Official DLSS Runs On AMD Navi 32 Via ZLUDA And Loses To FSR 4
By Indie Kings | October 4, 2026
Updated October 4, 2026: Modder countervolts ran official NVIDIA DLSS on AMD Navi 32 GPUs under Linux by routing DLSS calls through d4r, OptiScaler, Wine and ZLUDA, per PC Central. AMD FSR 4 stayed faster than that DLSS setup in every tested mode on the same RX 7700 XT, per PC Central.
Official DLSS on Navi 32
PC Central reports that a modder has run official NVIDIA DLSS on AMD Navi 32 GPUs using ZLUDA under Linux. The report is by Bartosz Mazur and was published Sep 29, 2026 at 21:20 UTC, per PC Central. PC Central summarizes the result in one line. A modder has successfully run official NVIDIA DLSS on AMD Navi 32 GPUs using ZLUDA under Linux, though FSR 4 remains faster in all tested modes.
PC Central says PC gamers using AMD Radeon RX 7000 series cards on Linux can now access NVIDIA DLSS through a community driven compatibility layer. That framing matters because it places Linux users at the center of the story. It also places community work at the center rather than vendor support.
PC Central says this offers an alternative upscaling method for users who prefer AMD hardware but want NVIDIA rendering features in supported titles. That claim is attributed to PC Central and is not expanded here. The article does not promise the same result in every title. It does not promise the same result outside the stated scope.
The project is known as d4r, per PC Central. PC Central says the project bridges the gap between competing graphics ecosystems without requiring official driver support. That bridge is software translation rather than vendor integration. It does not change the vendor driver stack itself.
PC Central says the d4r tool enables official NVIDIA DLSS to run on AMD RDNA3 GPUs under Linux and Proton environments. Linux and Proton are therefore the stated operating scope. Windows is not part of the stated scope in the source. Native AMD Windows driver behavior is not part of this claim.
It targets the Navi 32 architecture, per PC Central. Navi 32 powers the Radeon RX 7700 XT, RX 7800 XT, RX 7700, and Radeon PRO W7700, per PC Central. Those four products define the stated hardware scope. No other GPU is named in the source as supported.
Image: AMD Radeon RX 6800 XT graphics card. Credit: Wikimedia Commons, CC BY-SA 4.0, Vitaly Zdanevich.
The image above shows a Radeon RX 6800 XT. That card is a Navi 21 RDNA2 part, per its public product designation. It is not Navi 32 and it is not one of the supported cards named in this report. It is included only to illustrate an AMD Radeon card in general.
This distinction is important for readers who skim. The benchmarks below were recorded on an RX 7700 XT, per PC Central. The photo is not the test card. The photo does not show performance, compatibility, or physical installation.
Modder countervolts engineered the solution, per PC Central. PC Central says countervolts engineered the solution to intercept DirectX 12 calls and redirect them through a custom DLL that translates the workload for AMD hardware. That is the core technical claim. It describes interception plus translation rather than native execution.
The result is notable even before frame rates enter the discussion. Official NVIDIA DLSS logic is executing in a path that ends on AMD silicon. That path is maintained through several translation stages. Those stages are described in the next section.
The Shim Chain
PC Central describes a multi stage pipeline. Each stage has a defined role. No stage is presented as optional in the source.
The technical pipeline starts with the game sending DLSS inputs to OptiScaler, per PC Central. OptiScaler is therefore the first named receiver in the chain. The game itself is not named in the source. Any reference to a specific test game would be marked [NEEDS VERIFICATION] and is omitted here.
OptiScaler passes those inputs to d4r_nvngx.dll, per PC Central. That DLL is presented as a custom shim. It is not presented as an NVIDIA official binary. It is not presented as an AMD official binary.
This shim forwards the requests to NVIDIA official NGX core and nvngx_dlss.dll, per PC Central. That step maintains the standard CUDA execution path, per PC Central. The word standard matters here because it signals that NVIDIA logic is preserved rather than reimplemented.
