I mean the Doom developer response was: "we literally can't rely on baking things out like the previous game any more"
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
-
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 5 - Generative AI
- Thread starter sharknice
- Start date
I mean the Doom developer response was: "we literally can't rely on baking things out like the previous game any more"
Actually, the newest version of doom does not have a fallback. It requires raytracing.
NVIDIA or AMD hardware Raytracing-capable GPU with 8GB dedicated VRAM or better (examples: NVIDIA RTX 2060 SUPER or better, AMD RX 6600 or better)
- Joined
- Mar 11, 2007
- Messages
- 1,006
Yeah, Doom: The Dark Ages and Indiana Jones and the Great Circle both require ray-tracing hardware just to be able to run.
M76
[H]F Junkie
- Joined
- Jun 12, 2012
- Messages
- 15,385
This sounds like a made up narrative to justify laziness and passing off the compute costs to the consumer.
First off: 24TB storage for a large developer is like nothing, not a factor at all.
Second: Nobody says to use the exact same old tech unchanged. You can improve on it. You can also use RT while the art is in flux, then only compute baked in lighting after it is finalized.
Third: I'm no expert so correct me if this is impossible but I don't think anything is stopping you from computing the lighting with RT then instead of rendering to screen you write out the assets with the lighting already applied, effectively using RT to bake in the lighting for 99% of static assets.
- Joined
- Dec 19, 2005
- Messages
- 17,804
"Capcom Responds to Resident Evil Requiem DLSS 5 Backlash: "We Got the Design Right"
by Cpt.Jank Today, 20:46 Discuss (3 Comments)When NVIDIA first revealed DLSS 5 and its neural rendering, a lot of the online backlash that was leveled at the feature centered around Capcom's Resident Evil character, Grace, and how DLSS 5 changed her facial features. Now, in a recent interview with Eurogamer, Koshi Nakanishi and Masato Kumzawa, the game's director and producer, have responded to the DLSS 5 outrage and to the game's overall reception at and since launch. Addressing the DLSS 5 debacle, Kumzawa took the backlash as a positive indication, commenting that "the fact a lot of players commented they really liked the original design of Grace and didn't want to see it changed was a positive—it meant we got the design right," and that it was an indication that "Grace quickly established herself as a fan favorite." The Nakanishi and Kumzawa did not comment on previous allegations that Capcom had not involved game developers or even inform them of the potential changes to the character designs ahead of time.
[Editor's note: Our in-depth review of Resident Evil Requiem is now live]
Just 16 days after Resident Evil Requiem launched, it had already sold over 6 million units, making it the fastest-selling Resident Evil game to date. Commenting on the overall success of the game, the pair said that "we were very happy to see the reaction of players around the world after the game launched. In particular, because we were very tight-lipped in order to maintain the surprise for players, especially about what happens after the return to Raccoon City. So, watching people's clips and online reactions, we felt not only pleased that they were enjoying it, but relieved our strategy was right to try and save that for them." Notably, the Resident Evil developer publicly asked gamers not to spoil surprises about the game when the game's plot was leaked online shortly before the official launch."
- Joined
- Dec 19, 2005
- Messages
- 17,804
- Joined
- Dec 5, 2018
- Messages
- 5,032
"Capcom Responds to Resident Evil Requiem DLSS 5 Backlash: "We Got the Design Right"
This is a really oddly worded headline. It should say Capcom agrees with the backlash from fans. It sounds like the opposite here. Or I have terrible reading comprehension.
This is a really oddly worded headline. It should say Capcom agrees with the backlash from fans. It sounds like the opposite here. Or I have terrible reading comprehension.
Prince Valiant
Gawd
- Joined
- May 3, 2023
- Messages
- 816
It's a bit off in isolation. Makes sense with the full quote, third sentence in the first paragraph Erek included.
