AMD, Intel and Nvidia driver issues and last recommended version

[SamuriHL]

Curious if the drivers expose any additional functionality over the 2xxx cards. I don't think we've yet seen anyone get their hands on a card so it'll be nice to see if the drivers are stable for them, what functionality hdmi 2.1 exposes in the driver if any, and the performance in madvr. Can't wait to see that kind of info.

Sent from my SM-G975U using Tapatalk

[Manni]

I've only tested 8bits RGB with 456.38 (Studio / Standard) as that's what I use and don't have the time to do any other tests at the moment, but at least with that I can confirm that HDR works fine (with nvAPI) and that if using passthrough the HDR10 metadata is sent properly.

I don't know how it behaves in 12bits or in YCC.

[EDIT: Also I'm still using CRU so I haven't tested native frame rate.]

I don't have the issue of the "double switch" for HDR. It switches seamlessly between SDR and HDR, or SDR BT2020 and HDR passthrough.

So at least at first sight, it seems good to go :)

I'll post if I find any issue.

My full config is in my sig, YMMV.

I experienced issues with this driver and the following ones. It now outputs weird null HDR metadata and I can't get auto SDR tonemappint without forcing passthrough first then going to SDR tonemapping.

I tried most drivers before that and had to revert all the way back to 442.74 to get predictable, reliable HDR<>SDR auto switch with NV API in madVR. Even 442.92 had some issues.

Not sure why the more recent drivers stopped working, maybe an OS update?

This is with 1909 and a GTX 1080ti (full config in sig). YMMV.

[chros]

Even 442.92 had some issues.
Not sure why the more recent drivers stopped working, maybe an OS update?
This is with 1909
Like this (https://forum.doom9.org/showthread.php?p=1922074#post1922074)? I'm on 1809, so it seems it's not the newer windows.
And what about this (https://forum.doom9.org/showthread.php?p=1921810#post1921810) audio issue?

[Manni]

Like this (https://forum.doom9.org/showthread.php?p=1922074#post1922074)? I'm on 1809, so it seems it's not the newer windows.
And what about this (https://forum.doom9.org/showthread.php?p=1921810#post1921810) audio issue?

I can’t remember exactly, I think it’s more that when you start an HDR clip, by default it is outputting HDR passthrough here, and you have to manually switch to SDR tonemapped. Unlike some of the newer drivers, you can get the right mode manually, while some of the newer drivers are stuck in this weird NULL hdr metadata mode that prevents the right mode from being selected (I use a Maestro to switch calibration automaticallly, so I need no HDR metadata for my SDR tonemapped calibration to be applied.

I don’t have these audio issues but I use HDMI bitstream as I need it for immersive sound.

Bottom line is that here, with 1909 and a GTX 1080ti, only 442.74 or earlier works fully as expected (and I’ve only tested RGB 8bits).

I use CRU so the better refresh rate of the new drivers doesn’t make a difference.

[Klaus1189]

That's a bummer. Please let me know if I should update the Nvidia section.
If you have a spare SSD laying around, would be nice if you could set up Windows 10 2004 fresh on it and test the new driver. Would be interesting if the driver behaves the same on 2004. You can just put the old one back in the pc and the pc will work as nothing would had happend. I did this before on my own.
I know it will be some time, but what what do we all for testing our stuff ;-)

[Manni]

That's a bummer. Please let me know if I should update the Nvidia section.
If you have a spare SSD laying around, would be nice if you could set up Windows 10 2004 fresh on it and test the new driver. Would be interesting if the driver behaves the same on 2004. You can just put the old one back in the pc and the pc will work as nothing would had happend. I did this before on my own.
I know it will be some time, but what what do we all for testing our stuff ;-)

I don't have the time to use another drive at the moment, but I'm probably going to cancel my 3090 order as it looks like their ETA of the 5th was BS and I'm close to number 100 in a waiting list that doesn't move. Plus, they have already charged my VISA! This is SCAN in the UK, and I'm not impressed by the way they handle this. If you don't know when a product ships, don't charge the customer!

