Nvidia 3080 or 3080 TI 3D vision support using modded drivers

[skyrimer]

The only good thing is that 3080 availability is so low that I wouldn't be able to buy it even if 3d vision worked, and in case it doesn't, i will still wait for 3070 to be available so that 2080ti prices plummet even more and then grab one, so no hurries and just glad to leave the driver masters do their work, i'll sure donate for anyone that can fix it somehow, though i think there's no need for extra motivation here, but still.

[AlexTOPMAN]

The behaviour reported definitely sounds like when you have 3dvision going but games are running exclusive fullscreen.

Didn't you know that? (I`m about DX11)

[Losti]

OK, i have done some research here ref. the new 456 driver. I am pretty sure that there is no hope to bring the things back on an easy HexEditor based attempt. Installing 456 and 3D-Vision will work using my driver change tool, native files without any hack. No 3D-game will kick in stereo, no crash or something, just no Stereo (Stereo Disabled). I have compared the nvwgf2umx.dll from 452 and 456. There are huge changes. The major thing is, that in the 452 file you can find words like: "3d Vision" and the respective driver file nvSCPApi64.dll. In the 458 file those entries are completely missing. Nothing about the things like "anaglype" or "last sight" and so on, all the things are present in 452 file, but not in 458.

So as it seems they completely removed the call of the 3D-Vision driver from the GPU driver files

.
.
.

I have started a DX11 game without the nvwgf2um.... files. It will not work. I have started a DX9 game without theese files, it will work. Starting a DX11 game without the nvd3dum.... files,will work, but a DX9 game will not work. Starting a DX11 game win32 it will not work without nvwgf2um... files and also not without nvd3dum....files. So it Seems some games uses both files, some DX11 games needs only nvwgf2um but DX9 games will only use nvd3dum...files and HERE the Stero3D entries are still present. So thats why we can run DX9 games but not DX11 games.

.
.
.

So over all this is not a great thing to stick your heads into the sand We stay at 452 and update the profiles with the profiles from the latest driver and can play like before. MAY BE there are games that needs a newer driver to work with the lateset features...OK but most should work this way. The only thing is that we cant use a card later than the 20er series.

.
.
.

If there is no HERO injecting the things into the current driver that are needed for stero3D i cant see any hope at the end of the tunnel...NV definitively removed the Stero3D-DriverCall from the the 456 driver for DX10/11 ... currently not for DX9 ...


This is the comparison for the 452er version of the nvwgf2umx.dll and the one from 456....so in any way except the things from above you will see a massive change:

[skyrimer]

Terrible news then, looks like 2080ti it is, I still think that we can go forward with 2080ti performance and 452 driver for probably a few years in most games (even now 425 drivers work OK with many games) but yeah sad news. It's actually kind of a miracle that we've survived with hacked drivers for so long.

As people mentioned before, we can't do over 1440p in 3d, and 3080 is barely 15-20% faster at that resolution, so it wouldn't be a big improvement anyway.

I haven't seen Schwig around, I hope he has some trick to fix it, but this looks way more complicated than a simple hack.

The worst part is that I can't understand what could possibly Nvidia gain by removing all 3d support even if it was in such legacy, abandoned form, they truly wanted to wipe every sign of 3d like it was a curse or something, it's so stupid.

[TonyMontana]

I totally agree with Losti here, Nvidia has apparently overwritten/deleted the Stereo3D-DriverCall routine in nvwgf2um(x).dll, so maybe we could somehow identify/extract the code-fragment from a previous nvwgf2um.dll and then inject it in the current dll.

Luckily, Nvidia has not (yet) removed the Stereo3D-DriverCall from nvd3dum.dll. I am planning to analyse this dll and compare it with different versions of both nvd3dum.dll and nvwgf2um.dll to find the code responsible for Stereo3D-DriverCall.

