AMD, Intel and Nvidia driver issues and last recommended version

[huhn]

3D LUT gamut mapping.

it's "important" for PJ where some devices trade CR for gamut.
currently AMD choice the gamut as they wishes which can work but... with nvidia you can force BT 2020 under any circumstances which is quite a nice feature for calibration.

the real bad things about AMD currently are deint which is a joke but no one laughs.
the just bad HDR api which only works with 10 bit which is just dumb and needs FS if i recall correctly...

taking information from mclingo and the current user in the madVR thread there could be a horrific issue in the API.
i didn't had a HDR TV in the past just a 4K TV so i couldn't test it. BTW. my old TV had a better CR in BT 2020 mode i didn't test the new one yet waiting for some serious firmware fixes first before i touch calibration again.

[DMU]

the real bad things about AMD currently are deint which is a joke but no one laughs.
the just bad HDR api which only works with 10 bit which is just dumb and needs FS if i recall correctly...

All of the above can be applied to the Microsoft (WinAPI) HDR API. HDR only with 10 bits and in fullscreen. And given what is happening with the Nvidia drivers, they will soon come to this too.

[huhn]

win API clearly works in 8 bit windowed and so on so what do you mean?

[DMU]

This is not true. Are we talking about HDR API? Here's an example (https://github.com/microsoft/DirectX-Graphics-Samples/tree/master/Samples/Desktop/D3D12HDR) from Microsoft. Try to get HDR with true metadata on it at 8 bits and not in fullscreen.

[VBB]

Huhn, curious which new TV you ended up getting...

[huhn]

well as long as you are in window mode the HDR meta data would be wrong the file is not the only thing displayed and in the end no proper screen cares about meta data it has to measure far more accurate for every frame anyway. your sample is 2 years old and the windows DWM is evolving with pretty much every major windows update.

and i never even hinted that the windows API is any good.

edit: XH90 for my 120 HZ needs.

[oldpainlesskodi]

Oh boy....AMD, MadVR FSE, 10bt (10 bit at driver level too for 30hz and below....) passthrough AMD api HDR = perfect.

No need to continue this debate, I was just bored.

Update - I don't like anything other than pure refresh rate matching - on movies, anything else is a cluster f&^k.

[DMU]

your sample is 2 years old and the windows DWM is evolving with pretty much every major windows update.

This sample was updated (https://github.com/microsoft/DirectX-Graphics-Samples/tree/master/Samples/Desktop) 2 months ago. ;)

[huhn]

well look closer because tehy change something in thsi project doesn'T eman they touched the HDR code.

the d3d12HDR.ccp was lostz touched 2 years ago with this world changing edit: "Replace tabs with spaces"
and we have this change: <WindowsTargetPlatformVersion>10.0.17763.0</WindowsTargetPlatformVersion>
<WindowsTargetPlatformVersion>10.0.19041.0</WindowsTargetPlatformVersion>

the UWP version was changed but that'S not what you linked.
that still doesn't mean i'm correct i don't have the time to really check if this example you post is 2 years old i be like always be wrong.

[SamuriHL]

I'm showing 23.97585 on my 2070.

[huhn]

Oh boy....AMD, MadVR FSE, 10bt (10 bit at driver level too for 30hz and below....) passthrough AMD api HDR = perfect.

No need to continue this debate, I was just bored.

Update - I don't like anything other than pure refresh rate matching - on movies, anything else is a cluster f&^k.

so an API that only works in two mods is perfect compared to an API that literally can do HDR with everything under pretty much every condition?

if an image with zero control over it is perfect than yes the AMD API can do that and nothing else.

[Klaus1189]

I'm showing 23.97585 on my 2070.

Out of the box? No custom tweaking?

[SamuriHL]

Out of the box? No custom tweaking?

Correct. I don't use CRU or anything else. Just out of the box, that's what I'm getting.

