Performance difference between 425.31 and 452.06

[3DNovice]

...

[john105]

I think the problem lies with W10 or my motherboard USB driver heuristics, if USB ports are in a sleep state, it doesn't wake up to be available for it's install
or something weird in respect to it being recognized. Dunno, just guessing. But if I restart and then install, the install seems fine. Installs after the PC has
been running for awhile, seem to have issues. I also disconnect from the internet during installs, just an old habit.

Maybe it's related to https://www.nvidia.com/en-us/geforce/fo ... d-enabled/

[3DNovice]

....

[ZeGURU69]

Thanks for your answers but I don't think my problem comes from improper installation of the 3D USB controller.

My pyramid seems to be working well, Nvidia's 3D test works (while 3D does not launch in game) and I also have access to 3D discovery

I tested installing 425.31 drivers on 21H1, 21H2 and Windows 11 with the same result. On a 1809, no problem.

[RAGEdemon]

DID you use DDU in safe-mode and ensure you used the STANDARD NON-DCH drivers, opting for the "clean instal" option?

Also, you don't list your specs. Please add it to your sig so people can help troubleshoot.

[ZeGURU69]

DID you use DDU in safe-mode and ensure you used the STANDARD NON-DCH drivers, opting for the "clean instal" option?

Also, you don't list your specs. Please add it to your sig so people can help troubleshoot.

Yes, I have tested everything but a priori it is a "normal" problem on recent revisions of Windows 10. Losti has the same problem.

I recently bought an I7 12700KF and that's why I had to leave version 1809 because the performance with this processor is not good. I mostly played Day Gone and the performance is much better than my old setup

I had before an I7 8700K @ 4900 Mhz in some areas, I could go down to 20-30 fps (see less) while now I'm often at 60 fps and overall I don't go below 40 in most areas . On my backup I was at 30 fps on my old configuration and I went to 50 with the new processor.

So a very good upgrade but maybe a Ryzen 5 5800X does the same or better.

My config: RTX 2080 Ti, 32 Go @ 3600Mhz

[RAGEdemon]

That's really interesting. Maybe try a fresh instal of windows?

There was a recent discovery by some of our smart guys that the best performing driver is 425. After that, there is a 10-15%+ performance penalty by hacking 3DV to work with those drivers.

DJ-RK tested his 5600x against my 9900k. There was a consistent 20% performance difference, though I don't know how much of that is due to us using different drivers. There was a user who had a 9700k who consistently got 10% faster than my 9900k when in every test the 9900k ought to be the same or faster.

Results here:
viewtopic.php?p=177865#p177865

The new Alder-Lake ought to be at least as fast as Ryzen 5, but usually ~10% faster.

It is very interesting to see that your 8700k 30fps has jumped to 12700k 50fps! That's a 66% jump - wow! It's not a number of cores issue - just those huge cores brute forcing the 3DV CPU bug:


Do you have GTA5 installed? It has its own Built-in 3D Vision. Does that work? It would be awesome to get the above benchmark data from you so I can compile a 9900k vs 5600x vs 12700k chart

[3DNovice]

...

[RAGEdemon]

Indeed, the mitigation patches were a hypothesis, however I disabled them for further testing and still got the same results. More details in my last couple of post in the linked thread.

The only thing I can think of is that I had enabled 16x aniso, as a force of habit after every driver instal, or more likely, all 3 of us were using different nVidia drivers, which were affecting performance differently - I would like to get to the bottom of this, however our dataset is very limited

Thanks for the link.

[ZeGURU69]

That's really interesting. Maybe try a fresh instal of windows?

There was a recent discovery by some of our smart guys that the best performing driver is 425. After that, there is a 10-15%+ performance penalty by hacking 3DV to work with those drivers.

DJ-RK tested his 5600x against my 9900k. There was a consistent 20% performance difference, though I don't know how much of that is due to us using different drivers. There was a user who had a 9700k who consistently got 10% faster than my 9900k when in every test the 9900k ought to be the same or faster.

Results here:
viewtopic.php?p=177865#p177865

The new Alder-Lake ought to be at least as fast as Ryzen 5, but usually ~10% faster.

It is very interesting to see that your 8700k 30fps has jumped to 12700k 50fps! That's a 66% jump - wow! It's not a number of cores issue - just those huge cores brute forcing the 3DV CPU bug:


Do you have GTA5 installed? It has its own Built-in 3D Vision. Does that work? It would be awesome to get the above benchmark data from you so I can compile a 9900k vs 5600x vs 12700k chart

Hello RAGEdemon

First of all happy new year

I'm already on a fresh install of Windows, in fact I'm currently using Windows 11 for 3D. I failed to use the 452.06 drivers on this system because it is stuck on 3D discovery. If I select generic screen (my Optoma UHD40 is recognized like this) in the Nvidia panel, it refuses to save the changes and puts me back on 3D discovery.

