Two independent developers say they have gotten NVIDIA’s DLSS 5 neural rendering pipeline running on Apple Silicon, using Apple’s Metal graphics framework as a bridge. The claim, first reported by Technetbook and picked up across the hardware press on September 14, 2026, describes a chain of unofficial tools rather than anything sanctioned by NVIDIA or Apple. If it holds up, it’s a striking demonstration that DLSS 5’s neural models are portable well outside the GeForce ecosystem they were built for, even if the current result is closer to a proof of concept than a product.
The two names behind the work are developers using the handles @iamwavecut and @Mappsnet7. According to reports, @iamwavecut ported DLSS 5’s neural rendering models to run natively on Apple Silicon through Apple’s Metal framework, while @Mappsnet7 built the bridge that connects that Metal backend into Windows games, using ReShade together with Game Porting Toolkit 4.0b2. Stitched together, the two pieces let a Windows title’s rendering pipeline hand frames off to a DLSS 5 model executing on Apple’s own GPU architecture instead of an NVIDIA RTX card.
What DLSS 5 actually is, and why running it on a Mac is notable
NVIDIA describes DLSS 5 on its own research pages as “DLSS 5: 3D-Guided Neural Rendering,” a system that runs locally as a rendering stage inside existing game pipelines on GeForce RTX 50 Series GPUs. That framing matters here: DLSS 5 isn’t a single upscaling filter bolted onto the end of a frame, it’s a rendering stage, meaning it sits inside the pipeline and has to interoperate with the engine’s existing draw calls, motion vectors, and frame history. Reproducing that outside NVIDIA’s own driver stack and outside CUDA is a materially harder job than porting a simple image filter.
That’s also why the Apple Silicon experiment is being treated as a hardware story rather than a routine emulation trick. Apple’s M-series chips use a tile-based deferred rendering architecture and Apple’s own Metal shading language, both quite different from the CUDA and DirectX-adjacent tooling NVIDIA designs DLSS around. Getting a generative neural rendering stage to execute at all on that stack, even slowly, says something about how portable the underlying models are once someone is willing to hand-translate the runtime.
The hardware: Apple’s M5 Pro takes the load
Multiple reports name the specific chip involved as Apple’s M5 Pro, and at least one account specifies that the experiment used the 16-core GPU configuration of that chip. The M5 Pro sits in Apple’s current Mac lineup as the step above the base M5, aimed at prosumer laptops and desktops rather than the top-end Max or Ultra tiers. Choosing the Pro tier rather than a Max or Ultra chip suggests the developers were working with hardware they already had on hand, not chasing a best-case benchmark on Apple’s most powerful silicon.
That detail cuts both ways for how impressive the demo actually is. On one hand, getting any version of DLSS 5 to execute on a laptop-class GPU core count, without CUDA, without NVIDIA’s tensor cores, and without official driver support, is a nontrivial engineering feat. On the other hand, it means the headline performance numbers being reported almost certainly understate what the same workaround might achieve on Apple’s larger Max or Ultra chips, where GPU core counts and memory bandwidth scale up considerably.
The performance numbers: 240ms latency, 2 FPS
The reported numbers explain why nobody is calling this shippable yet. One report puts measured latency at 240 milliseconds, and a separate report describes frame rates dropping to 2 frames per second during the DLSS 5 pass. For context, competitive PC gaming setups typically target total system latency under 20-30 milliseconds, and NVIDIA’s own Reflex tooling exists specifically to shave single-digit milliseconds off that number. A 240ms figure, or a 2 FPS frame rate, puts this squarely in “it technically runs” territory rather than “you’d want to play like this” territory.
It’s worth being precise about what that tells us and what it doesn’t. It tells us the software path works end to end: a Windows game’s frames can be intercepted, handed to a Metal-side neural rendering model, processed, and returned. It doesn’t tell us whether the bottleneck is the translation layer (ReShade plus Game Porting Toolkit 4.0b2 doing extra work to bridge Windows and Metal calls), the neural model itself running unoptimized on unfamiliar silicon, or some combination of both. Nobody involved in the reporting has published a breakdown of where the 240ms actually goes.
