AMD, Intel and Nvidia driver issues and last recommended version

[Klaus1189]

Welcome!

In this thread you can check for issues with graphic drivers from AMD, Intel and Nvidia.
This thread is meant for use with madVR and its compatible players like Zoom Player, PotPlayer, KMPlayer, MPC-HC and my personal favourite MPC-BE (https://forum.doom9.org/showthread.php?t=165890).

I had to create an external file, because the first two posts are limited and had no more space for new driver releases.

View PDF (https://drive.google.com/file/d/1gQZLfJ0w5KT7tBHCKZASSVU8vTnyPe2n/view?usp=sharing)

But hey, maybe some Driver Dev may stumble over my thread here, please read this:

AMD
Known limitations of Radeon RX 5000 and 6000 GPUs (older generations are fine):
• Hardware deinterlacing is broken in every driver version -> It is poor quality when applied to
interlaced video and it is not adaptive anymore, which is a NO-GO for HTPC usage.
You can check this issue yourself: Download this short MBAFF sample file (https://drive.google.com/file/d/12wO8NJbIrfa2MD4yVaA38RrBBdziCda8/view)
The Hardware Deinterlacer should be aware that the characters above "WWW.D88.COM" are progressive and should not be deinterlaced, otherwise you will see flickering like in this picture (https://upload.wikimedia.org/wikipedia/commons/f/f0/Indian_Head_interlace.gif)
See here (https://forum.doom9.org/showpost.php?p=1928654&postcount=2155) for more details
• Color levels (full 0-255 vs. limited 16-235) are broken. That means limited 16-235 which is the standard range for videos are interpreted as 0-255 and this leads to elevated blacks and dimm whites.
This issue can be best seen on displays which want full range 0-255, like TVs in PC Mode (PC mode for full 4:4:4/RGB support, so you do not use color subsampling like 4:2:2 or 4:2:0)
Also overriding the range for test purposes in MPC-BE (video tab -> Output range) does not take effect.
Dear AMD, if any of your driver programmers read this: Please fix this fundamental video bugs. These are there since RX 5000 series launch.

I just tried the Windows default app "Movies and TV" and this app does not support frame adaptive deinterlacer at all. So best is to download MPC-BE and play the sample file. Windows default app "Movies and TV" really sucks.

Sorry Micro$oft, but with all the skilled and well paid developers you can only achieve this? This is ridiculous.

Intel
iGPU Intel UHD 770 in 12 gen CPU is broken for MPC-BE in use with EVR and EVR custom presenter. You get audio but video is black.
To "solve" the issue is to rename MPC-BE to MPC-HC. Are you serious Intel? Do you really think that is the way of fixing bugs? That is ridiculous for the well paid developers of Intel!
See here for details:
https://sourceforge.net/p/mpcbe/tickets/912/

Nvidia
I am really impressed of what you have done to fix all the HTPC issues of the last 3 years, but I think a little more tweaking of the 23.976 Hz (called 23 Hz, but not 24 Hz!) to get more than 10 hours of no framedrop/repeat like AMD would be nice. I know you can do that.

[Klaus1189]

https://forum.doom9.org/showthread.php?t=176013

[chros]

This old one for nvidia works fine for sure: v385.28

[Charky]

That's an excellent initiative. I've been looking for something like this for a long time since everyone seems to have an opinion on which driver version is the best, some people even using 2 yo+ drivers....

I've been using 417.35 on Win 10 1810 and haven't noticed any issues... NV HDR switches on & off fine, I created a 12bits 2160p23 custom resolution with custom timings CRU and it switches back and forth just fine.

[Klaus1189]

Do you need Fullscreen Exclusive Mode for NV HDR switching? Does it also work with windowed mode?

[Warner306]

Maybe you should highlight the ones that pass HDR correctly in green or some other color?

[Klaus1189]

good idea

[Klaus1189]

any info about 417.71 from 2019.01.15?