So I think I'm going to cancel next week and wait until we know more about the 3080ti as it might be on TSMC 7nm according to rumours, which might make it a bit less power hungry than the 3090. I don't really believe they'll move the already released card to 7nm, but that's still a slight possibility.

This means that I might attempt to upgrade the OS anyway, but it's unlikely I'll find the time until early November due to work. I'll let you know as soon as I can test this. I image my drives before any big change, so I can always go back if there are more issues with 2004 and no improvement.

For now, I would update the nVidia session and mark 442.74 as last known fully working driver re HDR for GTX and 1909 or previous, at least if your video chain detects content and automatically switches to the correct calibration (HD Fury Vertex/Diva/Maestro or new JVCs). Unless anyone confirms a more recent driver works, we can't rule out something in my config for course.

[SamuriHL]

I'll check later this morning to see if they broke it again. They had fixed that in one of these recent driver versions. I'll need to find a working hdmi cable to hook the integral back up.

Sent from my SM-G975U using Tapatalk

[Manni]

I'll check later this morning to see if they broke it again. They had fixed that in one of these recent driver versions. I'll need to find a working hdmi cable to hook the integral back up.

Sent from my SM-G975U using Tapatalk

You have an RTX GPU, my report is only about GTX (1080ti).

I don't think the issue is the same with more recent GPUs (or maybe a more recent OS).

[SamuriHL]

You have an RTX GPU, my report is only about GTX (1080ti).

I don't think the issue is the same with more recent GPUs (or maybe a more recent OS).

I can't imagine it'd be different between GTX and RTX on the HDR metadata. I could be wrong I suppose but I've not checked the last 2 driver versions since they fixed it. Still trying to hunt down a working HDMI cable in the house somewhere but I'll check anyway. Then we'll know for sure if it's different.

EDIT: Found a working cable. Yes, on build 19042 of Windows 10 I am indeed getting proper metadata with 456.71. So there does appear to be a difference after all.

[chros]

456.71 is performing better than I anticipated.
The audio problem no longer exists! (Stereo vs Multichannel dependent on AVR On/Off state). This was problematic for years. I no longer have to modify the installation with an older HDAudio driver.

Bottom line is that here, with 1909 and a GTX 1080ti, only 442.74 or earlier works fully as expected (and I’ve only tested RGB 8bits).

mark 442.74 as last known fully working driver re HDR for GTX and 1909 or previous, at least if your video chain detects content and automatically switches to the correct calibration (HD Fury Vertex/Diva/Maestro or new JVCs).

OK, thanks, I went back as well to 442.74 + hdmi audio from 456.71 (by copying over "HDAudio" directory), I'll report back in a week if all is good here.

I noticed 1 issue (?) straight away, that I haven't before (not even with 442.92 SD):
- 12bit nv output
- madvr OSD is enabled
- mpc-be/hc fullscreen window: madvr reports 10bit
- but when seekbar comes up at the bottom: madvr reports only 8bit!

I don't remember whether it worked like this or not previously. :)

I don’t have these audio issues but I use HDMI bitstream as I need it for immersive sound.

Cheers, then I guess it's my Win10 version :)
The issue with HDMI audio is:
- you have setup speaker config (e.g. to 5.1) and properties (e.g. to 41kHz) of NV HDMI device in Windows (under Sound in CP)
- if you click on Configure again and cancel the dialog (or hit escape) then it reverts back to default (!) stereo 48kHz config :D

This doesn't happen when you do the same with the Properties dialog.

[Manni]

As I said, I only tested 8bits RGB :)

[SamuriHL]

I'm only using RGB 8 bit myself these days.

[Manni]

I'm only using RGB 8 bit myself these days.

I was replying to Chros :)

[SamuriHL]

I know you were. I'm just agreeing it's the way to go. :D

[chros]

I disagree, 12bit is the ultimate solution :D

[Manni]

I had to make an image of my SSD so I thought I'd give a quick try to 2004.

