View Full Version : madVR - high quality video renderer (GPU assisted)
tp4tissue
29th June 2019, 00:25
Hey guys, which movie "teases out" the new dci-p3 gamut ?
I'm not seeing a huge difference between wide and standard gamut tv in movies, side by side, both ran through displaycal and i1d3.
xabregas
29th June 2019, 21:13
What gpu should i buy to play 1080p -> 4k sources and some 4k playback on madvr with decent (medium) settings??
Warner306
30th June 2019, 12:27
A GTX 1060 would do the job.
xabregas
30th June 2019, 18:08
A GTX 1060 would do the job.
So i assume a gtx 1660 would be a good choice?
ryrynz
30th June 2019, 21:32
More than adequate.
Charky
1st July 2019, 13:22
A GTX 1060 would do the job.
You're lucky.
Every time I say the same on this very topic, I get death threats from die hard madvr users yelling that madvr can't to anything with only a GTX 1060 and that anyone serious about picture quality should mortgage their house and buy a RTX 2080 TI...
yeah it's getting insane...
if you are ok with a 1060 look for used.
cork_OS
1st July 2019, 15:10
As 1060 3Gb user I would recommend 6Gb version.
Manni
1st July 2019, 15:39
You're lucky.
Every time I say the same on this very topic, I get death threats from die hard madvr users yelling that madvr can't to anything with only a GTX 1060 and that anyone serious about picture quality should mortgage their house and buy a RTX 2080 TI...
I don't think anyone made death threats, the question is what exactly do you want to do with madVR.
If you only use a few of the features, or are mostly playing 1080p content (or downscale 4K content) on a 1080p display then sure, a 1060 will do fine.
But if you have a 4K display/projector and want to use the latest HDR dynamic tonemapping, 3D LUT calibration, 12bits, some processing, black bars detection, NGU upscaling of 480p, 720p or 1080p content to 4K, then I doubt that a 1060 is enough, especially at 60p, unless you're happy to reduce the quality (and I'm not talking going from NGU Chroma very high to high, I'm talking settings that actually reduce the PQ).
I have a 1080ti and I have to limit the amount of processing I do in many situations (I use NGU chroma anti-alias high and NGU luma sharp very high). And at the moment, I only use 8bits due to a bug with the JVCs and not 12bits (which does save a significant amount of power. If I could use 12bits, I'd be even more limited).
So if you just want it to work and are happy to play with the settings until you don't get dropped frames, use only 8bits on a 1080p screen, are happy to not use a 3D LUT, can live without black bars detection or any other feature that doesn't work with D3D11 native and if you only use HDR passthrough, then sure, get a 1060.
If you want to get the best PQ available at the moment (especially for HDR with dynamic tonemapping) and use all the features listed above, then I'd suggest at least a 1080ti.
And if you want to be able to tap into the tensor cores potential that might be exploited by madVR at some point, then I wouldn't buy a GTX but would only buy an RTX.
Madshi himself recommends to get an RTX rather than a GTX. :)
SirMaster
1st July 2019, 15:59
You're lucky.
Every time I say the same on this very topic, I get death threats from die hard madvr users yelling that madvr can't to anything with only a GTX 1060 and that anyone serious about picture quality should mortgage their house and buy a RTX 2080 TI...
Lol, I have a laptop with a 1060 in it and I have dynamic tone-mapping enabled, NGU sharp medium chroma, NGU sharp very-high image, and error diffusion 1 dithering all enabled and there are no frame drops. Also all the trade quality for performance options are unchecked. It's most certainly fast enough.
Plays 1080p and 4K UHD 24Hz both just fine. If I want to do 60Hz (which is very rare), then I go with Lancosz instead of NGU for scaling and it works.
Manni
1st July 2019, 16:25
Lol, I have a laptop with a 1060 in it and I have dynamic tone-mapping enabled, NGU sharp medium chroma, NGU sharp very-high image, and error diffusion 1 dithering all enabled and there are no frame drops. Also all the trade quality for performance options are unchecked. It's most certainly fast enough.
Plays 1080p and 4K UHD 24Hz both just fine. If I want to do 60Hz (which is very rare), then I go with Lancosz instead of NGU for scaling and it works.
Do you use a 3D LUT for calibration? Do you need black bars detection or UHD Bluray menus, hence can't use D3D11 native? Do you use 12bits? Did you try upscaling a 480p DVD (50p or 60p) to 4K with your settings?