"the fact a lot of players commented they really liked the original design of Grace and didn't want to see it changed was a positive—it meant we got the design right,""
- Joined
- Dec 19, 2005
- Messages
- 17,804
“NVIDIA DLSS 5 Neural Rendering DLL Leak Hints at Nearby Launch
by AleksandarK Today, 05:19 Discuss (5 Comments)NVIDIA's DLSS 5 neural rendering technology might be one of the most controversial announcements, as the gaming community has been quite vocal about their opinions on the new rendering technique. However, the launch of DLSS 5 is expected this fall, and we are already seeing the first DLSS 5 DLL files leaked through games, as confirmed by EA's NBA 2K27 early access, which includes the new DLSS-NR file. The game is now in early access and is already shipping with DLSS Neural Rendering, essentially DLSS 5. This comes as a 158 MB package, which is substantially larger than any previous DLSS DLL file, indicating that NVIDIA has indeed packed a generative model inside the game, with only a tiny footprint considering that a whole generative AI model is running locally on the GPU.
However, this is about twice the size of the current official DLSS 4.5 Ray Reconstruction DLL file, which measures 72 MB. While size is not the only comparable quality of this DLL, we can assume that NVIDIA has included a massive feature set given the added storage space. If users recall, NVIDIA once again showcased its DLSS 5 neural rendering at the SIGGRAPH 2026 event, highlighting that one of the challenges faced with DLSS 5 was maintaining visual identity, object semantics, and character poses, along with output lighting semantics, motion, and art direction. Within these strict parameters, DLSS 5 applies material photorealism to enhance visual appeal. There has been some progress recently, but the fall launch is expected to shed more light on the entire situation and integration.”
https://x.com/renan_maniero/status/2092762553990447113
Last edited:
- Joined
- Dec 2, 2004
- Messages
- 32,863
Complete speculation by tpu "news" as is often the case. Neural rendering isn't the same thing as the generative Ai shown in dlss5 clips.
Reasonable speculation too, that the DLSSNR dll will have something to do with DLSS neural rendering.
RTX_neural texture, neural shaders and so on, has been available for a while to try:
https://github.com/NVIDIA-RTX/Rtxns
And do not require the nvngx_dlssnr.dll....
Down to 160mb, maybe it will still fit on regular 8GB vram gpu and that the first demo were using non quant-distiliated, etc..... model, those dlss model a full smaller magnitude smaller than even the very small LLM, could eventually fit on sram of a large enough but cheap old wafer node Groq like chiplet extension.
RTX_neural texture, neural shaders and so on, has been available for a while to try:
https://github.com/NVIDIA-RTX/Rtxns
And do not require the nvngx_dlssnr.dll....
Down to 160mb, maybe it will still fit on regular 8GB vram gpu and that the first demo were using non quant-distiliated, etc..... model, those dlss model a full smaller magnitude smaller than even the very small LLM, could eventually fit on sram of a large enough but cheap old wafer node Groq like chiplet extension.
- Joined
- Dec 19, 2005
- Messages
- 17,804
“NVIDIA DLSS 5 DLL Leaked by NBA 2K27 Early Access Build, Here's Our Analysis
EXCLUSIVE by AleksandarK Today, 10:35 Discuss (0 Comments)Earlier today, we reported that a DLSS 5 DLL has been leaked by an early access build of 2K's NBA2K27 early access. NVIDIA announced DLSS 5 earlier this year, and showed more details at SIGGRAPH this year, saying it would ship "this fall." However, we were surprised that it is shipping already, and with a title like NBA 2K27, without any major announcement or review coverage, pushing us to dig what is happening beneath the surface. Maybe the developers just put the wrong file into the release before pushing it out.
Copies of the DLL were mirrored by people and uploaded to Reddit, where we obtained ours. Just to be clear up front: NVIDIA didn't provide any information for this article, there was no briefing, no press build or anything. All findings reported here come from taking a publicly circulating file apart and trying to figure out what is inside it. Also, before you get too excited, the file cannot actually be used in the game—NBA 2K27 exposes no setting to turn it on, and nothing in the shipping build appears able to drive it. You also can't just drop it into any other game's folder—the game has to be coded specifically to load that DLL and feed it the right information, more on that later.“
Not sure about this part: shipping binary made it into a game months before anyone expected to see it
The video game ship this september and the article say that Nvidia showed more details at SIGGRAPH this year, saying it would ship "this fall."
Lot of people in business saying fall would have simply september-october-november in mind and not some astronomical equinox that occur september 22... and it is common for title that support latest Nvidia stuff to launch without them and get it months later with an update.