Bad news: the behaviour is the same with my 1080ti. 456.71 causes a fake HDR mode that prevents the SDR calibration to be selected.

456.71 was installed after the update, so I tried a full clean install with CRU, same results.

I had to revert to 442.74 to get the expected behaviour.

So the OS has nothing to do with this.

The good news (for me) is that the OS update finally resolved my super long start for the nVidia cPanel. I'm back to normal start time of a few secs.

So unless I find some unseen downsides, it looks like I'm going to stay on 2004. I'll update the thread if it wasn't the case.

Again, this is only for GTX as I haven't tested the new drivers with GTX.

I have also only tested RGB 8bits :)

[ryrynz]

The often unnecessary solution you mean.

[Manni]

The often unnecessary solution you mean.

No idea what you mean.

[SamuriHL]

That's seriously crazy that you have such wildly different HDR results on the 1080ti than I do on my RTX. What the hell is nVidia doing lately.....sigh

[ryrynz]

No idea what you mean.Was a reply to chros [emoji6]

Manni did you use madVRs calculated values with CRU for 24hz?

[JonnyVee]

Has anyone tested a 3080 or 3090 GPU yet with HDR - SDR? I can only get MPC-BE (1.5.5.5433) with its internal filters to convert HDR properly.

If I use anything with LAV filters such as MPC-HC or kodi DSplayer I get extreme blown highlights when in Fullscreen windowed and FSE. However, it’s fine when windowed. This is with the older internal LAV and newer 0.74.1 via external.

The attached pictures show the (1] excess highlights (not sure if this is the correct term) when fullscreen, (2) correction to normal when windowed, and (3) the correct conversion using MPC-BE’s internal filters.

This happens with MPC-BE as well when I use LAV External with 0.74.1. However, its fine with the included internal filters.

I’m using an EVGA 3080 FTW3 Ultra and this occurs with 456.38 and 456.71 drivers. I’ve also tried 8 and 10 bit, and RGB &YCbCr.

[Manni]

That's seriously crazy that you have such wildly different HDR results on the 1080ti than I do on my RTX. What the hell is nVidia doing lately.....sigh

Not really, it's a different architecture and most probably different drivers internally, even if the package is the same. Also they tend to spend more testing time on the more recent architecture with recent drivers, so it makes sense. There is often a point where it's preferable to stop updating, as only the new gens are improved, the older ones can get worse.

In the latest drivers, the 1080 GTX gen are the last supported.

Was a reply to chros [emoji6]

Manni did you use madVRs calculated values with CRU for 24hz?

Ah sorry, that makes sense :)

No I use my own values for the JVC rs2000. madVR has stopped working a long time ago re refresh rate optimisation here.

I get no drop/repeat for more than 3 hours or more (often no drop/repeat after a while) with 442.74 + CRU, so I'm a happy camper. :)

It's really nice to have the nVidial cPanel working normally again. Something had messed it up (I think one of the "bring 3D back" packages) and nothing could bring it back.

If there is no downside, that alone was worth the upgrade to 2004.

By the way, even with 442.74 I occasionally get the wrong HDR mode, but if I select HDR passthrough and then SDR pixel shader, I get the correct mode, contrary to the newest drivers that stay stuck in the wierg HDR null metadata mode.

[SamuriHL]

Not really, it's a different architecture and most probably different drivers internally, even if the package is the same. Also they tend to spend more testing time on the more recent architecture with recent drivers, so it makes sense. There is often a point where it's preferable to stop updating, as only the new gens are improved, the older ones can get worse.

In the latest drivers, the 1080 GTX gen are the last supported.


I suppose there is that but it's not like the 10xx cards are that old. Good god.