Again, people can be very happy with a 1060 if they only use a fraction of the features or are happy to reduce PQ (NGU Chroma medium or using Lancosz instead of NGU is reduced quality, though I agree not very noticeably), especially (but not only) at 60p.
It's all about which features you need, and how happy you are to compromise PQ. You seem to be very happy to compromise PQ and only use a subset of features, there is nothing wrong with that.
Ver Greeneyes
1st July 2019, 17:02
As 1060 3Gb user I would recommend 6Gb version.
Seconded. It's okay with smaller queues but having to compromise on those isn't great. Every major update to Windows 10 has also made things slower for me, although part of that may be CPU-related (I haven't done too much testing since the last time madshi fixed a function that started using a ton of CPU for no reason, so I'm not sure).
Do you use a 3D LUT for calibration? Do you need black bars detection or UHD Bluray menus, hence can't use D3D11 native? Do you use 12bits? Did you try upscaling a 480p DVD (50p or 60p) to 4K with your settings?
these settings cost "little" to nothing even for a way way older GPU.
d3d11 doesn't make a real major difference anyway if it would do that how could an iGPU even do a copyback operation at all?
rendering at 10 bit doesn't cost much more either hard to even measure it's pretty much writing an 10 bit output instead of 8 the math/work should be the same.
blackbar detection can even save performance.
even 480p60 upscaling can be done using NGU just tested 576p50 with ngu high quadruple with 13 ms.
currently tonemapping and 1080p->UHD scaling or quadrupling which has a huge diminishing return are very costly.
Manni
1st July 2019, 19:30
these settings cost "little" to nothing even for a way way older GPU.
d3d11 doesn't make a real major difference anyway if it would do that how could an iGPU even do a copyback operation at all?
rendering at 10 bit doesn't cost much more either hard to even measure it's pretty much writing an 10 bit output instead of 8 the math/work should be the same.
blackbar detection can even save performance.
even 480p60 upscaling can be done using NGU just tested 576p50 with ngu high quadruple with 13 ms.
currently tonemapping and 1080p->UHD scaling or quadrupling which has a huge diminishing return are very costly.
Yeah right, why don稚 you get a 1080ti and try to do what I知 doing instead of stating generic nonsense?
Black bar detection or UHD BD menus means no D3D11 native. In my setup, copyback is much slower than native. I知 not cropping black bars, I知 detecting them to do a vertical shift of the picture according to the A/R of the content.
12bits (there is no 10bits option with HDMI on nVidia) is significantly slower than 8bits here. Good for you if it痴 not the case with your setup.
A 3D LUT is also significantly slower, by this I mean a few ms.
Stack all this together and you go from no dropped frames to stutter.
Please post all your settings for the upscaling of 480p60 to 4K60 RGB 4:4:4 8bits with NGU luma very high and NGU chroma high in 13ms on an nVidia GPU (which one?) in D3D copyback on Windows 10 Pro x64 1809 or more recent (see my sig for details), at least that would be useful if true.
Yeah right, why don't you get a 1080ti and try to do what Iツ知 doing instead of stating generic nonsense?
so if i don't have the same card i can test with a different one? i don't have to because i don't have to use the option you are using the need for the individual user are very different and it's no secret that madVR can kill a 2080 ti with useful settings while on the other hand another user is totally happy what a 1060 can do at UHD.
Black bar detection or UHD BD menus means no D3D11 native. In my setup, copyback is much slower than native.
what do you understand as much slower? i even have issues to measure the difference d3d11 is faster and clearly more efficient but at the moment we are paying for an interop just for using it which make it not as good as it should/will be.
Iツ知 not cropping black bars, Iツ知 detecting them to do a vertical shift of the picture according to the A/R of the content.
then why don't you crop them? is it even possible to move them to let'S say the top but not cropping them? i had a discussion about this with madshi when it was a new feature where he claimed it should general save performance so generally be used because madVR is processing less pixel which is true where i was holding the argument that doesn't save performance if you need more scaling or worse just scaling because you are using blackbar detection.
12bits (there is no 10bits option with HDMI on nVidia) is significantly slower than 8bits here. Good for you if itテつ痴 not the case with your setup.
unrelated to madVR it doesn't matter what the GPU is sending the screen madVR is either rendering 8 bit or 10 bit nothing else and yes the performance difference is meaningless to my tested cards.