[Warner306]

You might have to collect your information from the official support thread. I don't think visibility is too high for other smaller threads.

[Charky]

That's an excellent initiative. I've been looking for something like this for a long time since everyone seems to have an opinion on which driver version is the best, some people even using 2 yo+ drivers....

I've been using 417.35 on Win 10 1810 and haven't noticed any issues... NV HDR switches on & off fine, I created a 12bits 2160p23 custom resolution with custom timings CRU and it switches back and forth just fine.

And now I quote myself...

When I wrote that last post, I hadn't tried my Sony HW40ES projector on my new setup yet.

Turns out, I forgot newest drivers are unable to create custom resolutions suitable for this projector, even with CRU. But they work fine with my OLED LG C8 TV. Go figure...

Anyway, to create a custom res with my projector, without encountering the "the GPU driver rejected this mode, for unknown reasons" error, I used the 385.28 drivers, whick I picked randomly after reading about them in madvr main topic.

Worked like a charm.

[Grimsdyke]

I use my Intel-IGPU system for around 12 months now and I did not noticed any real driver-problems although I also did not specifically looked for them.
Only Full-RGB produced a too dark picture inbetween but this was fixed soon too.
Besides that the performance is so good now that I have decided to not purchase a new board now but wait for ICE LAKE hopefully early next year.

[Asmodian]

The current 417.71 drivers are issue free for me on a RTX 2080 Ti, Win10 1809 17763.253. HDR requires 10-bit output but both D3D11 fullscreen windowed and exclusive work well, with the GPU sending 8-bit RGB HDR to my LG C7P. I can also add a custom mode for 23.976 Hz, optimize it, and madVR will switch to it later.

[chros]

I can also add a custom mode for 23.976 Hz, optimize it, and madVR will switch to it later.
Will it remain there after reboot?

[ashlar42]

Excellent initiative! I'm still currently using a 1080p SDR screen, so there's not much that I can contribute but I'm often thinking about the nightmare I'll face when I'll upgrade to 4K HDR, driver choice included. Thanks for making that bit a little less daunting.

[Charky]

Will it remain there after reboot?
On my setup, same drivers, it does.

[Asmodian]

Will it remain there after reboot?

It does! :D

[iSeries]

It does! :D

Is madVR able to create the custom timing or does CRU have to be used to get 12bit?

[Asmodian]

8-bit only. I don't use anything higher, my TV does not like >8-bit input. :p

[iSeries]

Mine neither lol but was just curious. Thanks:-)

[Oguignant]

wtf, none of these works with my 1080ti + win10 1803 + LG OLED TV

Nvidia GeForce Game Ready Driver

417.71 2019-01-15 HDR passthrough OK see here
417.35 2018-12-12 HDR passthrough OK
417.22 2018-12-03
417.01 2018-11-16
416.94 2018-11-13

[Oguignant]

wtf, none of these works with my 1080ti + win10 1803 + LG OLED TV

Nvidia GeForce Game Ready Driver

417.71 2019-01-15 HDR passthrough OK see here
417.35 2018-12-12 HDR passthrough OK
417.22 2018-12-03
417.01 2018-11-16
416.94 2018-11-13

eureka! 416.81 Work fine

[Manni]

Just to report that 417.71 doesn't work in passthrough with 1080ti, at least in 4K23 12bits RGB full (custom 23p res with CRU). It sends bogus metadata. See https://forum.doom9.org/showpost.php?p=1864303&postcount=54551

You might need a different table for GTX and RTX, and possibly for bit depth (8 vs 12), as drivers don't seem to behave the same.

[Klaus1189]

Anyone tested 418.81? Please let me also know which card, like GTX / RTX and OS version

[Manni]

Tested on Win 10 x64 build 1809, GTX 1080ti. HDR passthrough still broken (wrong metadata) but everything else seems to be working.

12bits at 4K23 RGB Full survives a reboot as long as it's your default refresh rate.