How the workaround is built, piece by piece
The chain described in reporting has three layers. First, @iamwavecut’s port takes DLSS 5’s neural rendering models, which NVIDIA ships to run inside GeForce RTX 50 Series pipelines, and re-implements the execution path so those same models can run natively through Apple’s Metal framework instead of NVIDIA’s own runtime. Second, @Mappsnet7’s contribution wires that Metal backend into actual Windows games using ReShade, the widely used post-processing injector that PC gamers already use for shader mods, combined with Game Porting Toolkit 4.0b2, Apple’s own compatibility layer for running Windows graphics code on Apple hardware.
Third, and implicitly, there’s the game itself, unmodified at the engine level, simply having its rendering calls intercepted and redirected. That’s the same basic approach mod communities have used for years to inject ray tracing or upscaling into games that don’t natively support it, just pointed at an entirely different chip architecture than anyone anticipated when DLSS 5 shipped for RTX 50 Series cards.
NVIDIA’s position: unofficial, unsupported, not a product
Every account of this story is explicit that the workaround is unofficial and does not carry NVIDIA support. NVIDIA has built DLSS around GeForce RTX 50 Series GPUs specifically, and the company’s public materials describe DLSS 5 as a rendering stage designed for that hardware. Nothing in the reporting suggests NVIDIA participated in, endorsed, or even necessarily knew about this port before it surfaced publicly. That distinction matters for anyone trying to gauge where this goes next: a hobbyist demo running at 2 FPS with 240ms of latency is a curiosity, not a signal that NVIDIA plans to bring DLSS to Apple hardware.
It also means none of the usual guardrails apply. There’s no driver validation, no game compatibility list, no support ticket to file if it breaks. Anyone attempting to reproduce this is working with a stack of independently built tools, any of which could stop working with the next macOS update, the next Game Porting Toolkit release, or the next DLSS revision from NVIDIA.
Historical context: modders have chased cross-platform DLSS before
This isn’t the first time DLSS 5 specifically has been dragged onto hardware NVIDIA never targeted. Earlier in 2026, modders got DLSS 5 running on older Ada Lovelace-generation RTX 4090 cards despite NVIDIA reserving the feature for RTX 50 Series hardware, and separately on RTX 20 through RTX 40 series GPUs more broadly, with reported frame rate costs as steep as roughly half the baseline performance. DLSS 5 has also shown up modded into emulators, including a PCSX2 build running on an RTX 4070 Super, well outside any context NVIDIA designed for.
What ties those efforts together is a pattern: NVIDIA ships a generative neural rendering feature tied to specific silicon, and within weeks a modding community strips the model out and forces it to run somewhere else, usually at a real performance cost. The Apple Silicon case fits that same pattern, except this time the “somewhere else” isn’t even NVIDIA hardware at all, it’s a completely different GPU architecture from a different vendor, which raises the technical bar considerably higher than porting between two generations of NVIDIA’s own cards.
Why cross-architecture portability is the bigger story than any single benchmark
The specific frame rate and latency numbers here are rough, but the underlying signal is more interesting than any single benchmark: neural rendering models, once trained, are not inherently locked to the silicon that trained them or the silicon they were shipped for. That’s consistent with how machine learning models generally work. A trained model is a set of weights and an execution graph, and in principle it can run on any hardware capable of executing that graph, whether that’s a GeForce RTX GPU, an Apple Silicon GPU through Metal, or in theory other accelerators entirely, with performance and efficiency varying enormously by platform.
NVIDIA’s commercial interest, understandably, runs the other direction: DLSS’s tight coupling to RTX hardware, including generation-gated features, is part of what has made upgrading to newer GeForce cards compelling for PC gamers chasing higher frame rates. Community efforts that decouple the software from the hardware, even clumsily and at steep performance cost, chip away at that argument, which is likely part of why this kind of unofficial port draws attention even when the actual playable result is nowhere near usable.