I imagine it could be true that everyone expected the feature to be late and not hit its Q3 ambition, but that sound a bit overselling how early this is showing up versus the official timeline.
The video game ship this september and the article say that Nvidia showed more details at SIGGRAPH this year, saying it would ship "this fall."
Lot of people in business saying fall would have simply september-october-november in mind and not some astronomical equinox that occur september 22... and it is common for title that support latest Nvidia stuff to launch without them and get it months later with an update.
I imagine it could be true that everyone expected the feature to be late and not hit its Q3 ambition, but that sound a bit overselling how early this is showing up versus the official timeline.
Remember boys and girls, these features exist to cut development costs.
Don’t pay artists for years, have them build a frame give it the most basic features, then tell the AI how you want it finished.
Why pay teams of level designers to build and scatter and create for years when you can tell an AI to do it.
This tech exists to remove people from the process to shrink teams and dramatically shrink the development cycle.
It is not sustainable to spent 5 years or more on a title, publishers know it, investors are figuring it out, studios know it, and the likes of Sony and Xbox are being forced to acknowledge it.
Frame rate or graphics improvements are secondary effects
Don’t pay artists for years, have them build a frame give it the most basic features, then tell the AI how you want it finished.
Why pay teams of level designers to build and scatter and create for years when you can tell an AI to do it.
This tech exists to remove people from the process to shrink teams and dramatically shrink the development cycle.
It is not sustainable to spent 5 years or more on a title, publishers know it, investors are figuring it out, studios know it, and the likes of Sony and Xbox are being forced to acknowledge it.
Frame rate or graphics improvements are secondary effects
Maybe eventually, but a title like NBA 2K27 will be almost all consoles sales and that the type of game that make a lot of money (and does not necessarily have a 5 years per title average), far ahead are the days of something like DLSS5 cutting dev cost much.
It could be the first time ever that is the case, that the cut of development cost is used to make game development cheaper instead of game bigger (like they always did until now pretty much), but I would not necessarily bet on it and more and more less but bigger AAA games could be what happen.
Using Unreal 5 or internal general engine instead of a per game affair exist to cut development cost, game cost did not go down that much.
Hollywood did go that road a bit, movie budget peaked (adjusted) probably in the 2004-2008 window, before the 2008 crisis and digital (from filming, to distribution, to VFX) made things cheaper if you do the same, the Marvel-StarWars big stuff did not got that much cheaper, just a little bit.
It could be the first time ever that is the case, that the cut of development cost is used to make game development cheaper instead of game bigger (like they always did until now pretty much), but I would not necessarily bet on it and more and more less but bigger AAA games could be what happen.
Using Unreal 5 or internal general engine instead of a per game affair exist to cut development cost, game cost did not go down that much.
Hollywood did go that road a bit, movie budget peaked (adjusted) probably in the 2004-2008 window, before the 2008 crisis and digital (from filming, to distribution, to VFX) made things cheaper if you do the same, the Marvel-StarWars big stuff did not got that much cheaper, just a little bit.
Last edited:
- Joined
- Dec 19, 2005
- Messages
- 17,804
View: https://streamable.com/q26fxz, higher res demo...
Seem impressively good limited to the characther face like that, when thing goes for a realistic look and photo-realist quality, it could get really good at it fast, frankly already quite something:
https://imgdiff.net/s/1b0b18d858fdc344d6acb8c31df376e0
A lot of it is getting the light to be close to a real simulation and not having a flattening effect it seem.
Last edited:
- Joined
- Aug 15, 2005
- Messages
- 9,192
Desaturates color, pushes black level and mid tones, shifts the white balance and color tone to be more "cold". Creates an illusion of a shifted/additional lightsource. Cuts framerate in half. So.......looks like a Reshade mod, But requires way more performance from your GPU.
And ultimately, it ruins the way the character looks. Shifting the light source and boosting mid-tones contrast, tends to over emphasize the model geometry and makes the character look older.
Also, this idea of darkening shadows, boosting mid-tones, and desaturatiing colors tends to counter the visual impact of Global illumimation. Which is a pretty big deal effect in gaming, the past couple of years.