[Klaus1189]

Which latest driver version of Nvidia has working HDR passthrough for users who use that.

[iSeries]

Which latest driver version of Nvidia has working HDR passthrough for users who use that.

398.11 is the last working driver for HDR passthrough, anything later has the bogus metadata issue.

Should add that later drivers will switch into HDR fine, and displays will switch into HDR, however any driver later than 398.11 is sending bogus metadata to the display.

[brazen1]

I'm using 416.81. It switches HDR mode including at panel level (auto increases backlight, etc.) with no problems and behaves with 3D stereoscopic switching perfectly as well. This includes the resolution change bug from 2160p to 1080p and vice versa postage stamp issue many experience. I do edit the installer to use 388.59 HD Audio driver because every audio driver since has multichannel problems under certain conditions. I've tried every driver release after these and none of them passthrough HDR to all players and the audio problem persists. DVDFab Media Player v.3 to be exact. Using madVR compatible players like MPC-HC/BE were also hit and miss until using this combination. A good fallback if you don't want to edit the installer is 385.28, an older driver. The advantage to using the newer combination is frame drops that you can count in madVR's OSD. For most that do not use a reclock method of some sort, you will get a frame drop every 4.37 mins. This has always been the norm for nVidia. The improvement using this newer combo is the drops decrease to one every 13 mins or so. This improvement is only using 3D though. 2D still drops one every 4.37 mins. Many are using newer drivers and reporting no problems except they are confined to very limited use with very specific caveats. This combo is much more diverse imo.

[Zetti]

418.91 is released.

https://us.download.nvidia.com/Windows/418.91/418.91-win10-win8-win7-desktop-release-notes.pdf

[Warner306]

Maybe you should add this link next to the description for 398.11 to explain what this is the last working driver: https://forum.doom9.org/showthread.php?p=1865444#post1865444

[Klaus1189]

Thanks, added.

[Klaus1189]

Is HDR passthrough (https://forum.doom9.org/showthread.php?p=1865444#post1865444) fixed in 419.17?

[huhn]

can someone test WFS 10 bit for banding especially at 12 bit or even better 10 bit RGB output on nvidia?

you can test it with this image: http://www.bealecorner.org/red/test-patterns/Gradient-16bit.png

it's not easy to do but there is a trick that will help send 12 bit RGB (a really good 1080p screen is mostly better) use MPC hc/be use WFS 10 bit with the image and and now right click.
when you right click you should see heavy banding when it switches to 8 bit output and shortly after that it should disappear and should have less banding or no banding in 8 bit WFS mode so it should have less banding then WFS 10. FSE 10 bit should show less banding too.
tested on 419.17 if this is a GPU driver bug everyone should be affected.


capturing the image should be the easiest way to show it but that's not really possible.

[Manni]

How do you set 10bits RGB out on nVidia? The only options at least here on a 4K display are 8bits and 12bits.

The display must support 12bits natively in order for 10bits content to show without banding when 12bits is selected.

Recent JVC projectors support 12bits from the input to the panels, but not many displays do. Most support only 8bits or less, a few support 10bits, very few actually support 12bits natively, even if they accept a 12bits input.

So if you send 12bits to a panel that doesn't support 12bits, or only supports a 12bits input but then dithers to the panels because they are only 10 or 8bits, then you get this minor banding you are seeing sending 12bits to your display.

Are you 100% sure that your display supports 12bits from the input to the panels, through the whole internal chain, the way recent JVC projectors do?

I don't see the point of the test if you're not asking people to test on native 12bits displays.

[Manni]

Is HDR passthrough (https://forum.doom9.org/showthread.php?p=1865444#post1865444) fixed in 419.17?

No it's not, at least not on my 1080ti

[chros]

can someone test WFS 10 bit for banding especially at 12 bit or even better 10 bit RGB output on nvidia?
I can test it this evening but with the old driver (v385.28) set to 12 bits (no 10bits) and madvr set to 10bits. My 1080p TV (8bit+FRC) accepts 10 bit signal, that's how I use it.