The system then uses nvcuda.dll as a Wine bridge to connect to ZLUDA, per PC Central. Wine and Proton therefore provide the operating system compatibility context. ZLUDA provides the compute translation context. Those roles are distinct in the source.
ZLUDA translates the PTX code into native RDNA3 kernels, per PC Central. PTX is NVIDIA portable parallel code in this context. RDNA3 kernels are the AMD native execution target in this context. That translation is what lets AMD hardware do the work.
This architecture allows the GPU to execute DLSS 4 and 4.5 network layers directly on AMD silicon, per PC Central. Directly here means after translation rather than without translation. It does not mean native NVIDIA hardware execution. It does not mean AMD endorsement.
The table below restates the pipeline and cost elements exactly as published by PC Central. Standing is included for every row. No figure has been rounded, converted, or reinterpreted here.
| Element | Detail | Standing |
|---|---|---|
| DLSS 4.5 cost | about 7.5 ms per frame | Published by PC Central |
| DLSS 4 cost | fixed around 2.8 ms | Published by PC Central |
| Chain | game to OptiScaler to d4r_nvngx.dll to NGX core and nvngx_dlss.dll, CUDA path kept | Published by PC Central |
| Bridge | nvcuda.dll as Wine bridge to ZLUDA | Published by PC Central |
| Translation | ZLUDA turns PTX into native RDNA3 kernels | Published by PC Central |
| Scope | Navi 32: RX 7700 XT, RX 7800 XT, RX 7700, PRO W7700; Linux and Proton | Published by PC Central |
The chain is long, but the logic is linear. Game output enters OptiScaler. OptiScaler enters the d4r shim. The shim enters NVIDIA NGX logic. NVIDIA logic enters the CUDA path. The CUDA path enters Wine and ZLUDA. ZLUDA exits to RDNA3.
Readers should not mistake length for inefficiency alone. Some cost comes from model complexity rather than shim overhead. PC Central separates those costs by DLSS version. The Quality and Ultra Performance numbers make that separation concrete.
It is also worth stating what the chain does not claim. It does not claim official driver support, per PC Central. It does not claim endorsement by NVIDIA or AMD in the source. It does not claim a simplified setup path. This article gives no installation steps and no configuration.
Quality Mode Scoreboard
Benchmarks on the RX 7700 XT at 1440p were reported as follows, per PC Central. DLSS 3 reached 72.4 FPS in Quality. DLSS 4 reached 69.4 FPS in Quality. DLSS 4.5 reached 51.5 FPS in Quality.
Those three numbers define the Quality mode story. DLSS 3 leads the translated DLSS group in Quality. DLSS 4 trails DLSS 3 by 3 FPS in Quality. DLSS 4.5 trails by a much larger margin in Quality.
DLSS 4.5 is significantly slower because of approximately 7.5 ms per frame computational cost, per PC Central. That cost is large relative to a 60 FPS frame budget. A 60 FPS target allows about 16.7 ms per frame in total. A 7.5 ms upscaler cost therefore consumes a large share before rendering, submission, and display.
The practical effect is visible in the Quality result. DLSS 4.5 at 51.5 FPS is below the 60 FPS threshold often used as a smoothness reference. That does not make it nonfunctional. It does make it less attractive where frame rate is the priority.
DLSS 4 has a fixed cost of around 2.8 ms, per PC Central. That fixed cost explains part of its behavior across modes. A fixed cost matters less when the base rendering load is heavy. It matters more when the base rendering load is light.
PC Central presents DLSS 4 as more efficient for heavy upscaling modes like Ultra Performance. That efficiency claim is relative within the tested DLSS versions. It is not a claim that DLSS 4 beats FSR 4. FSR 4 remains ahead in the source across all tested modes.
