn’t enough.
According to the reporting from IT BOLTWISE, the modder had to identify and rewrite CUDA instructions inside nvngx_dlssnr.dll that check for Blackwell-specific device identifiers before the neural rendering pipeline will initialize. Once those checks were patched out and the instruction paths adjusted to route through Ada Lovelace’s available tensor core operations, the pretrained FP8 neural model inside the DLL could execute, just far more slowly than its design target.
// Conceptual outline of the patch workflow described in reporting
// 1. Extract leaked nvngx_dlssnr.dll from the Control build
// 2. Locate hardware-ID gate checking for Blackwell GPU signature
// 3. Patch/NOP the gate check in the compiled binary
// 4. Remap FP8 tensor instructions to Ada-compatible execution paths
// 5. Replace the game's stock DLL with the patched version
// Result: neural rendering initializes on RTX 4080/4090, but at
// roughly 40% lower frame rate than native RTX 50 executionTom’s Hardware’s coverage noted that modders were already testing the patched DLL in titles beyond Control within a day of the leak, suggesting the technique is portable across DLSS 5 integrations rather than a one-off fix specific to a single game engine.
Historical precedent: DLSS 3 Frame Generation on RTX 20 and RTX 30
This is not the first time the modding community has cracked open an Nvidia feature gate. When DLSS 3 Frame Generation launched exclusively for RTX 40-series cards, Nvidia’s stated justification was the Optical Flow Accelerator, a hardware block Nvidia said was necessary for smooth frame interpolation. Modders responded with tools that redirected frame generation calls to run on RTX 20 and RTX 30 cards anyway, and separate community projects built wrapper DLLs that translated Nvidia’s frame generation calls into AMD’s FSR 3 pipeline, sidestepping the hardware check entirely.
Those older workarounds generally produced usable, if imperfect, results. Frame generation on unsupported RTX 30 cards came with added input latency and occasional artifacting, but it ran at frame rates gamers could live with. The DLSS 5 patch on RTX 4000 cards tells a different story: a roughly 40% FPS penalty is not a minor compromise, it is closer to running a demanding path tracing setting than a frame-rate enhancement feature. That gap suggests Nvidia’s FP8 tensor throughput argument for restricting DLSS 5 to Blackwell has real technical teeth, even if the hardware gate itself can be bypassed.
Market and competitive impact
The timing of this leak carries commercial weight. Nvidia’s DLSS 5 rollout is one of the headline features tied to justifying the RTX 50 series’ pricing, and reports earlier in 2026 already pointed to RTX list prices climbing sharply this year amid tight GPU and memory supply. A working, if flawed, DLSS 5 patch for RTX 4000 cards undercuts part of Nvidia’s pitch that neural rendering is a reason on its own to upgrade to Blackwell, since it demonstrates the feature is not fundamentally impossible on older silicon, just inefficient. Nvidia’s own September 4, 2026 confirmation that official RTX 40 support is coming reinforces that same point, this time from the company itself rather than the modding scene.
That said, the roughly 40% FPS penalty measured by IT BOLTWISE likely limits how much this mod actually threatens RTX 50 sales. A gamer weighing whether to buy an RTX 5080 instead of keeping an RTX 4090 is unlikely to settle for a DLSS 5 experience that costs that much performance, especially when DLSS 4.5’s Ray Reconstruction and Super Resolution already run natively and efficiently on Ada Lovelace hardware. The mod is more proof-of-concept than a genuine substitute for a Blackwell purchase, at least in its current form.
For competitors, the episode adds a small data point to the broader graphics card conversation. AMD’s FSR 4 and Sony’s PSSR 2.0 have both been positioned as cross-generation alternatives that don’t require a hardware refresh to access the newest upscaling improvements, a contrast to Nvidia’s generation-gated rollout strategy for DLSS 5. Whether that contrast moves purchasing decisions depends heavily on how quickly Nvidia’s promised RTX 40 support actually materializes, and whether the company treats the leaked DLL as a security matter or lets the modding scene continue iterating on it.
What this means for RTX 4000 owners right now
Anyone holding an RTX 4080 or RTX 4090 today has a genuine, if rough-edged, option to try DLSS 5’s neural rendering before Nvidia’s own optimized version reaches RTX 40 cards. That said, several practical caveats apply based on what has been reported so far:
- The patch currently depends on a leaked DLL that was never intended for public distribution, which raises both legal and stability questions for anyone applying it.