[huhn]

How do you set 10bits RGB out on nVidia? The only options at least here on a 4K display are 8bits and 12bits.
displayport monitors.
10 bit PC dispalys are a thing for a very very long time now.

The display must support 12bits natively in order for 10bits content to show without banding when 12bits is selected.
which display even does processing at less then 12 bit?
Recent JVC projectors support 12bits from the input to the panels, but not many displays do. Most support only 8bits or less, a few support 10bits, very few actually support 12bits natively, even if they accept a 12bits input.
not sure why this would matter or why you think that's the case.
So if you send 12bits to a panel that doesn't support 12bits, or only supports a 12bits input but then dithers to the panels because they are only 10 or 8bits, then you get this minor banding you are seeing sending 12bits to your display.
dithering will not produce banding that's the point of dithering.
Are you 100% sure that your display supports 12bits from the input to the panels, through the whole internal chain, the way recent JVC projectors do?
my TV is labelled as a 10 bit 120 HZ panel (orion if i remember corretly) i'm pretty sure it's 8 bit +HI FRC
there are no 12 bit displays even the most professional displays normal user can buy are 10 bit and i'm talking about grading displays here these often have 12 bit hardware LUT or better so sending 12 bit is clearly intended.
12 bit input support is a thing since for ever.

so even if JVC supports 12 bit presentation it doesn't matter even 8 bit can trick human eyes and measuring device to thing is 12 bit plus. and that why the only way to figure out if a end device (not a display anymore we talk about projector now) can do 12 bit is checking the part if it supports 12 bit direct input from the processor words from a manufacturer are not enough.
checking for 12 bit input doesn't tell you anything and using measuring device can't figure it out and the human eye well it'S the human eye.

if your logic of dithering 12 to 10 will create banding then madVR is the biggest banding producing program ever made by using 16/32 bit to 10 bit dithering i'm pretty sure we both can agree that'S not the case.

BTW there where devices that clearly accepted 12 bit for the panel(you can check the parts) and these devices biggest short coming was bit deep related. i'm talking about Panasonic plasmas they often support 12 bit for the panel which is pretty worthless thanks to the noise level well i guess this is faster/cheaper then dithering before which was not a huge thing back then.
I don't see the point of the test if you're not asking people to test on native 12bits displays.
as you said before nvdia doesn't like the idea of sending 10 bit using HDMI. so we do it with 12.
and as 10 bit FSE doesn't show banding but WFS for me i clearly see a problem in WFS 10 bit.

or with other word i can easily see the difference between 10 bit dithering and 10 bit no dithering which makes it very clear that my TV can show 1024 steps as a 10 bit panel would be able to do

[chros]

can someone test WFS 10 bit for banding especially at 12 bit or even better 10 bit RGB output on nvidia?

you can test it with this image: http://www.bealecorner.org/red/test-patterns/Gradient-16bit.png

it's not easy to do but there is a trick that will help send 12 bit RGB (a really good 1080p screen is mostly better) use MPC hc/be use WFS 10 bit with the image and and now right click.
when you right click you should see heavy banding when it switches to 8 bit output and shortly after that it should disappear and should have less banding or no banding in 8 bit WFS mode so it should have less banding then WFS 10. FSE 10 bit should show less banding too.
tested on 419.17 if this is a GPU driver bug everyone should be affected.

Good find, @huhn! I could recreate the issue. (see my signature for more details on the setup I use)
- 1080p 10 bit (8bit+FRC) TV
- nvidia v385.28 driver is set to Full range, RGB, 12 bit
- MadVR is set to:
-- "10 bit or higher"
-- "Ordered dithering" is On!

It didn't matter whether I set dithering to None or Ordered!

FSE 10 bit should show less banding too.
I couldn't really tell the difference in banding between FSE and WFS, I could clearly see it in both.

and as 10 bit FSE doesn't show banding but WFS for me i clearly see a problem in WFS 10 bit.
It does for me, see above.