[Klaus1189]

@SamuriHL: Thanks

@all: please post your refresh rate for huhn and madshi

[jmone]

Here is a summary of my quick test with the new nVidia drivers after a couple of hours playing a movie. I was hoping for no frame repeats/drops @ 23.976 (4K) out of the box with no tweaks... close but no cigar. Also the Refresh Rates move around a little over the playback period.
- 23.97531 to 23.97536 : 1660Ti to a Sony OLED - 1 Repeated Frame in 2hrs of playback
- 23.97452 to 23.97458 : 1660Ti to a JVC PJ - 2 Dropped Frames in 2hrs of playback

Also, It seems the ability to custom tweak from inside madVR is broken on this driver. You can "Test" just fine, but when you go to save it fails "The GPU driver rejected this mode, for unknown reasons".
https://jmone.org/jmonenextcloud/index.php/s/SztdBYbtekQR4H6/preview

[cableffm]

23.97576 / 1 drop in 1.1hours
without tweaking.

sometimes it's moving between .97572 and .97576.

(1 repeated in 5 mins before)
2080ti-->Sony Oled
456.38
12bit/RGB full/23p
HDR output in madvr

Win10 2004, no HDR probs :D n
NVHDR works, no switch to OSHDR

[VBB]

Same here with my GTX 960. Went from 4.5 minutes to a little over an hour. I'll take that :)

Also, for me HDR goes straight into HDR mode, not SDR first.

[NikosD]

Paper launch, limited stock by design, scalpers with bots, ridiculous prices.

They reminded me a recent article:
https://www.mooreslawisdead.com/post/nvidia-s-ultimate-play

Also, the queues outside physical retail stores were not good reminding me Apple's situation.

Some people discovered a new use for 3080 at 50100$
Money laundering...Interesting.
https://i.postimg.cc/m2mnrK3X/75242-02-zotac-geforce-rtx-3080-trinity-graphics-card-on-ebay-for-over-50-000-full.png

[VBB]

I think this also happened with previous drivers, but for some reason I cannot get YCbCr 4:4:4 8-bit to play nice with HDR in normal or PC mode. It always ends up with everything being green. If anyone knows how to prevent that from happening, please let me know. I tried exclusive, windowed, FS windowed, D3D9 overlay. All the same.

[nevcairiel]

Some people discovered a new use for 3080 at 50100$
Money laundering...Interesting.


I read an article just now that apparently some people are running anti-resale bots that make fake accounts and jack up the ebay listings of the cards to ruin the scalpers days, since it wont actually be sold anytime soon that way.
They just need to keep it up for long enough until supply normalizes, and the scalpers might be stuck with their cards and no profit to make.

[chros]

Also, for me HDR goes straight into HDR mode, not SDR first.
Which OS version do you use?

Some productivity benchmarks have shown the real power of the massive number of FP32 CUDA cores, but of course for that to be meaningful an application has to be able to use it.
Good point, do you have a link?

(Also, I can't watch Jims videos, his accent just drives me crazy)
:D Scottish, I really like it, I got kinda used to it during the years due to couple of colleagues. But check out one of his early videos :D
I like to read your insight as well, maybe you should also start a podcast :) (I got to the point that I mostly interested in theories rather actually using stuff :D )

You only need to read my posts here, I've been telling you the same things from day one BEFORE he releases his videos.
Your write ups would be also interesting to read if you wouldn't want to push them into others' throat :)

Now, regarding the gaming workload analysis, the 25% of INT commands mentioned above is wrong.
The figure given by nVidia when Turing was released back then at 2018 that was the first architecture to execute INT+FP at the same time was around ~35 % or so to speak 1/3.
No, it's ~26% (https://youtu.be/nJBggUfYozY?t=724) (not that it matters much): they didn't say 36 out of 100, but additional 36 on top of 100, meaning 36/(100+36) = ~25.9.
But of course it's only an average, as Nev said it can be completely different based on a given application, e.g. can be even different when using jinc vs NGU Sharp upscaling in madvr.

I'm under the impression that will beat 3080 at both 1080p and 1440p resolutions but will lose a little at 4K.
AMD will release later the full die of 80CUs using HBM2e memory and will beat even the 3090 card even at 4K.
I hope so, I like competition! Although I'm not sure about the HBM2 ones.

[NikosD]

We are talking here about gaming Geforce cards aka gaming workloads only.

And still, nobody has explained why Ampere is underperforming so badly in lower resolutions like 1080p and 1440p and this has nothing to do with CPU bounded gaming cases at lower resolutions.

This thing doesn't scale proportionally.