- Confirmed testing has centered on Control and the RTX 4080 and RTX 4090; broader game compatibility has not been independently verified across a wide title list.
- The nearly 40% FPS penalty measured on an RTX 4090 makes this impractical for anyone chasing higher frame rates rather than simply testing the feature out of curiosity.
- Nvidia has not commented on the leaked DLL or the community patch specifically, but on September 4, 2026 the company confirmed that official DLSS 5 support is coming to RTX 40-series cards after further optimization on RTX 50, reversing its Blackwell-only stance from the September 3 launch just a day earlier.
DLSS generational restrictions: a pattern comparison
| DLSS Generation | Officially Restricted To | Modded to Run On | Reported Result |
|---|---|---|---|
| DLSS 3 Frame Generation | RTX 40 series | RTX 20 / RTX 30 series | Usable performance, added latency |
| DLSS 4.5 Ray Reconstruction | RTX 20 series and newer | N/A (broad official support) | Native support, no mod required |
| DLSS 5 Neural Rendering | RTX 50 “Blackwell” series at launch; RTX 40 confirmed to follow | RTX 4080 / RTX 4090 (Ada Lovelace) | Functional but ~39% FPS penalty |
Why Ada Lovelace tensor cores struggle with DLSS 5’s FP8 model
The technical explanation for the performance gap comes down to how Nvidia optimized DLSS 5’s neural model. The renderer relies on a pretrained FP8 model built around Nvidia’s second-generation transformer architecture for Super Resolution, which the company detailed when it previewed DLSS 5 in March 2026, and Blackwell’s fifth-generation tensor cores were built specifically to push high volumes of FP8 operations with minimal overhead. Ada Lovelace’s fourth-generation tensor cores can technically process similar workloads, but without the same dedicated throughput, meaning the same neural network runs, just through hardware that has to work harder and slower to complete each pass.
That’s a meaningfully different situation from a simple artificial software lock. If DLSS 5 were purely gated by a driver check with no real hardware dependency, a patched DLL would likely perform close to native speed once the gate was removed. Instead, the roughly 40% FPS drop measured on an RTX 4090 strongly suggests genuine architectural limits, not just a marketing decision to push Blackwell sales. That distinction should temper expectations for how far the modding community, or even Nvidia’s own engineers, can push an RTX 40 implementation once optimization work is done.
What comes next: predictions for DLSS 5 modding and Nvidia’s response
Based on how Nvidia has handled similar situations with DLSS 3 Frame Generation, and what has happened since DLSS 5’s launch, here is where things stand and what looks likely in the weeks ahead:
- Nvidia moved faster than expected on RTX 40 support. Rather than staying quiet, the company confirmed just one day after DLSS 5’s September 3, 2026 launch that RTX 40-series cards will get official support too, once further optimization work on RTX 50 wraps up. Nvidia still has not addressed the leaked DLL or the community patch directly.
- Expect incremental optimization from modders, not a fix for the core bottleneck. Community patches to DLSS 3 Frame Generation improved over time, and similar refinement attempts on the DLSS 5 DLL are likely, though the FP8 throughput gap on Ada Lovelace is a hardware limitation that software patching alone probably cannot close.
- More games will get the patched DLL tested against them. Tom’s Hardware already noted modders moving beyond Control within a day of the leak, and that expansion is likely to continue as more DLSS 5-integrated titles ship now that the feature is officially live on RTX 50.
- DLSS 4.5 remains the practical choice for RTX 4000 owners in the meantime. With Super Resolution and Ray Reconstruction running natively and efficiently on Ada Lovelace, most RTX 4080 and RTX 4090 owners are likely to stick with DLSS 4.5 rather than accept a 39% performance hit for early, unofficial DLSS 5 access, especially with Nvidia’s own optimized RTX 40 version now confirmed to be coming.
- This episode adds pressure on Nvidia’s RTX 50 pricing narrative. With GPU prices already climbing through 2026, both the community workaround and Nvidia’s own confirmation that RTX 40 will eventually get official DLSS 5 support give upgrade-skeptical RTX 4000 owners another data point when deciding whether Blackwell’s premium is worth paying right now.
The broader pattern: feature-gating as a GPU sales strategy
Nvidia is far from alone in tying flagship software features to the newest hardware generation, but the company has faced this exact criticism before. When DLSS 3 Frame Generation launched as an RTX 40-exclusive feature, critics argued the Optical Flow Accelerator justification was as much a sales lever as a technical necessity, a claim modders partially validated by getting frame generation running on older cards. The DLSS 5 situation complicates that narrative somewhat: the steep performance penalty on Ada Lovelace suggests the FP8 tensor throughput argument for Blackwell exclusivity holds up better than the Optical Flow Accelerator claim did for DLSS 3.