The table below lists the reported frame rates. The Standing column marks each figure as published by PC Central. FSR 4 is listed without FPS because the source prints no FSR FPS figures.
| Mode at 1440p on RX 7700 XT | FPS | Standing |
|---|---|---|
| DLSS 3 Quality | 72.4 | Published by PC Central |
| DLSS 4 Quality | 69.4 | Published by PC Central |
| DLSS 4.5 Quality | 51.5 | Published by PC Central |
| DLSS 3 Ultra Performance | 88.9 | Published by PC Central |
| DLSS 4 Ultra Performance | 94.5 | Published by PC Central |
| FSR 4, any mode | faster than the DLSS figures in every tested mode, no FPS printed | Published by PC Central, figures unstated |
Quality mode is often the mode buyers ask about first. It preserves more source resolution than Performance or Ultra Performance. That makes it a useful reference for image stability. It also makes it the mode where heavier models show their cost most clearly.
In that Quality reference, the translated DLSS stack still works. It produces playable frame rates on the stated card at 1440p. It simply does not match AMD own upscaler on AMD hardware in the same test set. That gap is the central finding.
The source does not state FSR 4 FPS figures, per the limits of the source. This article therefore does not print them. It does not estimate them. It does not chart them. The correct statement is directional only. FSR 4 was faster in every tested mode.
Ultra Performance Flips 3 and 4
In Ultra Performance, DLSS 4 hits 94.5 FPS compared to 88.9 FPS for DLSS 3, per PC Central. That is a reversal from Quality. DLSS 3 led in Quality by 3 FPS. DLSS 4 leads in Ultra Performance by 5.6 FPS.
The reversal is consistent with the fixed cost account in the source. DLSS 4 has a fixed cost of around 2.8 ms, per PC Central. A fixed cost penalizes lighter upscaling less when the render load per frame is already low. Ultra Performance renders at a lower internal resolution than Quality, so the surrounding workload changes.
This does not mean DLSS 4 is free. A 2.8 ms fixed cost is still material at high frame rates. At 90 FPS, the total frame budget is about 11.1 ms. A 2.8 ms slice is about one quarter of that budget. At 120 FPS, the total budget is about 8.3 ms, so the same slice is larger in proportion.
The comparison is still useful because it shows version behavior is not uniform. DLSS 4 is slower than DLSS 3 in Quality in this test. DLSS 4 is faster than DLSS 3 in Ultra Performance in this test. Both statements are attributed to PC Central figures.
Readers should avoid generalizing beyond those two modes. The source does not provide a full mode by mode table in the excerpt verified here. Balanced and Performance modes are not listed with FPS in the verified fact block. This article does not fill that gap.
The Ultra Performance result also clarifies the DLSS 4.5 cost problem. DLSS 4.5 costs about 7.5 ms per frame, per PC Central. That cost is nearly three times the stated DLSS 4 fixed cost. A heavier model can still be justified by image quality in some contexts, but the source gives no image quality verdict here. The source gives frame rates and costs only.
For Linux users on Navi 32, the practical read is narrow. If the goal is maximum frame rate under heavy upscaling, DLSS 4 beat DLSS 3 in the one Ultra Performance pair printed. If the goal is Quality mode frame rate, DLSS 3 beat DLSS 4 in the printed pair. In both cases, FSR 4 beat the DLSS figures.
That last sentence is repetitive by design. The source repeats the FSR 4 lead as the bottom line. This article preserves that emphasis rather than burying it. The novelty is that DLSS runs at all. The result is that FSR 4 still wins on speed.
And FSR 4 Wins Anyway
AMD own FSR 4 remains faster than this DLSS implementation across all tested modes, per PC Central. That is the punchline the source keeps repeating. NVIDIA upscaler runs on AMD hardware and loses to AMD own upscaler on that same hardware.
The wording deserves care. Faster here means higher frame rate in the tested modes. It does not mean better image quality in the source. PC Central does not provide an image quality comparison in the verified facts. This article makes no image quality claim.
Faster also does not mean DLSS failed. A compatibility layer can succeed technically while trailing natively tuned code on speed. Translation has cost. Native paths often have less overhead. The source numbers fit that pattern.