[chros]

By the way, even with 442.74 I occasionally get the wrong HDR mode, but if I select HDR passthrough and then SDR pixel shader, I get the correct mode, contrary to the newest drivers that stay stuck in the wierg HDR null metadata mode.
Really?! I just spent on it 2 hours this morning :) Bloody hell, what a mess :D Thanks for reporting, I'll keep an eye on it then.
Maybe we should go way back to 430.53... Unfortunately I can't easily test HDR metadata, if I could then I'd go back and try bunch of them.

@JonnyVee, upload the images to an external image host and link them here.

[VBB]

Manni is correct as far as support when it comes to the Nvidia Studio Drivers. The Game Ready Drivers still offer full support all the way back to the 600-series:

"Effective April 12, 2018, Game Ready Driver upgrades, including performance enhancements, new features, and bug fixes, will be available only for desktop Kepler, Maxwell, Pascal, Volta, Turing and Ampere-series, as well as for systems utilizing mobile Maxwell, Pascal, and Turing-series GPUs for notebooks."

[SamuriHL]

That's right I forgot he's using studio drivers. I've been using game ready lately cause I'm getting ready for Cyberpunk 2077.

[blu3wh0]

So I just upgraded from my 1080 Ti to a 3080 and finally got around to testing MadVR. Good news, the issues with HDR on my 1080 Ti where I had to open the file twice to get HDR properly working has been resolved, although it only seems to work in fullscreen and any opening of the menu or looking at the status bar breaks it, but at least it switches really quickly. However, the card introduced really annoying micro stutters when playing at 23Hz mode (fullscreen windowed mode, in normal windowed it's fine). The stutter becomes significantly worse if you just move the mouse or start bringing up the menu. Switching to 60Hz with smooth motion does seem to mostly remove the problem, but I think it still stutters a bit. At 100Hz it's no longer noticeable to me. I believe the solution previous brought up for the 20 series for a similar problem was to set present ahead frames to 1, and this appears to have alleviated the problem. For now, this setting and watching at 60Hz instead is a good solution, but this really should not be a problem. It seems like it's always one step forward and two steps back with these things...

[JonnyVee]

Really?! I just spent on it 2 hours this morning :) Bloody hell, what a mess :D Thanks for reporting, I'll keep an eye on it then.
Maybe we should go way back to 430.53... Unfortunately I can't easily test HDR metadata, if I could then I'd go back and try bunch of them.

@JonnyVee, upload the images to an external image host and link them here.

Here’s links to the full size images

https://imgur.com/e7S8uoQ

https://imgur.com/Fucko6q

https://imgur.com/pWeCDJ6

[ryrynz]

I suppose there is that but it's not like the 10xx cards are that old. Good god.

True but he has a point. Every update has that possibility to potentially break things that haven't been tested and given these changes are generally made for newer cards there's likely greater chance for breakage of functionality for older cards. Still on 1809 and 385.28 for exactly this reason, it just works.

[SamuriHL]

Sure but the alternative point of view is that it should be a very well known quantity at this point. But I get it.

[chros]

Still on 1809 and 385.28 for exactly this reason, it just works.
Really? Do you have multimonitors, e. g. TV + monitor? It didn't work here e.g. after sleep, I got blank screen on monitor if video played only on TV. This was flawless with 1607.
That's why I upgraded the driver when SamuriHL stated that 442.92 is good again.

[el Filou]

I went back as well to 442.74 [...] I noticed 1 issue (?) straight away, that I haven't before (not even with 442.92 SD):
- 12bit nv output
- madvr OSD is enabled
- mpc-be/hc fullscreen window: madvr reports 10bit
- but when seekbar comes up at the bottom: madvr reports only 8bit!I think it's normal behaviour because 10bit only works in fullscreen windowed, and once you display GUI elements it swtiches to windowed instead of fullscreen windowed. At least that was the case in 2018: https://forum.doom9.org/showpost.php?p=1838798&postcount=50141

[Klaus1189]

I think it will take some time till I get one ...
https://www.proshop.de/RTX-30series-overview

And I am still open for RDNA2

[SamuriHL]

I'll continue to wait until the cards actually exist in the real world. LOL

[chros]