I will compare at least 10-20 versions of this dll from previous driver generations. In theory the Stereo3D-DriverCall should be the same in every dll generation, so I will try to identify it that way, asking myself what parts of the file have never changed?

I wonder, if we will not find a solution to modding 456 drivers to enable 3DV, wouldn't it work the other way around by modding the 452 driver to support the RTX30xx-cards? I will also take a turn in this direction...

[Lysander]

Man, I wonder if the nVidia driver devs were shaking their heads at the 3DV removal if they knew what proper 3dv looks like. I know I would. But I know it's due to low adoption

[skyrimer]

It sounds promising, best luck TonyMontana!

[AlexTOPMAN]

We still have а hope that another fork of the driver will not contain the new NvTOPPS plugin + SQL database, as was added in 456.38. I think that the problem with 3D stereo appeared precisely because of these innovations.

[P.C.Zen]

Not the greatest news but good to hear people are still working on finding a fix.

If there's ultimately no way to enable 3D with cards later than the 2000 series is SLI an option? Is it possible to make and/or refine SLI profiles for newer games? Might that be an option moving forward...? ...to extend the 3D vision's lifespan on the hardware side of things?

[Necropants]

Believe me, SLI is not an option.... I was an avid user and even I gave up in the end. No developer support.... I'm supremely annoyed, this seems almost malicious....

Okay Nvidia you don't want me give you any more money, obviously....

[AlexTOPMAN]

IMHO, someone with good programming skills should think about how to make a new stereo driver through the use of MS DWM (with its strategy of removing exclusive fullscreen mode) to merge stereo pairs and SLI frames from different video cards (maybe two+), without having any problems with the NVidia video driver as it is now. You also need to think about where and how to intercept control from the video driver in order to send a second frame with a shifted camera (+ separation and convergence settings) to the same video card or to another one in SLI: leave the previous method or have a better one.

[AlexTOPMAN]

Okay Nvidia you don't want me give you any more money, obviously....

It`s time to talk with AMD about vacant niche of: MultuGPU (it`s a real chance to leave NVidia behind in gamers top performance graphics systems and price/performance benchmarks), 3dstereo glasses and "VR performans gaming" (with mGPU VR of course).

[Robert256]

Interesting work Losti !

Btw, I found strings nvVideoProcessorSetOutputStereoMode and nvVideoProcessorSetStreamStereoFormat in the new drivers, but don't know what they do and if it would be of any help...

Also "Only two views supported for stereo high profile", and a few other hits on "stereo".

[Lysander]

Okay Nvidia you don't want me give you any more money, obviously....

It`s time to talk with AMD about vacant niche of: MultuGPU (it`s a real chance to leave NVidia behind in gamers top performance graphics systems and price/performance benchmarks), 3dstereo glasses and "VR performans gaming" (with mGPU VR of course).

I'd go for it, but it's not really nVidia that's the problem, it's just not enough ppl care about stereo 3d. Were really a small niche. Of all the ppl I know personally only one has had a 3d tv, I believe paired with a playstation. Other than my demo, nobody really experienced proper 3dvision. Ppl dont care, especially if u ask them to put stuff on their head, keep it charged, pay for it in the first place...

[RAGEdemon]

There is also the possibility of hacking old drivers to work on new generations of cards: I remember modifying inf/ini files to soft-mod my GTX into Quadro for professional applications...

It would be interesting to see of we can mod older drivers into running on a 3080 while having 3DVision still work...

[skyrimer]

The thing is that if we can get to work just this one driver, even if not a single one after this works, the whole 30XX would work with it in 3D, easier said than done of course.

The old driver compatibility is an interesting approach, however without a 3080 it will be hard to test, though I bet if anyone can get their hands on one they could easily sell it even for a profit since they are impossible to find.

Just out of curiosity I tested the driver that preceded my 1080ti release and it didn't recognize it, so some kind of hack would probably be necessary.

[Necropants]

This was my intention and I secured a card, but was so angry at this. (not nessesary that it broke but how it broke.) I cancelled my order.