The gap also matters for purchase logic. PC Central frames the d4r work as an alternative for users who prefer AMD hardware but want NVIDIA rendering features in supported titles. That alternative now exists in the stated scope. It is simply not the fastest option in the same test set.
Some readers may ask for the size of the FSR 4 lead. The source does not state FSR 4 FPS figures. This article therefore cannot state a percentage lead. It cannot state a frame delta. It cannot rank FSR 4 Quality versus FSR 4 Performance by FPS. Those figures are unstated.
The absence of FSR FPS is not a reason to dismiss the claim. Directional claims can still be informative when the source is explicit. PC Central is explicit that FSR 4 remains faster in all tested modes. That statement is carried here with attribution and without inflation.
The angle is therefore balanced rather than triumphal. Community engineering achieved something difficult. Vendor native code still won on speed. Both facts can be true at once. Both facts are true in the source.
For AMD GPU owners, the takeaway is simple. The test shows choice rather than necessity. DLSS can run through translation on Navi 32 under Linux and Proton in the stated setup. FSR 4 remains the faster path where it is available in the tested modes.
Unofficial and Unlinked
This is an unofficial community compatibility layer, per the nature of the project described by PC Central. It shims NVIDIA proprietary DLLs and the CUDA path onto AMD hardware under Linux. That description is technical rather than legal advice. It is included so readers understand what kind of software this is.
It is not endorsed by NVIDIA or AMD in the source. The source states no vendor statement from either company. This article therefore attributes no approval, partnership, or support to either vendor. Any claim of endorsement would be marked [NEEDS VERIFICATION] and is omitted.
This article gives no installation steps. It gives no configuration. It gives no command lines. It gives no file paths for installation. It gives no workaround for errors. That omission is intentional.
This article links no code, repository, build, DLL, or download of any kind. It names d4r, countervolts, OptiScaler, ZLUDA, Wine and Proton in prose without hyperlinking any of them. It does not link any GitHub page, repository, release, or file. It does not describe how to obtain NVIDIA DLLs.
The reason is scope and legal care. Readers should not treat this report as a setup guide. Readers should not treat mention of a tool as permission to copy proprietary files. Readers who need policy detail should consult vendor terms and the project documentation directly, outside this article.
The hardware scope also needs restating. The stated target is Navi 32 under Linux and Proton, per PC Central. The named cards are RX 7700 XT, RX 7800 XT, RX 7700, and PRO W7700, per PC Central. Use outside that scope is not established in the source.
The operating system scope needs the same care. Linux and Proton environments are stated, per PC Central. Other operating systems are not stated. Other compatibility tools are not stated. Driver versions are not stated. Kernel versions are not stated.
What Is Not Established
Several facts readers often expect are absent from the source. They are listed here so the article does not imply them. Absence is itself useful information.
The source does not state FSR 4 FPS figures. It states only that FSR 4 was faster in every tested mode. This article repeats that directional result without adding numbers. Any FSR FPS number seen elsewhere should not be attributed to PC Central without checking.
The source does not state the game tested, in the verified facts supplied here. If this article mentioned a test game at all, it would mark it [NEEDS VERIFICATION]. No game name is therefore given. No level, scene, or benchmark mode is given.
The source does not state driver or kernel versions, in the verified facts supplied here. Graphics driver behavior can change results. Kernel behavior can change compatibility. Neither is specified here because neither is specified in the source.
The source does not state whether this works outside Navi 32. RDNA3 is mentioned as the translation target, per PC Central. Navi 32 is the stated architecture scope, per PC Central. Other AMD architectures are not confirmed in the verified facts.