Really?! I just spent on it 2 hours this morning :) Bloody hell, what a mess :D Thanks for reporting, I'll keep an eye on it then.
Maybe we should go way back to 430.53... Unfortunately I can't easily test HDR metadata, if I could then I'd go back and try bunch of them.
Yep, after a week without reboot the 442.74 is shite as well :)
Although I didn't have the SDR in HDR issue as before, but madvr doesn't change the metadata for sure when modifying the display peak nits (checked with the second code value (https://forum.doom9.org/showthread.php?p=1882713#post1882713) on B8).
So, what's next? :)

True but he has a point. Every update has that possibility to potentially break things that haven't been tested and given these changes are generally made for newer cards there's likely greater chance for breakage of functionality for older cards. Still on 1809 and 385.28 for exactly this reason, it just works.
You forgot to answer to the question below, thanks.
Really? Do you have multimonitors, e. g. TV + monitor? It didn't work here e.g. after sleep, I got blank screen on monitor if video played only on TV. This was flawless with 1607.

[Klaus1189]

Yep, after a week without reboot the 442.74 is shite as well :)
Although I didn't have the SDR in HDR issue as before, but madvr doesn't change the metadata for sure when modifying the display peak nits (checked with the second code value (https://forum.doom9.org/showthread.php?p=1882713#post1882713) on B8).
So, what's next? :)

What Nvidia driver is recommended now? :scared:

[SamuriHL]

I'm still recommending 456.71 which has proper metadata, good clocks, and a good audio driver. But I know not everyone agrees with that.

EDIT: FOR RTX CARDS cause who cares about GTX. :p :D LOLOLOLOL

[chros]

How would I know? :D That's what we try to find out...

[Manni]

I'm still recommending 456.71 which has proper metadata, good clocks, and a good audio driver. But I know not everyone agrees with that.

You can’t make the same recommendation for GTX and RTX...

[SamuriHL]

You can’t make the same recommendation for GTX and RTX...

Then Klaus1189 needs to understand that and will have to split the list. Which makes maintaining it slightly more complicated. I'm not saying I disagree, just that it needs to be well understood that RTX and GTX cards should be treated as completely different.

[Klaus1189]

That's no problem, just give the info and I update the Nvidia section ;)

[SamuriHL]

You see my recommendation for RTX. Manni and chros (or others) will have to give a definitive recommendation for GTX, as manni is very correct, they are not working right with the latest driver builds.

[Manni]

That's no problem, just give the info and I update the Nvidia section ;)

You have already added a note for 442.74, but 1909 is not related as I installed 2004 and confirmed that past 442.74 the issues were similar with 2004.

So my most recent driver recommendation for GTX is 442.72, but it works a little bit better with 1909 than with 2004.

However, I made it clear that it was my recommendation when using a HD Fury to select the correct calibration according to content for a JVC. Others with different needs or using different hardware might have different results and might be able to use different drivers.

HTPCs are so complex that it’s really hard to make blanket recommendations.

For now, I’m on 442.72 on 2004, but this might change as I have little time for testing.

I’m also hoping that I’ll be able to upgrade to RTX and use a more recent driver without having to do more tests with GTX :)

[SamuriHL]

I’m also hoping that I’ll be able to upgrade to RTX and use a more recent driver without having to do more tests with GTX :)

:D Most excellent. LOL

[huhn]

isn't RTX and GTX very misleading? isn't this more about turing and pascal. i mean there is the 1600 series and i have a hard time believing they behaviour different then the "RTX" cards. and who knows what't wrong with ampare in theory it exists but that's it for now.

[SamuriHL]

I don't know. Someone with a GTX 16xx card needs to chime in on that one. It's very possible it's a pascal vs turing thing but we need someone to test that theory.