I have GTA 5, I reinstall it and I found the integrated benchmark, but there is no result screen afterwards? I launched it and I get between 65 and 120 fps with all graphics options enabled except display distance.

Otherwise, I also tested Jedi Fallen Order which runs in the tested portion of the game at constant 60 (but with a sometimes heavy GPU load at 98%). I also tested The medium which went very badly on my old configuration. It runs better but it seems to be GPU limited since there are portions where I am at 40 fps with a GPU load at 98%. In Day Gones also there are places that are GPU limited (but always above 50-55 fps).

[RAGEdemon]

Are you using an EDID override from a compatible display? IIRC, UHD40 is not a 3DV compatible projector.

You'll need to install an external benchmark app called CapFrameX.

GTA5 benchmark settings: 720p default settings except VSync off, 3DV on. Advanced graphics settings - everything disabled.
NVCPL: Everything default.

Happy new year to you too!

[ZeGURU69]

No I don't use EDID Override, it is not necessary on the Optoma UHD40, I select generic CRT screen and I am in 3D Vision.

I installed CapFrameX and I launched a bench with your setting. I put the file in the message.

[RAGEdemon]

Thank you. To be clear, is this with 3DV ENABLED and WORKING?

Apologies for my lack of understanding, however from what I understand, your 3DV is not currently working, so I'm not sure about the validity of these results if 3DV is not kicking in in comparison to DJ-RK and my results where 3DV is working (and severely impacting GTA5 performance as a result)

Or are you saying that 425 works fine, and it's only 452.06 which doesn't work?

[ZeGURU69]

3D Vision works, I have to use Losti's tool after every system start to repair the driver (there is an option in its tool to repair the driver). Tests have been done on the 425.31 with 3D vision enabled and functional.

[RAGEdemon]

Thank you. I have posted the results in the CPU thread. Wow!

viewtopic.php?p=183108#p183108

[john105]

You'll need to install an external benchmark app called CapFrameX.

GTA5 benchmark settings: 720p default settings except VSync off, 3DV on. Advanced graphics settings - everything disabled.
NVCPL: Everything default.

In case you're looking for results on more systems, I attached what I got on mine.
Core i9-9900KS (5GHz, no overclock), DDR4 32GB 3600Mhz, GeForce GTX 1080 Ti SLI

Not sure how accurate CapFrameX compares results though. Wouldn't it be easier to just use the benchmark results that the game saves to C:\Users\<username>\Documents\Rockstar Games\GTA V\Benchmarks? Here's what I got from the same run with CapFrameX.

Code: Select all

Frames Per Second (Higher is better) Min, Max, Avg
Pass 0, 44.337463, 91.184883, 79.815498
Pass 1, 50.725891, 181.874039, 82.907211
Pass 2, 10.564146, 223.880981, 81.799934
Pass 3, 66.777748, 210.957809, 106.234787
Pass 4, 33.840359, 188.895874, 94.469299

[RAGEdemon]

@john105, Your 9900KS results graph are averaging the same as the 12900K, which is 66% faster than our 9700K and 9900K results OC to 5.1GHZ All cores with 4000MHz memory. Are you SURE your 3D Vision was ENGAGED during the benchmark? Thanks.

@ZeGURU69, would you kindly also confirm that 3D Vision was ENGAGED (double image on the screen) when the benchmark was in progress? Also, presumably you're on DDR5?

Thank you.

[john105]

Are you SURE your 3D Vision was ENGAGED during the benchmark?

Yes, of course. Don't forget that I have SLI. It makes a big difference in GTA5

[RAGEdemon]

Are you SURE your 3D Vision was ENGAGED during the benchmark?

Yes, of course. Don't forget that I have SLI. It makes a big difference in GTA5

No it doesn't. The whole point of testing this scenario is that the GPU has no impact because we are testing at default at 720p. Only the CPU is being stressed

Unfortunately, there is no way your 9900KF results are valid - sorry. Something is amiss.

[john105]

No it doesn't. The whole point of testing this scenario is that the GPU has no impact because we are testing at default at 720p. Only the CPU is being stressed

Unfortunately, there is no way your 9900KF results are valid - sorry. Something is amiss.

The GPU always has an impact, it's just less at 720p and lower settings. You can check my results in an earlier thread viewtopic.php?f=105&t=23289 especially the last post. I can record a video with SLI on and off if you don't believe

[RAGEdemon]

No it doesn't - maybe a small percentile within margin of error, which doesn't have any practical significance. If you look at your GPU usage, it ought to be <90% on both cards.