View: https://www.youtube.com/watch?v=oxXdIVc9Suo
And ultimately, it ruins the way the character looks. Shifting the light source and boosting mid-tones contrast, tends to over emphasize the model geometry and makes the character look older.
Also, this idea of darkening shadows, boosting mid-tones, and desaturatiing colors tends to counter the visual impact of Global illumimation. Which is a pretty big deal effect in gaming, the past couple of years.
View: https://www.youtube.com/watch?v=oxXdIVc9Suo
Last edited:
You can play with different styles and intensities for everything you described (as shown in the GUI), or just toggle it to 'off' all together
Even more controls/selections seen in the original video LukeTbk posted:
Last edited:
- Joined
- Dec 19, 2005
- Messages
- 17,804
i don’t like it, seen this same look years ago with gta 5 online mods
it’ll get old fast
it’ll get old fast
- Joined
- Dec 19, 2005
- Messages
- 17,804
Integrating DLSS 5 Neural Rendering
Driving nvngx_dlssnr.dll from an application — NGX does not officially support this
Written 27 August 2026 · Verified end-to-end on Vulkan (Minecraft / Caustica) and D3D12 (standalone harness) · NVIDIA driver 616.56 · RTX 5080 · NGX 310.8.0
1. What This Feature Actually Is
DLSS Neural Rendering is not a denoiser and not an upscaler. It is an image-space neural post-process applied to a finished frame.
Streamline registers exactly four required buffers for it, which is the clearest statement of its contract:
kBufferTypeUpliftInputColor
kBufferTypeUpliftOutputColor
kBufferTypeMotionVectors
kBufferTypeDepth
No albedo, no normals, no roughness. If your renderer already feeds DLSS Ray Reconstruction, you already produce a superset of what NR needs.
Expectation setting: The model is trained to reconstruct organic detail — skin, hair, faces, fabric. On content with little sub-pixel structure to infer (stylised or low-frequency textures), the honest result is local contrast shaping and mild sharpening, not transformation.
2. The Three Obstacles, in the Order You Will Hit Them
Symptom Result code Real cause
Create fails, driver predates DLSS 5 0xBAD0000C FAIL_OutOfDate Version gate. Note this proves the feature ID is correct.
Create fails on a current driver 0xBAD0000B FAIL_UnableToInitializeFeature The NGX core has no feature→snippet mapping for NR and rejects it during validation without ever opening a library.
Calling the snippet directly fails 0xBAD00002 FAIL_PlatformError The snippet checks who is calling it. This is the real wall.
Do not chase the wrong causes. Every one of these was independently ruled out with evidence: driver version, GPU generation, per-application entitlement (all 31 app IDs in the local OTA manifest were swept), NvAPI initialisation, parameter-block flavour, init signature, D3D12 vs. Vulkan, and the driver profile matched by executable name.
None of them move 0xBAD00002.
3. The Caller Check
Disassembling NVSDK_NGX_D3D12_Init_Ext in the snippet shows the gate in the prologue:
GetModuleHandleExW(FROM_ADDRESS, <return address>, &hModule)
; on failure -> "Error: Unable to determine calling module" -> 0xBAD00002
GetModuleFileNameW(hModule, path, MAX_PATH)
wcsstr(path, L"nvngx.dll")
; no match -> "Error: Not called from NGX runtime - %S" -> 0xBAD00002
The snippet resolves the module owning its return address, takes that module's file path, and requires it to contain the literal substring nvngx.dll.
The driver's core is _nvngx.dll, which contains it. Your shim does not.
The Fix
Make the call from a module whose file name carries that substring.
Copy your shim beside itself as:
nvngx.dll_<yourname>.dll
Load the copy and forward the NR calls through it.
Same code, different module identity.
Two variants that look equivalent but are not:
A directory named nvngx.dll also satisfies the substring test but breaks process startup.
An executable named with it fails to start at all.
Only the calling DLL's own name is safe to change.
4. Step by Step
Place the snippet. Put nvngx_dlssnr.dll in a directory your code controls. NGX searches the application directory for feature snippets, which is why community instructions say to drop it next to the game.
Let the NGX core initialise normally. You still need it: it owns the parameter block you will pass to the snippet.