Still, the fact that a single modder could produce a functioning, if degraded, version of Nvidia’s newest flagship rendering feature within roughly a day of a leak underscores how quickly gated software features get reverse engineered once source material becomes available. For a company that has increasingly used exclusive software capabilities to differentiate each GPU generation, that speed is a recurring headache, even when the underlying hardware gap turns out to be real.
Frequently asked questions
Does DLSS 5 officially support RTX 4000 series GPUs?
Not yet, but official support is confirmed to be coming. DLSS 5 launched on September 3, 2026, exclusive to the RTX 50 “Blackwell” series, spanning the RTX 5060 through the RTX 5090, and by September 2026 Nvidia had extended it to titles like NBA 2K27 starting with driver 616.64 WHQL, still limited to RTX 50 Series desktop and laptop GPUs. One day after launch, Nvidia said DLSS 5 will officially come to RTX 40-series cards too, after further optimization work on RTX 50, though as of September 2026 the company has given no ship date.
How did modders get DLSS 5 running on RTX 4000 cards?
According to Tom’s Hardware and IT BOLTWISE, a modder took a leaked DLSS 5 library file, nvngx_dlssnr.dll, originally tied to Remedy’s Control, and patched its internal CUDA instructions to remove the Blackwell hardware check, allowing it to execute on Ada Lovelace GPUs.
What is the performance penalty for running the patched DLSS 5 mod?
IT BOLTWISE’s testing found an RTX 4090 dropping from roughly 135 FPS to about 82 FPS in Control when the patched neural renderer was enabled, a decline of about 39%. Russian outlet 3DNews independently reported the same 135-to-82 FPS numbers.
Which RTX 4000 cards have been confirmed to run the mod?
Reporting from Shazoo confirmed the patched DLL running on a GeForce RTX 4080, while IT BOLTWISE’s benchmark testing used an RTX 4090.
Is this similar to how DLSS 3 Frame Generation was unlocked on older cards?
Yes, in spirit. Modders previously got DLSS 3 Frame Generation running on RTX 20 and RTX 30 series cards despite Nvidia restricting it to RTX 40-series hardware. The DLSS 5 case follows the same pattern of a hardware-gated feature getting patched to run on unsupported GPUs, though with a much larger performance penalty this time.
Has Nvidia responded to the DLSS 5 leak or the RTX 4000 mod?
Not to the leak or the mod specifically. But on September 4, 2026, one day after DLSS 5’s official launch, Nvidia confirmed that RTX 40-series cards will get official DLSS 5 support after further optimization work on RTX 50, reversing its initial Blackwell-only stance.
Should RTX 4000 owners install the patched DLSS 5 mod?
Given that it relies on a leaked file and delivers a roughly 39% frame rate penalty in early testing, most RTX 4080 and RTX 4090 owners are better served sticking with the officially supported DLSS 4.5 feature set, which includes Super Resolution and Ray Reconstruction, until Nvidia ships its own optimized DLSS 5 support for RTX 40 hardware.
When does DLSS 5 officially launch?
DLSS 5 officially launched on September 3, 2026, exclusively for RTX 50 “Blackwell” cards. Nvidia said the following day that official support for RTX 40-series GPUs is coming later, after further optimization work on RTX 50.
Will RTX 40-series cards get official DLSS 5 support?
Yes, according to Nvidia. On September 4, 2026, the company confirmed that DLSS 5 will officially come to RTX 40-series GPUs after further optimization work on RTX 50, a reversal of its Blackwell-only launch stance from the day before. Nvidia has not given a timeline or said whether the official version will avoid the roughly 39% FPS penalty seen in the unofficial RTX 4090 patch.
Related Coverage
- RTX 4090 vs RTX 5080: 15% RT Gap, DLSS 5 Snub [2026]
- RX 9070 XT vs RTX 5070 Ti: $380 Price Gap, 6% FPS [2026]
- MSI Katana 15 HX C14: DDR4 RTX 5070 Cuts RAM Cost 2x [2026]
- GTA 6 Locked at 30 FPS on Consoles, PS5 Pro No Exception
- Nvidia RTX Prices Jump 39%, AI Chips Rise 15% [2026]