[ryrynz]

Really? Do you have multimonitors


Just a TV, sleep functions fine here.
Waiting for someone to say they can get a 30 series running perfectly with no frame drops/repeats for almost 3 hours by whatever means.
Maybe next year.. Along with a new madVR version.. :eek:

[SamuriHL]

Wouldn't that all be nice. I'm wanting to do that myself but can't get a hold of a 3080. I think next year is probably realistic for the earliest on both of those wish list items. Let's hope they don't break the damn drivers in the meantime.

[glc650]

I don't know. Someone with a GTX 16xx card needs to chime in on that one. It's very possible it's a pascal vs turing thing but we need someone to test that theory.I have a 1660 and was stuck on 442.74 for the longest time due to the FS windowed 10 bit HDR issues documented on this site. I haven't noticed any issues with latest Windows 10 and latest nVidia drivers. Latest drivers (even before latest Windows 10) increased time before frames repeated/drop on 23/24p content (which doesn't matter to me since I use madVR to optimize my resolutions with most of them being fully optimized) and fixed the audio issue. And I like that mpc-hc now leaves HDR when pausing/seeking/right click menu - much easier on the eyes that way. I have a Sony 75Z9F and Marantz SR1812. My 1660 is connected directly to both (HDMI to TV and HDMI (via DP to HDMI adapter) to AVR) because I've had issues passing all resolutions/frame rates/etc through my AVR.

[Manni]

I have a 1660 and was stuck on 442.74 for the longest time due to the FS windowed 10 bit HDR issues documented on this site. I haven't noticed any issues with latest Windows 10 and latest nVidia drivers. Latest drivers (even before latest Windows 10) increased time before frames repeated/drop on 23/24p content (which doesn't matter to me since I use madVR to optimize my resolutions with most of them being fully optimized) and fixed the audio issue. And I like that mpc-hc now leaves HDR when pausing/seeking/right click menu - much easier on the eyes that way. I have a Sony 75Z9F and Marantz SR1812. My 1660 is connected directly to both (HDMI to TV and HDMI (via DP to HDMI adapter) to AVR) because I've had issues passing all resolutions/frame rates/etc through my AVR.

Thanks for the feedback, but are you using HDR passthrough or pixel shader tonemapping?

In any case yes I would expect the issue I experience with recent drivers to be a pascal vs turing issue (didn't think nVidia was stupid enough to call a pascal card RTX or a Turing card GTX, but I guess in the 16xx case they disabled raytracing, is that why?).

Anyway, there shouldn't be any issue with Pascal if you use HDR passthrough because that's the issue I have with my GTX 1080ti: HDR is incorrectly detectly by my video chain despite using pixel shaders (hence sending SDR content) because the driver sends bogus HDR metadata along with the tonemapped SDR content. I have to select HDR passthrough and then pixel shaders again to get the correct mode and calibration. It's only two buttons on the iPad, but still, it's annoying.

This almost never happens with 472.74 on 1909, it happens occasionally with 472.74 and 2004, but I can toggle manually as indicated above when it does.

Although the wrong mode is almost always selected by default with more recent drivers due to the bogus metadata, I can still toggle manually with some of them, but the most recent ones are stuck in HDR due to the bogus metadata and no amount of toggling will bring back SDR content detection (i.e. get the driver to stop sending the bogus HDR metadata). They keep sending bogus HDR metadata that force the HDR calibration to be selected. Hence why there is no problem if you use HDR passthrough, as this is the calibration you want anyway. You might even get the correct HDR metadata when using HDR passthrough.

Also, if your screen/display doesnt change calibration automatically according to content, either on its own or using an HD Fury device, then you might not notice any problem either.

I had tried to qualify this before, but as always with HTPCs you can't make generalisations...

@Klaus1189, maybe you could put a link to this post with your note that 472.74 *might* be the last driver to use for Pascal cards for pixel shader tonemapping (unless it's confirmed it's a GTX vs RTX issue, which is unlikely) with displays/video chains that automatically detect the content to select a calibration automatically, if you want to put such a longwinded warning. I honestly don't test drivers for all use cases, it's time-consuming enough to test them for my own use :).