Interesting thing is that using the same TV with my old laptop (optimus system, with intel driver + win8.1) I didn't have this issue with 10bit FSE.

[ChaosKing]

Here's also a good tool if you don't want to install too much crap which are bundled nowadays in nvidia drivers such es telemetry https://www.techpowerup.com/forums/threads/nvcleanstall-clean-installer-for-nvidia-drivers-alpha.249085/
Remove your old driver with DDU first https://www.techpowerup.com/forums/threads/nvcleanstall-clean-installer-for-nvidia-drivers-alpha.249085/

[chros]

Here's also a good tool if you don't want to install too much crap which are bundled nowadays in nvidia drivers such es telemetry https://www.techpowerup.com/forums/threads/nvcleanstall-clean-installer-for-nvidia-drivers-alpha.249085/
Remove your old driver with DDU first https://www.techpowerup.com/forums/threads/nvcleanstall-clean-installer-for-nvidia-drivers-alpha.249085/
Interesting, thanks!

[huhn]

if you see it on both FSE and WFS you should assume it is the TV.

[Zetti]

419.35 is released.

https://us.download.nvidia.com/Windows/419.35/419.35-win10-win8-win7-desktop-release-notes.pdf

[SamuriHL]

They need to fix the HDR metadata issue one of these days. Those of us that do other things besides watch movies really could use a fixed updated driver. And those poor (after spending the kind of money they did :D) RTX 20x0 owners can't really use older drivers anyway. I have not checked the latest driver to see if the metadata has been fixed. Anyone have time to do so?

[VBB]

I think Manni is the only one with the means to test it. Too bad Nvidia doesn't care about HTPC users whatsoever...

[SamuriHL]

We're certainly not a priority for them, no. Hopefully Manni will get a minute to check. I'm not super hopeful given that it's been broken for quite some time now.

[Manni]

We're certainly not a priority for them, no. Hopefully Manni will get a minute to check. I'm not super hopeful given that it's been broken for quite some time now.

Madshi has reported the issue to his contact at nVidia, so I'm hoping it will get fixed at some point. I'll try to test the latest version later today and I'll report back. I had recently reverted to 385.28 because of this, as I needed to run some tests in passthrough and the recent drivers are not usable for this. It should be a priority for them!

[nevcairiel]

Chances for non-criticial things like that to get fixed are higher in a new major driver branch then a point release. But who knows.
I still think it could help to actually tell the OS about the metadata instead of hoping the driver just ignores the OS.

If I had a device to test this, I would try to proof my idea with some code, but alas I do not, so..

[SamuriHL]

Tell Jim to expense you one. :D LMAO I would be very curious what the results of such a test show cause I have a feeling you could be right. I think nVidia changed this back when Windows added support for HDR. Things haven't been quite right since.

[Manni]

Chances for non-criticial things like that to get fixed are higher in a new major driver branch then a point release. But who knows.
I still think it could help to actually tell the OS about the metadata instead of hoping the driver just ignores the OS.

If I had a device to test this, I would try to proof my idea with some code, but alas I do not, so..

Please contact me by PM/email if you wish, I can send you a HD Fury Vertex :)

[Manni]

We're certainly not a priority for them, no. Hopefully Manni will get a minute to check. I'm not super hopeful given that it's been broken for quite some time now.

Okay I tested 419.35, and either there is a bug in the latest madVR build, or the new driver is even worse. It's not that it passes bogus metadata, it doesn't switch to HDR anymore :(

The OSD reports NV HDR and HDR passthrough, but the GPU isn't sending HDR, so no HDR metadata reported at all...

Back to 385.28, though I might try 397.31 to see one more time if I can get better 3D and most of everything else, but last time I tried I had to go back to 385.28 anyway.

[SamuriHL]

FFS they broke THAT again? I give up...thanks for testing.