Please take some kind of camera footage of the benchmark running, showing a double 3D Vision image. Please also make sure NVCPL and all driver settings / game settings / config tweak are removed/DEFAULT.

Edit: This is how your capture looks as it stands. As you can likely agree, there is no way this is accurate (you are light orange - unknown)

[john105]

Please take some kind of camera footage of the benchmark running, showing a double 3D Vision image. Please also make sure NVCPL and all driver settings / game settings / config tweak are removed/DEFAULT.

Here's the video https://easyupload.io/6zitx6

Edit: This is how your capture looks as it stands. As you can likely agree, there is no way this is accurate (you are light orange - unknown)

I'm actually surprised that the i7-12700KF is that good without SLI in this game, because GTA5 is one of a very few cases where SLI makes a big difference. If we tested 480p and min settings, what you said earlier would've been closer to the truth, but it was still 720p + default (mid) settings. I wonder who would win if we tested in 1080p + max settings. SLI should give much more advantage in such case, but GTA5 is too unpredictable to know for sure

[RAGEdemon]

Thanks for the upload and effort. You seem to be correct - well done.

Generally, a more powerful GPU / SLi shouldn't matter if the GPU usage in the game being benchmarked is low - i.e. the GPU is not the bottleneck. In GTA5, my 2080 Ti stays at 38% GPU - it doesn't even max the clock. That's a 62% headroom remaining. Maybe 3D Vision does something weird with GTA5 -I don't know. If I had the resources, I would love to investigate

This is what GTA5 SLi vs non-SLi looks like at lower resolutions in 2D: virtually no difference:



Due to these revelations, and the fact that there seems to be huge difference between 425.31 and later hacked 3DV drivers, I think it would be prudent if we only benchmark CPUs with systems running 2080 Tis on 425.31 - last official 3DV drivers.

Here is the new graph of my 9900K @5.1/4000CL16 Memory vs 12700K Memory unknown:
Driver 425.31 on 2080Ti:





That's a 29% improvement.

I'll change my results in the other post to mitigate misinformation...

[john105]

Maybe 3D Vision does something weird with GTA5 -I don't know. If I had the resources, I would love to investigate

I remember reading somewhere a long time ago that NVIDIA driver has a lot of SLI-specific code. I guess that both SLI and non-SLI code paths for 2D are optimized enough that the main limiting factor is the CPU. But we all know that the 3D Vision driver has a bottleneck. I think that the SLI 3D code path just has less of a bottleneck in this case. Keep in mind that GTA 5 is not a typical game because it uses 3D Vision Direct instead of Automatic that >99% games rely on for 3D.

[ZeGURU69]

@john105, Your 9900KS results graph are averaging the same as the 12900K, which is 66% faster than our 9700K and 9900K results OC to 5.1GHZ All cores with 4000MHz memory. Are you SURE your 3D Vision was ENGAGED during the benchmark? Thanks.

@ZeGURU69, would you kindly also confirm that 3D Vision was ENGAGED (double image on the screen) when the benchmark was in progress? Also, presumably you're on DDR5?

Thank you.

Yes, 3D Vision was engaged !

[Lysander]

@john105, Your 9900KS results graph are averaging the same as the 12900K, which is 66% faster than our 9700K and 9900K results OC to 5.1GHZ All cores with 4000MHz memory. Are you SURE your 3D Vision was ENGAGED during the benchmark? Thanks.

@ZeGURU69, would you kindly also confirm that 3D Vision was ENGAGED (double image on the screen) when the benchmark was in progress? Also, presumably you're on DDR5?

Thank you.

Yes, 3D Vision was engaged !

Hi, do you have God of War? We were chatting on discord about CPU bottlenecks in that game and were wondering what it looks like on the 12900K, apparently the first boss fight (not the troll but the man that comes to Kratos's house, so you have to play for like an hour or so) gets bad FPS drops on 8700k. The WIP 3D fix for that game is available on this forum.

thx

[Lysander]

I've done some of my own testing on this and these my observations:

Days Gone - largest observed benefit - on 452, during the opening bike chase, the FPS dips to 32, whereas on 425 the lowest is 41; in the first little tent camp you get to, on 452 I get dips to 54 whereas on 425 it's constant 60
The Ascent, Chernobylite, Blair Witch - no observed gains

So looks like UE4 games don't really benefit from 425. Losti observed something different, perhaps his testing was more comprehensive. I didn't run benchmarking software - I simply loaded the game in the same spot/sequence and observed the Afterburner FPS.

[masterotaku]

Days Gone - largest observed benefit
So looks like UE4 games don't really benefit from 425.

Days Gone is a UE4 game :p. Although with a custom made fix unrelated to the universal fix.