The source contains no statement from NVIDIA or AMD, in the verified facts supplied here. No approval is reported. No objection is reported. No comment is reported. Silence in the source is not evidence of a position.
| Unknown | Status in Source | Standing |
|---|---|---|
| Test game name | not stated in verified facts | Would be marked [NEEDS VERIFICATION] if mentioned |
| FSR 4 FPS figures | direction only, faster in every tested mode, no numbers printed | Published by PC Central, figures unstated |
| Driver and kernel versions | not stated in verified facts | Not established |
| Support outside Navi 32 | not stated, scope is Navi 32 only | Not established |
| NVIDIA or AMD statement | no vendor statement in source | Not established |
The table above is a guardrail. It tells readers what not to assume. It also tells future coverage what to verify before updating this story. A follow up with a named game, driver stack, and full settings would be more useful than speculation.
Readers should also avoid treating one RX 7700 XT result as a product line verdict. The RX 7800 XT, RX 7700, and PRO W7700 are named as Navi 32 scope, per PC Central. They were not the benchmark card in the printed figures. Their results are not printed here because they are not in the source.
Finally, performance in one title does not predict performance in all titles. Different engines stress upscalers differently. Different scenes stress fixed costs differently. The source figures are specific to the undisclosed test case behind them. They are reported here as specific rather than general.
FAQ
What did countervolts achieve on Navi 32?
Countervolts ran official NVIDIA DLSS on AMD Navi 32 GPUs under Linux through d4r, OptiScaler, Wine and ZLUDA, per PC Central.
Which cards are in the stated Navi 32 scope?
RX 7700 XT, RX 7800 XT, RX 7700 and PRO W7700 are named by PC Central, with benchmarks printed only for the RX 7700 XT at 1440p.
What were the reported Quality mode frame rates?
DLSS 3 reached 72.4 FPS, DLSS 4 reached 69.4 FPS and DLSS 4.5 reached 51.5 FPS in Quality at 1440p on the RX 7700 XT, per PC Central.
What happened in Ultra Performance?
DLSS 4 reached 94.5 FPS and DLSS 3 reached 88.9 FPS in Ultra Performance at 1440p on the RX 7700 XT, per PC Central, consistent with the stated 2.8 ms fixed cost for DLSS 4.
Why is DLSS 4.5 slower in the report?
DLSS 4.5 carries about 7.5 ms per frame of computational cost in the report, per PC Central, which weighs heavily on Quality mode frame rates.
Is FSR 4 faster and is this official?
FSR 4 was faster in every tested mode with no FPS printed, per PC Central, and the DLSS path is an unofficial community layer with no NVIDIA or AMD endorsement stated in the source.
Bottom Line
Official NVIDIA DLSS now has a reported path onto AMD Navi 32 under Linux through community translation. That path runs from game to OptiScaler to d4r_nvngx.dll to NGX core and nvngx_dlss.dll, then through nvcuda.dll and Wine to ZLUDA, which turns PTX into native RDNA3 kernels, per PC Central. The feat is real in the source and specific in scope.
The numbers keep the story grounded. DLSS 3 at 72.4 FPS, DLSS 4 at 69.4 FPS and DLSS 4.5 at 51.5 FPS in Quality at 1440p on the RX 7700 XT show a working but costly translation, per PC Central. Ultra Performance at 94.5 FPS for DLSS 4 versus 88.9 FPS for DLSS 3 shows version behavior can flip by mode, per PC Central. FSR 4 stayed faster in every tested mode, per PC Central, which is the punchline.
Readers should treat this as news rather than a guide. No installation steps are given here. No code or DLL source is linked here. The named cards, operating environments, pipeline stages, costs and frame rates above are the verified record. Everything else remains unstated until a fuller test report appears.
Sources: PC Central report by Bartosz Mazur, Modder runs official NVIDIA DLSS on AMD Navi 32 GPUs using ZLUDA, published Sep 29, 2026 21:20 UTC, https://pccentral.net/modder-runs-official-nvidia-dlss-on-amd-navi-32-gpus-using-zluda/. Technical roles for d4r, countervolts, OptiScaler, ZLUDA, Wine and Proton, plus NVIDIA and AMD hardware context, are described in prose per that source without code or download links.