[TonyMontana]

I am currently analysing earlier versions of the "nvwgf2um.dll" to find the Stereo3D-DriverCall. The picture above shows a comparison between 452.06 vs. 431.36:

- Although same hex-adresses do not match same binary content, both dll files work completely with 3DV in DX 11 (with their respective drivers of course)
- I therefore assume that the 3DV code is not "fixed" to specific addresses in the file, as adresses for 3DV-related binary content actually (almost) never match the same adresses in different dll generations
- I even tried merging binary content from 452.06 to 456.38 and surprisingly it did not crash when starting a DX10/11 game with the patched dll !
- Nvidia seemed to regularly have changed huge chunks of this dll even in older versions in the past, so it was just a matter of time for the 3DV code to be overwritten, which ultimately happened now
- e.g. in v375.57 it had a file size of "only" 14 MB, now it has grown to a monstrous 38 MB

So I can confirm that 456.38 driver allows the dll to be modified without crashing.


I will try now to find the Stereo3D-DriverCall. So here is my question for the 3DV experts: Do you know what exact hex-code makes the API-Call for nvSCPAPI.dll and nvSCPAPI64.dll? Would this somehow be related to the STEREO_API_ACTIVATION_CONTROL (as seen above)?

Please let me know if you can tell what piece of code could make the call - even if you do not know exactly but have a rough assumptionI I will be happy to test it out. Thank You.

[bo3bber]

For testing purposes, you might want something simpler than games or other stuff. Here's my 3D_Vision_Direct sample app, compiled as Debug, including the pdb files for any crashes or errors you might see. This includes the HelixVision/3DFM 3DMigoto to do the driver override, and the Schwing hacked nvwgf2umx.dll file, so you can just double click the Tutorial07.exe and it works, when you have 452.06 installed.

http://bo3b.net/3D_Direct_Test_452.06.7z


This just enables 3D Vision using the DirectMode in the driver, and draws a spinning cube in 3D.

The DirectMode fails in the same way that Automatic mode fails, so this should be a good test case. When I tested on 456.38, the call fails:
status = NvAPI_Stereo_CreateHandleFromIUnknown(g_pd3dDevice, &g_StereoHandle);

But curiously the calls to SetDriverMode and NvAPI_Stereo_IsEnabled both succeed.


It might be possible to track what happens in the NvAPI_Stereo_CreateHandleFromIUnknown in the success case, and then compare to what happens in the fail case.

In any case, maybe this simple test app makes testing easier.


@TonyMontana: I'd expect this NvAPI_Stereo_CreateHandleFromIUnknown to call that STEREO_API_ACTIVATION_CONTROL in the dll, but don't know for sure.

@Losti: I think I have to agree that the code has been pulled out of the dll altogether. While binary comparisons are generally useless because small code changes change the actual bytes and make for messy compares, the lack of strings in the file does seem pretty telling.
If I open the file in Notepad++, and just search for the word 'stereo', I get:
452.06: 235 hits
456.38 10 hits

So obviously there's a lot that's been killed. Also from a software development perspective, if I was changing code in there for new features, I would probably go ahead and kill code like 3D Vision that was no longer supported.

However- we do see that we can call the stereo calls successfully. They return a generic 'not found' error, but it doesn't crash. This suggests that we might be able to redirect those calls to working driver or somehow restore the dead code. As noted, the code itself has not changed in years.

[TonyMontana]

@bo3bber: Thank you very much for supplying your 3D_Vision_Direct sample app! I am sure this will help us testers a lot! I just had a look at it on 452.06 (with 3DV set up normally) but it crashes shortly after starting without displaying the cube animation. Is your app compatible with Win10 1709 ? If not, I will switch to a higher version of Win10. Are there any software-dependencies the app might need besides from the 452.06 driver?

This suggests that we might be able to redirect those calls to working driver or somehow restore the dead code.