Market impact: what this does and doesn’t change
For NVIDIA, the near-term impact is close to zero. A 2 FPS demo built from ReShade, Game Porting Toolkit 4.0b2, and hand-ported neural models isn’t a competitive threat to RTX 50 Series sales, and nothing here suggests Apple is building official support for NVIDIA’s upscaling stack into macOS or Metal. NVIDIA’s DLSS ecosystem remains tied to GeForce hardware for anyone who actually wants playable frame rates.
For Apple, the story is a mixed bag. It’s a reminder that Apple Silicon’s GPU cores are capable of executing genuinely demanding neural workloads, even ones never designed with Metal in mind, which is a point Apple has been keen to make as it pushes M-series chips into AI and creative workflows. But it also underscores that Apple still has no first-party equivalent to DLSS for gaming, and that whatever gaming credibility Apple Silicon has continues to rely on unofficial community bridges like Game Porting Toolkit rather than native support from the graphics vendors that matter most to PC gamers.
For the broader modding and emulation community, this adds one more data point to a growing body of evidence that generative neural rendering features are becoming portable targets, the same way ray tracing shaders and texture mods were in earlier console generations. Expect more attempts at porting DLSS 5, and rival upscaling systems like AMD’s FSR line, across architecture boundaries as the tooling around Game Porting Toolkit and Metal shader translation matures.
Competitive comparison: DLSS 5 on native RTX hardware vs. the Apple Silicon workaround
| Factor | DLSS 5 on GeForce RTX 50 Series (native) | DLSS 5 on Apple Silicon (unofficial port) |
|---|---|---|
| Official NVIDIA support | Yes, first-party feature | No, unofficial and unsupported |
| Execution path | Local rendering stage inside the game pipeline via NVIDIA’s runtime | Ported Metal backend bridged into games via ReShade and Game Porting Toolkit 4.0b2 |
| Target hardware | GeForce RTX 50 Series GPUs | Apple M5 Pro (16-core GPU used in reported tests) |
| Reported latency | Not disclosed in this reporting; NVIDIA markets DLSS alongside Reflex for low-latency play | Reported as high as 240ms |
| Reported frame rate impact | Not disclosed in this reporting | Reported dropping as low as 2 FPS |
| Stability and compatibility list | Validated by NVIDIA drivers and game partnerships | None, dependent on independently maintained tools |
| Availability | Ships with supported GeForce RTX 50 Series systems | Not confirmed as publicly available or supported |
Timeline: DLSS 5’s modded history in 2026
| Approximate period (2026) | Reported development |
|---|---|
| Early 2026 | DLSS 5 launches tied to GeForce RTX 50 Series GPUs, marketed as a generative neural rendering stage |
| Mid 2026 | Community mods bring DLSS 5 to older Ada Lovelace RTX 4090 hardware, outside NVIDIA’s official RTX 50-only positioning |
| Mid 2026 | Broader mods extend DLSS 5 down to RTX 20 through RTX 40 series cards, with reported frame rate costs near half of baseline |
| Mid 2026 | A modded DLSS 5 build appears running inside the PCSX2 emulator on an RTX 4070 Super |
| September 14, 2026 | Reports surface of DLSS 5 neural models ported to Apple Silicon (M5 Pro) via Metal, bridged into Windows games with ReShade and Game Porting Toolkit 4.0b2; unofficial, with latency and frame rate figures well outside playable ranges |
What’s confirmed and what isn’t
It’s worth separating what multiple reports agree on from what remains genuinely open. Confirmed by the current reporting: DLSS 5 is the technology involved, NVIDIA’s own materials describe it as a generative neural rendering stage for GeForce RTX 50 Series GPUs, the Apple-side hardware is an M5 Pro with a 16-core GPU, the developers behind the port use the handles @iamwavecut and @Mappsnet7, the bridge tools are ReShade and Game Porting Toolkit 4.0b2, and every outlet covering this frames it explicitly as unofficial and unsupported by NVIDIA.
Not confirmed: whether NVIDIA has any plan, formal or informal, to support Apple Silicon officially. Whether this workaround will ever be packaged into something the public can install and run themselves. And there’s no confirmed pricing or launch date for any Apple-side DLSS support, because none exists, this remains a proof-of-concept built by independent developers rather than a shipped product.