And the gaming workloads using 36 INT instructions per 100 FMADD instructions can't explain why in the best case scenario of 4K resolution, the 3080 of 30TFLOPS is on average only 70% faster than the 2080 of 10TFLOPS card.

A lot of hardware inside Geforce Ampere is underutilized in gaming workloads and at the same time the figures written in Ampere's specs are no real-world or even optimized or even best-case maximum figures.

The 3x of TFLOPS number hasn't been reported anywhere even in the most compute-intensive floating point productivity benchmarks like CUDA rendering using V-RAY and Blender.

[SamuriHL]

Same here with my GTX 960. Went from 4.5 minutes to a little over an hour. I'll take that :)

Also, for me HDR goes straight into HDR mode, not SDR first.I'll try rebooting and see if that fixes it for me.

Sent from my SM-G975U using Tapatalk

[NikosD]

Judging by the eBay reselling of 3080 cards at 1700$ to 2000$ (initial price 700$) and the pre-orders of PS5 reselling at 800$ to 1000$ (initial price 500$) it is safe to say that marketing departments should be closed and overloaded hype should be banned from everywhere.

But all these tricks are the base of the so called "free market" in order to create inflated demand, resulting in inflating prices.

[SamuriHL]

We'll see if the scalpers are brave enough to try this s**t with AMD cards. LMAO

[VBB]

Which OS version do you use?

Windows 10 Enterprise 64-bit 2004 19041.508

[VBB]

chros, SamuriHL, when you get a chance, could you please test YCbCr 444 8-bit with HDR to see if you also get that green screen issue? Thanks!

[Klaus1189]

https://www.youtube.com/watch?v=yE9n_llJOVQ
https://www.youtube.com/watch?v=qHogHMvZscM

[chros]

@VBB: Sorry, but I haven't upgraded it and I don't have time to fool around with it, hopefully others will.

We'll see if the scalpers are brave enough to try this s**t with AMD cards. LMAO
:D

[el Filou]

So, are we back to using NVidia's API for HDR with this driver, or does it still switch to OS HDR?After reading this KB article: https://nvidia.custhelp.com/app/answers/detail/a_id/5072 (linked to in the release notes for the last drivers), as they're suggesting to sync the Windows HDR setting with the application HDR setting (in this case, madVR), I am under the impression that NVIDIA is trying to dynamically switch Windows HDR itself on or off. Don't know if this is correct but clearly something has changed compared to the old NVHDR way that directly controlled the display from the app.Oh boy....AMD, MadVR FSE, 10bt (10 bit at driver level too for 30hz and below....) passthrough AMD api HDR = perfect.
No need to continue this debate, I was just bored.I get how people with some hardware that suits their use case can think some other hardware has bugs that they consider showstoppers, but sincerely, the best way to avoid tiring debates is to not start them by proclaiming that a personal use case equals the perfect hardware config for everyone. ;)

[NikosD]

People and users of this forum don't seem to understand even the basics of the covered monopoly game of Intel and nVidia.
Both these companies during the last few decades (!) were spending a huge amount of money in marketing and PR (BRIBING) of tech journalists and OEMs in order to do two simple things:

1) Avoid AMD's products inside their devices (OEMs) even if they are better (recent example of Ryzen processors in laptops, EPYC processors in servers etc and Threadrippers in HEDT) and exaggerate over any weakness (tech journalists) compared to theirs (Intel & nVidia) - recent example of 5% faster Intel processors in 1080p gaming.

2) Produce a lot of hype for their products even for useless features (hairworks, physx, RTX in 2018 for nVidia/ AI, AVX-512 inside Tiger lake and desktop processors for Intel etc) and bury or underestimate any AMD's better feature.

So, even after a desktop processor of 8C/16T for desktop Zen1 Ryzen which later became 16T/32T for Zen 2 even after a 32C/64T processor for servers which later became 64C/128T processor and the same for HEDT market with Threadrippers, the OEMs and tech journalists didn't have any intense positive reaction.