These are both very interesting ideas.

As Nvidia did not delete the code from the DX9-dlls, could this help us in any way?

[RAGEdemon]

“Der Weise kann vom Narren lernen” - "The wise might learn something from a even a fool".

Would it perhaps make any sense to attempt to decompile the dll? Some reading shows that it is quite easy for .net, more difficult for C#, and very difficult for C++; but tools do exist, e.g:

https://www.hex-rays.com/products/decompiler/
https://github.com/0xd4d/dnSpy
https://github.com/icsharpcode/ILSpy

What crackers do is use a disassembler.

On the other hand, maybe it's time to abandon 3DV altogether for a better modern hardware agnostic solution which also performs very well on the CPU? Helifax's wrapper is a great success; i.e. maybe instead of focusing on trying to get 3DV working again, it might make sense to only try to get 3DV glasses/monitor output activated and then write wrappers for DX10/11/12? Maybe even abandon 3DV monitor output altogether if it's too difficult and focus on VR as a direct output? I think most of us have VR in some form nowadays.

iZ3D were also successful in creating their own; and they had the best driver.
VorpX has been successful.
Tridef/DDD were successful too.

I am not a coder; just some alternative suggestions from a fool that might inspire the wise...

[AlexTOPMAN]

This suggests that we might be able to redirect those calls to working driver or somehow restore the dead code.

To place the old driver pack into subfolder (for example) and redirect calls to the new path... it should work, I think.
...
Ok, the other way we have is to find the right person and invite them to our forum to help us solve our problem.
What about him? Ingo Esser
https://developer.nvidia.com/blog/author/iesser/
https://www.crunchbase.com/person/ingo-esser
Could someone try to contact to him? (sorry, but I don't speak English fluently and my vocabulary leaves much to be desired ).

[AlexTOPMAN]

maybe it's time to abandon 3DV altogether for a better modern hardware agnostic solution

Better modern hardware solution is a VR. As expected, VR things are not as good as we would like them to be, either (I`m talking about "the same" 3dstereo 2d games supporting, VR SLI etc.).

which also performs very well on the CPU

CPU solutions is a bad idea. The CPU main rendering thread in most cases is a bottleneck in terms of high performance. Another solution is multithreading, but it is very difficult and problematic to add it to an existing software architecture. Not our way, IMHO.

get 3DV glasses/monitor output activated and then write wrappers for DX10/11/12?

This is the path through Microsoft DWM, as I wrote above.

iZ3D were also successful in creating their own; and they had the best driver.

I met the founder and owner of iZ3D Vadim Asadov in 2007 and spoke personally. I can try to search for him again and restore contact, and also ask him which of the guys wrote this driver for him.

I am not a coder too, although I was a beginner in the past.

[Losti]

I am currently analysing earlier versions of the "nvwgf2um.dll" to find the Stereo3D-DriverCall. The picture above shows a comparison between 452.06 vs. 431.36:

- Although same hex-adresses do not match same binary content, both dll files work completely with 3DV in DX 11 (with their respective drivers of course)
- I therefore assume that the 3DV code is not "fixed" to specific addresses in the file, as adresses for 3DV-related binary content actually (almost) never match the same adresses in different dll generations
- I even tried merging binary content from 452.06 to 456.38 and surprisingly it did not crash when starting a DX10/11 game with the patched dll !
- Nvidia seemed to regularly have changed huge chunks of this dll even in older versions in the past, so it was just a matter of time for the 3DV code to be overwritten, which ultimately happened now
- e.g. in v375.57 it had a file size of "only" 14 MB (still huge), now it has grown to a monstrous 38 MB

So I can confirm that 456.38 driver allows the dll to be modified without crashing.

IMHO I believe that if we knew the exact hex-code that makes the call for 3DV to be activated, we could solve this compatibility issue.