Create the compliant caller. On first use, CopyFile your shim to nvngx.dll_<yourname>.dll in the same directory and load it. Refresh the copy every run so it cannot lag a rebuild.
Resolve the snippet's own exports from nvngx_dlssnr.dll. It ships a full Vulkan surface (14 entry points, more than its 10 for D3D12), so Vulkan is first-class:
NVSDK_NGX_VULKAN_Init_Ext2 / Init_Ext
NVSDK_NGX_VULKAN_CreateFeature
NVSDK_NGX_VULKAN_EvaluateFeature
NVSDK_NGX_VULKAN_ReleaseFeature
Use the snippet-build signatures. The public headers branch on NGX_SNIPPET_BUILD, and the snippet exports that variant — version and parameter block, not the application form taking FeatureCommonInfo. Guessing wrong here is silent stack corruption, not a compile error.
Pass the core's capability parameter block. A freshly allocated block lacks the snippet and preset callbacks features expect. Where NVSDK_NGX_Parameter_SetUI is not available (it is a static-library helper, not a core export), set through the vtable: NVSDK_NGX_Parameter is a pure-virtual class and slot 3 is Set(const char*, unsigned int).
Create with feature ID 18 and keep the handle.
Evaluate per frame with colour, depth, motion vectors, and an output image that is not the input.
Create once; never create-then-release. CreateFeature records initialisation work into the command buffer. Releasing before that buffer executes leaves the GPU referencing freed resources and produces VK_ERROR_DEVICE_LOST.
Release only at teardown, with the device idle, and before destroying the images the feature was created against.
5. Parameters
Parameter Notes
DLSSNR.Color / Depth / MVec / Output Output is the only read-write resource.
*SubrectBaseX/Y, *SubrectWidth/Height Per resource. Depth and motion may be render resolution while colour and output are display resolution — describe each at its true extent.
DLSSNR.MVecScaleX/Y Motion vectors in render pixels consumed against a display-resolution image scale by the upscale ratio, not 1.0.
DLSSNR.Width / Height / DepthInverted / Reset Reset on the first frame after a feature rebuild.
DLSSNR.Hint.Render.Preset Its own namespace — see below.
Intensity, Style, UseAutoMask, LocalStructureStrength, LocalToneStrength, SkinStructureStrength Tuning. LocalStructure drives detail synthesis; LocalTone drives local contrast and reads as ambient-occlusion-like shading.
6. Colour Space Matters
NR expects a display-referred image.
Running it on linear HDR before tone mapping does not merely look different — the model reads values as far brighter than intended and blows out colour instead of adding detail.
Place the pass after your tone mapper and before any debug or UI composite.
7. Preset Letters Are Per-Feature Namespaces
A preset letter means different things to different features. Do not carry one across.
Feature Library Presets
Super Resolution nvngx_dlss.dll Where the J/K/L/M letters live.
Ray Reconstruction nvngx_dlssd.dll A/B/C removed. D = 4, E = 5 (latest transformer). F and above are documented as “do not use, reverts to default” — and a shipping integration's log confirms it: CreateFeature FAILED on preset F … retrying on preset E.
Neural Rendering nvngx_dlssnr.dll Its own DLSSNR.Hint.Render.Preset table.
A screenshot showing “Preset F” under DLSSNR says nothing about RR's preset F.
8. Streamline Is a Dead End Today
Streamline knows the feature as kFeatureDLSS_NR, ID 1004 (from the embedded manifest in sl.dlss_nr.dll).
But slIsFeatureSupported returns eErrorFeatureNotSupported, and the log explains why:
Could not find version matching for plugin: dlss_nr_0
Ignoring plugin 'sl.dlss_nr' since it is not supported on this platform
The OTA manifest carries dlss_0, dlss_d_0, dlss_g_0 and others, but no dlss_nr_0.
The plugin loads and verifies as NVIDIA-signed, then disables itself.
The direct route works; the Streamline route does not.
9. Diagnostics Worth Wiring First
NGX will tell you what it is doing if you ask.
Route its logging through NVSDK_NGX_FeatureCommonInfo.LoggingInfo at verbose level, and mirror it to a file — a shim's stderr often does not reach the host application's log.
Lines worth watching:
NGXSecureLoadFeature: unable to read the deny list. Assuming that the feature is allowed.