[gghocker]

Hi,

as I read this thread and saw the confirmed perfomance improvments with 425.31 over 452.06, I tried to test it by myself without success.

My Setup is (nvidia drivers as mentioned by the 3D Fix Manager):
Win10 2004
Nvidia driver: 452.06
Nvidia 3DV driver: 425.31

I did use the button "Install 425.31" inside 3D Fix Manager.
The Nvidia Setup did popup with the message "this driver cannot be installed on this windows version".

What is the easiest way to install 425.31 in my case?

Best regards

[Lysander]

Days Gone - largest observed benefit
So looks like UE4 games don't really benefit from 425.

Days Gone is a UE4 game :p. Although with a custom made fix unrelated to the universal fix.

Oh yeah, good point =)

[b4thman]

I my opinion (I did not make any comparison about pfs), it is very noticeable the difference in practically every game. I am using 425.31 driver since 2/3 months, even though I have to "repair" (Losti's utility) the driver everytime I turn on the PC and I want to play (otherwise 3D does not work properly). I have tried to use 452.06 a couple of times, and soon I regret and go back to 425.31

It is up to anyone, but I recomend at least to try and compare. Every game I have played so far is perfectly compatible the same way as using 452.06 driver, and there is always an increase of performance using 425.31.

Why a later driver si slower than a previous one? That is something that makes me wonder how planed obsolescence works in this tecnologic world. No surprise if soon later Nvidia decided to turn his back on the people of this community.

[Losti]

Hi,

as I read this thread and saw the confirmed perfomance improvments with 425.31 over 452.06, I tried to test it by myself without success.

My Setup is (nvidia drivers as mentioned by the 3D Fix Manager):
Win10 2004
Nvidia driver: 452.06
Nvidia 3DV driver: 425.31

I did use the button "Install 425.31" inside 3D Fix Manager.
The Nvidia Setup did popup with the message "this driver cannot be installed on this windows version".

What is the easiest way to install 425.31 in my case?

Best regards

Seems you have a DCH System ? Check the driver forum for my driver changer tool, this is also easy and may work for you.

[gghocker]

Thank u,

it did not work, because windows did directly overwrite the nvidia driver again
after installing windows 1809 and nvidia 425.31 I got it running with this setup.

I did not make benschmarks, but I now can run FF7 with 4k 60fps. I know before it did shutter a lot at 4k. So assume it runs better on 425.31.

best regards

[Lysander]

I did not make benschmarks, but I now can run FF7 with 4k 60fps. I know before it did shutter a lot at 4k. So assume it runs better on 425.31.

best regards

4k at 60 in 3d? Interesting, can you put your system specs in your signature?

[Muojo]

Hello,

I upgraded to Windows 10 21H2 and since unable to use 3D on 425.31 after reboot.

I have blocked the update of the drivers and when I restart, I still have the 425.31 drivers but the 3D does not engage anymore. The only solution (which does not always work) is to reinstall the 425.31 drivers.

Anyone have a solution?

Did you use DDU to remove the drivers and then reinstall 425?

[Losti]

Hello,

I upgraded to Windows 10 21H2 and since unable to use 3D on 425.31 after reboot.

I have blocked the update of the drivers and when I restart, I still have the 425.31 drivers but the 3D does not engage anymore. The only solution (which does not always work) is to reinstall the 425.31 drivers.

Anyone have a solution?

Did you use DDU to remove the drivers and then reinstall 425?

Same here. The only solution i have made for this is build in my tool:

viewtopic.php?t=25361

It also reinsalls the driver with the repair function but without restart and over all its more quick than manual installation.

[Lysander]

I think we can safely add Batman: Arkham Knight to that list, too, i.e. it seeing benefits in 425.31. I was testing it quite a bit over the last 2 days and I definitely get lower FPS on 452 than on 425. This game experiences significant frame drops during driving and gliding but on 425, the lowest drop I observed was 48 (and this was only for a split second) whereas on 452, it went down to 38 at times and was below 60 much more than on 425. I don't have graphs, but just observation of FPS in-game while moving around the same areas. Also, I didn't find the benchmark reflective of the gameplay (of course I did not test the entire game), but I made tons of smoke with the batmobile and it didn't drop as much as in the benchmark.

[gghocker]

I did not make benschmarks, but I now can run FF7 with 4k 60fps. I know before it did shutter a lot at 4k. So assume it runs better on 425.31.

best regards

4k at 60 in 3d? Interesting, can you put your system specs in your signature?

i9 11900, 2080ti SLI
but my joy about 4k 60fps was too early, it was 4k 60 fps at the map I was in FF7. Some "level" later at other maps there are no stable 60 fps possible with 4k but at least still 45fps, with which I play, because it depents on the map and where you looking at, so I play 40 till 60 fps at 4k