I will try now to find the Stereo3D-DriverCall. So here is my question for the 3DV experts: Do you know what exact hex-code makes the API-Call for nvSCPAPI.dll and nvSCPAPI64.dll? Would this somehow be related to the STEREO_API_ACTIVATION_CONTROL (as seen above)?

Please let me know if you can tell what piece of code could make the call - even if you do not know exactly but have a rough assumptionI I will be happy to test it out. Thank You.

Yeah of course you cannot search for entries in the DDL from different driver at the same adress, its not unusial that that will not match. So a merge will not be possible, also we do not know any depending settings or things in the DLL without decompiling it. I have tried to decompile the ddl files without sucess, usingsome quickly found "DLL exporers" or something similar.

[AlexTOPMAN]

Did someone try 460.20? https://developer.nvidia.com/cuda/wsl/download

[RAGEdemon]

which also performs very well on the CPU

CPU solutions is a bad idea. The CPU main rendering thread in most cases is a bottleneck in terms of high performance. Another solution is multithreading, but it is very difficult and problematic to add it to an existing software architecture. Not our way, IMHO.

Indeed, the CPU bottleneck of 3DVision is well known. I did an investigation into it with the immense help of our community here:
https://www.nvidia.com/en-us/geforce/fo ... n-thread-/
nVidia confirmed it and were working on a fix/rewrite, before they abandoned 3DV altogether

What I meant by what I said above is that we know 3D vision is bottlenecked by the CPU - if we are designing a new driver/wrapper then we ought to plan to avoid this known CPU bottleneck. For example, Helifax's Vk wrapper apparently gives exactly half FPS compared to 2D, which seems to indicate that there is no CPU bottleneck there - amazing work.

On a related note, Win10 now offers independent GPU selection per app - perhaps the stereo3D overheads can be passed onto another GPU, e.g. every intel CPU has an iGPU that is never used.

Unfortunately, VorpX has huge performance issues. This graph is from another of my investigations into performance:
viewtopic.php?f=105&t=25199

The developer of VorpX actually rewrote parts of VorpX to improve performance in that thread.

Interesting iZ3D/DDD/3DV/2D performance comparison graphs also further down that thread.

iZ3D were also successful in creating their own; and they had the best driver.

I met the founder and owner of iZ3D Vadim Asadov in 2007 and spoke personally. I can try to search for him again and restore contact, and also ask him which of the guys wrote this driver for him.

I am not a coder too, although I was a beginner in the past.

This sounds like a fab idea. I believe they tried to auction the iZ3D driver for ~$500,000 but were unable to sell it so it has been abandoned. You *might* be able to persuade him to release the code as open source? That would be brilliant.

A word of caution against contacting nVidia:

1. Bo3b already tried - they told him to "pound sand". No harm in asking again of course...

2. We are reverse engineering which is expressly forbidden by nVidia's EULA. They will likely have a team of lawyers with cease and desist letters, to shut us down if not sue us into oblivion, if they even catch a whiff that we are trying to poke into their driver. Perhaps this might be another reason to start something new as well as trying to salvage the old...

[AlexTOPMAN]

Helifax's Vk wrapper apparently gives exactly half FPS compared to 2D, which seems to indicate that there is no CPU bottleneck there - amazing work.

Using only 50% of the maximum possible performance in fps - is not the best result overall. The correct way is to set another path to another video card at the API level with all the same data as for the first video card, and an additional command to convert all 3d coordinates in the game (before they are converted by the video card into triangles), taking into account the necessary shift and rotation of the player's camera to draw the image for the second eye. All of this can be done by the video card, not the CPU. In this case we going to effectively utilize CPU up to 100% with no more "waste" work (trying to do the same on other CPU core to keep one thread performance at high will cause us many other tough problems).

Win10 now offers independent GPU selection per app

Sounds great!