No overloaded hype was created, no inflated prices for AMD products.
Even for the best of the best CPU processors that mankind has ever created by far - the whole Zen 2 architecture implementations - no particular hype was ever created by OEMs and tech journalists.
Because Intel spends BILLIONS of $$$ to BRIBE these two - OEMs and tech journalists.
The margins of AMD are still low - even for Zen 2 CPUs.
So, after the next release of the absolute monster of CPU processors, the divine Zen 3 CPUs that will be released on 8th of October, no extraordinary hype will be created, no limited stock by design, no inflated prices.
Just the best processor you could ever have, in right stock and price.
The same with GPUs although not even this time nVidia is in position of Intel, unfortunately...

[VBB]

After reading this KB article: https://nvidia.custhelp.com/app/answers/detail/a_id/5072 (linked to in the release notes for the last drivers), as they're suggesting to sync the Windows HDR setting with the application HDR setting (in this case, madVR), I am under the impression that NVIDIA is trying to dynamically switch Windows HDR itself on or off. Don't know if this is correct but clearly something has changed compared to the old NVHDR way that directly controlled the display from the app.

Yeah, I'm sure MS is pushing OS HDR, and Nvidia will likely abandon their API eventually. With this new driver, though, at least madVR is still able to use NV HDR.

[el Filou]

2) Produce a lot of hype for their products even for useless features and bury or underestimate any [competitor]'s better feature.Congrats, you just described marketing. Every company is doing it when they can, even AMD. Remember the x3 multitexturing no devs used? Truform tessellation? AVIVO that wasn't even fixed-function video decode? TrueAudio? The one-die 4-cores when Intel "used glue" for its 4-cores even though it didn't matter much as the market for 4-cores was so small at the time? The supposedly "future-proof 8-core" FXs ?So, even after a desktop processor of 8C/16T for desktop Zen1 Ryzen which later became 16T/32T for Zen 2 even after a 32C/64T processor for servers which later became 64C/128T processor and the same for HEDT market with Threadrippers, the OEMs and tech journalists didn't have any intense positive reaction.
That's because for most consumer workloads until now, adding more cores had diminishing returns apart from things like two-pass video encoding (live encoding for streaming is best done on the GPU), 3D rendering, and pure compute, which are things most people aren't doing. Maybe it will change now even consoles have 8 cores and games start using them, but until now it wasn't a game changer. And no, the journalists did have intense positive reaction for the high-core count AMD models, but only for specific use cases."best of the best CPU processors that mankind has ever created" ... "absolute monster" ... "divine"You do realize you're talking like someone who's in a cult, right?

[Klaus1189]

If there wasn‘t AMD, mainstream Intel would still have 4c/8t...

[huhn]

yes AMD was the reason that intel didn't need to bring more cores to the market.

[almawdah]

So what the recommended , can we upgrade the driver to latest driver for nvidia to use it with MadVR ??

[Manni]

So what the recommended , can we upgrade the driver to latest driver for nvidia to use it with MadVR ??

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.

[el Filou]

If there wasn‘t AMD, mainstream Intel would still have 4c/8t...But they only started releasing mainstream CPUs with more than 4 cores when AMD became competitive overall, it wasn't just the mere existence of more than 4 cores from AMD that made them do it: Phenom IIs had 6-core versions on the mainstream platform, but that didn't cause Intel to release mainstream 6-cores, they reserved them for the HEDT platform.
My point is that what's important is if the number of cores can actually benefit your use cases. It's good that AMD was able to release a 16-core on a mainstream platform with Zen 2, but with mainstream apps (i.e. Web, games, productivity apps) you hardly see the benefit anyway, heck in some cases the 12- and 16-cores can be slower than the 8-core. So while you have to salute the technological prowess, and 16 cores can be useful for content creators or devs or people who run a whole lab in VMs, I see no need for journalists to have an "intense reaction" (or is that a euphemism for something?).

[almawdah]

I've only tested 8bits RGB 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.

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.


Thanks very much brother


Sent from my iPhone using Tapatalk

[Manni]

Thanks very much brother


Sent from my iPhone using Tapatalk

You're welcome :)

One last thing, I'm still using CRU so I haven't tested native frame rate.

I'll edit my post to reflect that.

[SamuriHL]

chros, SamuriHL, when you get a chance, could you please test YCbCr 444 8-bit with HDR to see if you also get that green screen issue? Thanks!