Predictions: where this goes from here
- Expect the ReShade and Game Porting Toolkit 4.0b2 bridge code to circulate more widely in modding communities within weeks, even if it stays unpolished and far from playable frame rates.
- Expect NVIDIA to stay silent on official Apple Silicon support; the company’s business model depends on DLSS driving GeForce RTX upgrades, not on decoupling the software from the hardware.
- Expect follow-up attempts on Apple’s higher-end M5 Max and M5 Ultra chips, where more GPU cores and memory bandwidth could meaningfully cut the reported 240ms latency and 2 FPS frame rate, even without becoming genuinely playable.
- Expect similar unofficial ports targeting other non-NVIDIA architectures, following the same pattern already seen with DLSS 5 mods on older RTX cards and inside the PCSX2 emulator.
- Expect Apple to keep leaning on Game Porting Toolkit improvements for its broader Mac gaming push, without officially endorsing or blocking community bridges like this one.
Should anyone actually try this today
For nearly everyone, no. A 240ms latency figure and a 2 FPS frame rate make this unusable for actual gameplay, and the underlying tools, an unofficial Metal port of DLSS 5’s neural models plus a ReShade and Game Porting Toolkit 4.0b2 bridge, come with no support path if something breaks. This is a demo for people who want to see whether the idea works at all, not a setup anyone should expect to play games on.
That said, the fact that it works at all, however badly, is the actual news. Getting a generative neural rendering stage designed around NVIDIA’s CUDA and RTX Tensor Core pipeline to execute through Apple’s Metal framework on a completely different GPU architecture is a genuinely hard engineering problem, and the fact two independent developers cracked open enough of it to get frames rendering, even at 2 FPS, says the underlying models are more portable than NVIDIA’s tight hardware coupling suggests.
Frequently asked questions
Is DLSS 5 officially supported on Apple Silicon?
No. Every report on this workaround describes it as unofficial, built by independent developers, with no support or endorsement from NVIDIA.
What Apple chip was used in the demo?
Reports point to Apple’s M5 Pro, specifically the configuration with a 16-core GPU.
Who built the DLSS 5 Apple Silicon port?
Two developers using the handles @iamwavecut and @Mappsnet7 are credited in reporting. @iamwavecut is described as porting DLSS 5’s neural rendering models to run on Apple Silicon via Metal, while @Mappsnet7 built the bridge connecting that backend into Windows games using ReShade and Game Porting Toolkit 4.0b2.
How fast does DLSS 5 run on Apple Silicon compared to a GeForce RTX GPU?
Reports describe latency as high as 240 milliseconds and frame rates dropping as low as 2 FPS, far below what’s usable for actual gameplay. Native performance figures on GeForce RTX 50 Series hardware were not disclosed in this reporting.
Can I install this DLSS 5 Apple Silicon workaround myself?
There’s no confirmation that the tooling is packaged for public, general use. It’s currently reported as a developer proof of concept rather than a distributed, supported release.
Does this mean NVIDIA is bringing DLSS to Mac officially?
Nothing in current reporting suggests that. NVIDIA continues to build DLSS around GeForce RTX 50 Series hardware, and this remains an unofficial third-party effort.
Has DLSS 5 been ported to other unsupported hardware before?
Yes. Earlier in 2026, modders got DLSS 5 running on RTX 4090 cards outside its official RTX 50-only positioning, extended it to RTX 20 through RTX 40 series GPUs with steep performance costs, and got it running modded inside the PCSX2 emulator on an RTX 4070 Super.
What is Game Porting Toolkit 4.0b2?
It’s Apple’s compatibility layer for running Windows graphics code on Apple hardware, used here as one of the bridge components connecting the ported Metal DLSS 5 backend into Windows games.
Sources: Technetbook; NVIDIA’s DLSS technology page and research division; NVIDIA’s RTX DLSS developer documentation; Apple’s Metal developer page; additional coverage via Tom’s Hardware.