SnippetLocationInfo::load: Module not found at <path>
NVSDK_NGX_CreateFeature_Validate: app id is <id>
The absence of a SnippetLocationInfo line for your feature is itself the finding: it means the core never looked for a library, because it has no mapping for that ID.
10. Recovering an Unknown Feature ID
If a future feature's ID is unknown, recover it from any binary that already uses it.
On x64 Windows, the second integer argument to CreateFeature lands in edx, so scan for mov edx, imm32 immediately preceding a call, and anchor on a feature you already know:
18007dc72: movl $0xd, %edx
18007dc7a: callq *0x12a7e0(%rip) ; -> CreateFeature
; 13 = RayReconstruction (known-good anchor)
18007c0ba: movl $0x12, %edx
18007c0c2: callq *0x12c398(%rip) ; -> same pointer
; 18 = the unknown feature
Same call target, different ID.
That is the identification.
Reconstructed from first-hand debugging. Result codes, strings, and offsets quoted here were observed directly.
- Joined
- Apr 25, 2001
- Messages
- 15,906
LOL. We heard the same sour grapes about DLSS when it came out. DLSS is now the standard. This will be the same.
Shows the various presets in the leaked DLL (which again all have individual adjustments/dials that can be tweaked for each as seen in prior videos - he plays with some of them too)
View: https://www.youtube.com/watch?v=hFIpT2he2DE
He shows some of the dials in action at/during this timestamp:
View: https://youtu.be/hFIpT2he2DE?t=309
View: https://www.youtube.com/watch?v=hFIpT2he2DE
He shows some of the dials in action at/during this timestamp:
View: https://youtu.be/hFIpT2he2DE?t=309
Last edited:
I don’t like it, but you’re not wrong.
And tech like it will form the backbone of future game engines.
- Joined
- Dec 19, 2005
- Messages
- 17,804
what happens when the dll is deleted from the game files?
It won't launch, but really, why would you want to? There are two situations for a game that uses something like this:
- What we are going to see for a long time, maybe forever, is an option that you can just turn off. You know, like the other DLSS features. So turn it off if it offends you. I mean there are people who refuse to use upscaling. They are silly people, but you can be silly if you like. Just turn it off.
- If it ever becomes required you wouldn't want it off because it would be an integral part of the art style choice and disabling it would mean the game would look wrong. It would be like removing the normal or specular textures in a game now: You COULD do it, but you wouldn't want to, it would look wrong.
Whatever, I'm sure people will cream, cry and whine when it comes out, find the examples of its worst issues while ignoring the things it does best, declare it crap, and refuse to use it. Just like upscaling. I'm also sure it will be useful, and for the people who aren't dogmatic about it, they'll find games that are uplifted by it and offer visuals that are just not possible with brute-force rendering.
- Joined
- Aug 15, 2005
- Messages
- 9,192
It all just looks like the cover of a body building magazine.
Emphasize every contour/add specular highlights. Boost contrast of shadows and mid-tones. desaturate color.
In a movie, a DP would have to be extremely specific about lighting, to get people's Philtrum to look boldly contoured like that. And it wouldn't look more/less static like that, as they moved around the scene.
On the lowest "0" settings, its not necessarily egregious. But, the color is still heavily altered and desaturated (And Control already isn't very colorful!). And a lot of the intentional colored light bleed/global illumination, is sucked out.
Even giving this my most generous take--------not worth literally cutting FPS in half.
Mind you you're still looking at it backported/implemented into a game(s) it was never designed/intended to be implemented into to begin with
Just like there are downsides implementing 'regular DLSS-pre5' into non-DLSS games and artifacts/downsides that occur when doing that
- Joined
- Aug 15, 2005
- Messages
- 9,192
lol there are some cursed ass images starting to circulate
There's a few where it looks actually pretty good, but it cannot be trusted near faces. It's either super uncanny or fucks them up.
- Joined
- Mar 20, 2017
- Messages
- 434
Might be bull shots. I'd rather check the videos...
I just write protect the directory. I ran and old copy of BG3 for months that way until BG3 for my friend with a Mac finally matched. I have also download old copies from Steam repos.Steam will just redownload it on startup/file consistency check