[TonyMontana]

OK, as far as I understand this we now have the following options:

1) decompile / disassemble the dll and try to restore the "dead" code
2) redirect api calls to a previous working dll
3) modding 452 drivers to support RTX 3000
4) hire / contact a software engineer
5) abandon Nvidia 3DV driver and develop something similar on our own
6) hitting the jackpot by nvidia accidentally restoring the code in a future driver release for whatever reason

I will personally focus on options 1)-2) and try to find a way to restore / "revive" the dead code. Not sure if option 2) would be easier to achive?

I have to say that I am not an expert for dll decompilation / disassembling. If someone of you knows a software engineer / coder / cracker please let them know and if we're lucky they could help us out. Thank you!

[bo3bber]

iZ3D were also successful in creating their own; and they had the best driver.

I met the founder and owner of iZ3D Vadim Asadov in 2007 and spoke personally. I can try to search for him again and restore contact, and also ask him which of the guys wrote this driver for him.

I am not a coder too, although I was a beginner in the past.

That would be brilliant. Please try to contact Vadim and see if there is any chance we could get their code. At this point it's clear it has no value, so maybe he'd enjoy seeing it live on as an open-source project. We can also easily do closed source if he prefers.

Point him to my and DarkStarSword contributions on 3Dmigoto as proof of our skill and interest. We can make anything work no matter how messy, no documentation, broken tools. None of that matters, we walk through walls. Having that great code base would save thousands of hours of work.

[RAGEdemon]

===================
Edited
===================

[skyrimer]

Maybe if there's a branch of HelixVision using iZ3D code, share a percentage of the benefits?

[3DNovice]

....

[Necropants]

In other words. not at all.
Also not only is it NVlink only (this is actually a good thing.) But they changed the form factor of the connection AGAIN!
These bridges are not cheap. Time to put a fork in SLI sadly. That said I am likely to just get another 2080ti when they drop for backup purposes...

[bo3bber]

@bo3bber: Thank you very much for supplying your 3D_Vision_Direct sample app! I am sure this will help us testers a lot! I just had a look at it on 452.06 (with 3DV set up normally) but it crashes shortly after starting without displaying the cube animation. Is your app compatible with Win10 1709 ? If not, I will switch to a higher version of Win10. Are there any software-dependencies the app might need besides from the 452.06 driver?

Well, not much of a sample app if it crashes for other people. Not sure what happens there, although it is a full debug version. Could easily be something installed/setup on my machine that I don't recall. I don't think the OS version of 1709 will matter, although maybe. The schwing hack was not necessary on that version of the OS, so try renaming the nvwgf2umx.dll in the folder out of the way so it won't load. Could be the hack is not compatible on 1709.

OK, as far as I understand this we now have the following options:

1) decompile / disassemble the dll and try to restore the "dead" code
2) redirect api calls to a previous working dll
3) modding 452 drivers to support RTX 3000
4) hire / contact a software engineer
5) iz3D somehow magically agrees to give us their code
6) abandon Nvidia 3DV driver and develop something similar on our own

I will personally focus on options 1)-2) and try to find a way to restore / "revive" the dead code. Not sure if option 2) would be easier to achive?

I have to say that I am not an expert for dll decompilation / disassembling. If someone of you knows a software engineer / coder / cracker please let them know and if we're lucky they could help us out. Thank you!

That seems like a reasonable list, I tweaked. I'm taking a look at redirection, option 2. But don't hold your breath.

[helifax]

Helifax's Vk wrapper apparently gives exactly half FPS compared to 2D, which seems to indicate that there is no CPU bottleneck there - amazing work.

Using only 50% of the maximum possible performance in fps - is not the best result overall. The correct way is to set another path to another video card at the API level with all the same data as for the first video card, and an additional command to convert all 3d coordinates in the game (before they are converted by the video card into triangles), taking into account the necessary shift and rotation of the player's camera to draw the image for the second eye. All of this can be done by the video card, not the CPU. In this case we going to effectively utilize CPU up to 100% with no more "waste" work (trying to do the same on other CPU core to keep one thread performance at high will cause us many other tough problems).