Oh for crying out loud. LOL Yea I can look into it in a bit. That requires a lot of setup. lol

[VBB]

:D What do you have to change? I appreciate it. I'll do some more testing tonight. Perhaps it just requires a reboot on my end, which I didn't get to last night.

[SamuriHL]

I still haven't rebooted, either and probably should. As for what I have to change, a bunch of crap. LOL First YCbCr 444 mode, then changing to PC mode on the TV, which apparently reset all my damn picture settings completely screwing up my calibration (SIGH, screw you LG)....Now that I fixed all that and set it up, I threw on 1917 just to test it. No green, and PQ looks pretty damn impressive. This isn't a movie where I'm likely to notice a lot of banding so, um, you're making me mad by having me test this seeing how good it COULD look if LG hadn't completely blew it on PC mode. LOL So uh yea thanks for that :D

[VBB]

LOL well thanks for letting me now that it works for you. I'm sure you do know that PC mode is completely separate as far as settings and calibration go. On my C7, the two modes even need different brightness settings to have the same black level.

Also, I noticed again last night that YCbCr 4:4:4 actually has very, very little banding in both SDR and HDR, in PC mode. That's why I'm now trying to get it to work.

[SamuriHL]

Yes I do know that which is why I said I had a bunch of setup to do. :) Not HARD to do but I just had to remember everything to set without looking. Anyway, it wasn't a big deal and I was able to test it.

I'd have to do a lot more testing on that theory. PC mode has been an unmitigated mess for LG. ALL of us have WANTED it to work but we've always found issues. I'm assuming you're sticking to 8 bit for this venture yes?

[VBB]

Oh, 8-bit for sure. It's pretty much established now that PC mode switches the TV into 8-bit mode anyway, and of course my old C7 and GTX 960 can't do full chroma at higher than 60Hz anyway. And, as we know by now, the new Nvidia driver did not enable 10-bit for anything lower than 120Hz outside of 4:2:2. So, nothing's really changed for us C7/C8 users, except a much more accurate 23Hz mode :)

I'm trying to switch from normal HDMI mode, 4:2:2 10-bit at 60Hz, with smooth motion, to PC mode 4:4:4 8-bit, with frame rate switching.

[SamuriHL]

So far I like this new driver a lot. I should grab the studio version just to have it. The improved 23Hz mode is a welcome change. We can thank madshi for pushing them on that one. :)

[VBB]

Definitely. BTW, now that I'm thinking about using PC mode again, has anyone ever had an issue with the lack of 5:5 pull down, since Real Cinema is permanently OFF?

[SamuriHL]

I use real cinema always so I can't tell you on that one. But that could be a real problem.

[jmone]

Here an update on my testing with the new nVidia drivers after a couple of hours playing a movie. I was hoping for no frame repeats/drops @ 23.976 (4K) out of the box with no tweaks... close but no cigar. Also the Refresh Rates move around a little over the playback period.
- 23.97525 to 23.97434 : 1660Ti to a Sony OLED - 2 Dropped Frame in 2hrs of playback (Decoding Audio)
- 23.97531 to 23.97536 : 1660Ti to a Sony OLED - 1 Repeated Frame in 2hrs of playback (Decoding Audio and Running JRivers "Video Clock")
- 23.97452 to 23.97458 : 1660Ti to a JVC PJ - 2 Dropped Frames in 2hrs of playback (Bitstreaming Audio)

My conclusions so far are:
- Out of the box: The drivers seem to adjust the Refresh Rate during playback to try to avoid Repeated/Dropped frames and it works pretty well... or is it just clock Drift? Anyway, the good news is that over the length of a 2hr movie I get 2 Dropped Frames on a couple of different setups.
- JRiver Video Clock: When I played a movie also using JRivers "Video Clock" (similar to Reclock), I then got 1 Repeated Frame. I don't know if they are "fighting" each other but historically, Video Clock would give me 0 Dropped/Repeated frames on my OLED Setup. Maybe JRiver Video Clock needs a tweak to work a bit better with this new driver???
- madVR Custom Modes: Can't save any custom modes anymore to tweak the refresh rate which previously also got me to 0 Dropped / Repeated frames (on my Bitstreaming PJ setup).

So it is a big improvement but for the perfectionist..... still a bit to do (I guess I'll have to look at CRU unless others have suggestions???)