Win10 now offers independent GPU selection per app

Sounds great!

Hi, no offense, but I am not sure you understand the full picture: Getting the 50% FPS in 3D vs 2D is the MAXIMUM you can get. For Stereo 3D you need to render the image TWICE right? So for 60 FPS in 2D you want 60 FPS in 3D, but you render 2 eyes right? This translates in x2 GPU usage. So rendering 120FPS in 2D will give you 60 FPS in 3D - since you are rendering 2 perspectives. Hence what Rage was talking about the 50% of the 2D frames.
Also, what you are describing there is exactly what we are doing.... Have you read the papers on how 3D Vision does the stereo?

Cheers,
Helifax

[TonyMontana]

Well, not much of a sample app if it crashes for other people. Not sure what happens there, although it is a full debug version. Could easily be something installed/setup on my machine that I don't recall. I don't think the OS version of 1709 will matter, although maybe. The schwing hack was not necessary on that version of the OS, so try renaming the nvwgf2umx.dll in the folder out of the way so it won't load. Could be the hack is not compatible on 1709.

Thank you for your quick response!
On both Win10 1709 and 2004 (both freshly installed and properly set up with 452.06 and 3DV), when starting the "Tutorial07.exe" it displays an error message saying that "MSVCR120D.dll" is missing. Although I have pasted a version (from the internet) of this missing dll in the sample-app folder, the application then opens a window, but crashes shortly afterwards.
So I think this could be solved by including the "MSVCR120D.dll" from your system? It would be great if you would upload it again together with this dll. (Maybe you could test it on another computer before).
Besides, does your sample app also load the 32-bit version "nvwgf2um.dll" or only 64-bit "nvwgf2umx.dll"?
Does the app support anaglyph (discover) mode?

[Marin3D]

I assume you are missing ms redistributables VC12(2013), you may see which library is compiled on VC12 using Dependency Walker http://www.dependencywalker.com/ or install the MSVC 2013 redistributables download from here https://www.microsoft.com/en-us/downloa ... x?id=40784

[helifax]

"MSVCR120D.dll" is the DEBUG VERSION (notice the D at the end) of the Microsoft C Runtime Library. 120 refers to the version of the toolkit/compiler that was used to build the code: 120 stands for the Toolkit that comes with Visual Studio 2013/2015 - or the Microsoft Visual C 2013/2015.
The DEBUG version you cannot download as a Redistributable! You need to install Visual Studio 2013 to get access to it.

This means the sample was built as a Debug Version, so... in order to be used by anyone that doesn't have VS 2013/2015 or the toolkit installed, it should be build in Release but with Debug Symbols (I am not sure if this make sense for most people )
I am sure Bo3b can address this very quickly though as he knows exactly what to do and well.. to debug it -------- obviously you need Visual Studio installed anyway!

[TonyMontana]

I am sure Bo3b can address this very quickly though as he knows exactly what to do and well.. to debug it -------- obviously you need Visual Studio installed anyway!

Thank you for explaining this. But even with VS2013 installed it still crashes, so I would be grateful for a recompiled version. In the meantime, I will do the testing with DX 10/11 games.

I have also found a resource I would like to share with you. It explains the process of DLL redirection and may be useful for our purposes:

https://www.exploit-db.com/docs/english ... ection.pdf

[AlexTOPMAN]

So rendering 120FPS in 2D will give you 60 FPS in 3D...

The problem is - only few games can scale with 2x performance. Often, games have no support SLI at all. I mean, new 3Ds SLI can work in any case, because doubling data happens only at CPU to GPU transfer point. CPU+GPU natively balanced games (optimized by developers) have more chance to double fps in 3Ds SLI.

I remember modifying inf/ini files to soft-mod my GTX into Quadro for professional applications...

Here are the instructions (sorry, that in Russian): https://forums.overclockers.ru/viewtopi ... #p17065337