A 3D LUT is also significantly slower, by this I mean a few ms.
it's clearly not free but even a 960 can do a 3D LUT for like 2 ms
Stack all this together and you go from no dropped frames to stutter.
my 960 can do a 3d lut, blackbar detection plus moving the image so you have to do something far far more demanding like tone mapping NGU very high scaling error diffusion or what ever countless option i'm missing here.
Please post all your settings for the upscaling of 480p60 to 4K60 RGB 4:4:4 8bits with NGU luma very high and NGU chroma high in 13ms on an nVidia GPU (which one?) in D3D copyback on Windows 10 Pro x64 1809 or more recent (see my sig for details), at least that would be useful if true.
i said high not very high which can't be highlighted enough here a OSD from 1060 it's based on default madVR settings no calibration nothing on this system all setting should be visible in the OSD: https://abload.de/img/osd2ejwd.png
it's a 576i50 file which should give higher render times then 480p60 while obviously still easier to do in real-time.
Manni
1st July 2019, 21:45
so if i don't have the same card i can test with a different one? i don't have to because i don't have to use the option you are using the need for the individual user are very different and it's no secret that madVR can kill a 2080 ti with useful settings while on the other hand another user is totally happy what a 1060 can do at UHD.
Do you ever read what people write before writing nonsense? This is EXACTLY the point I was making. That for some people who don't use all the features and are ready to compromise on PQ, a 1060 is fine. And for others, who use all the features and don't want to compromise on PQ, a 1080ti can be just enough.
what do you understand as much slower? i even have issues to measure the difference d3d11 is faster and clearly more efficient but at the moment we are paying for an interop just for using it which make it not as good as it should/will be.
Here D3D11 copyback is at least 15-20% slower than D3D11 native. It's even more obvious when measuring files with MadMeasureHDR. It takes 10-15mn in D3D11 native with my 1080ti, at least 20-25mn in copyback, and more than 30mn with my iGPU on the MBP 13" 2018.
then why don't you crop them? is it even possible to move them to let'S say the top but not cropping them? i had a discussion about this with madshi when it was a new feature where he claimed it should general save performance so generally be used because madVR is processing less pixel which is true where i was holding the argument that doesn't save performance if you need more scaling or worse just scaling because you are using blackbar detection.
Of course it is possible to shift them without cropping them, I do it all the time. I tried cropping and it doesn't make much of a difference. In any case, you need to test with 1.78 pictures otherwise it looks like you're fine and then you play Pacific Rim in HDR and you don't understand why your frames are dropping. That's why I don't bother cropping the frames. Whatever settings I choose have to work with 1.78 content, so even if I could save a few ms with cropping wide ARs it wouldn't help my 1.78 titles.
In any case, I only mention the black bars detection because it's a common software feature that requires copyback. I need copyback for UHD Bluray Menus in jRiver, so black bars isn't really relevant.
unrelated to madVR it doesn't matter what the GPU is sending the screen madVR is either rendering 8 bit or 10 bit nothing else and yes the performance difference is meaningless to my tested cards.
Good for you. It's not meaningless here. That's why I suggest you test a 1080ti with my OS and then report back, instead of claiming that there is no difference.
it's clearly not free but even a 960 can do a 3D LUT for like 2 ms
And what difference do you make between "a couple of ms" and "2ms"? That's exactly what I said.
my 960 can do a 3d lut, blackbar detection plus moving the image so you have to do something far far more demanding like tone mapping NGU very high scaling error diffusion or what ever countless option i'm missing here.
What you're missing is the worst case scenario. I don't care what you can do with your 960 without using HDR dynamic tonemapping. I'm only telling you that when you try to play say Pacific Rim in 4K HDR with dynamic tonemapping in 12bits with a 3D LUT in D3D11 copyback using NGU Chroma high or very high, it's much more demanding than when you play Lawrence of Arabia in 8bits without a 3D LUT in passthrough with lower chroma settings.
It's all the extra features that push a perfectly fine 35ms to a stuterring 42ms. You only need a few ms here and a few ms there to end up on the wrong side of 41ms, especially with 1.78 content in HDR.
A lot of people brag about their lower GPU, but they either don't use all the features or they compromise on quality. Or they don't notice frame drops.
i said high not very high which can't be highlighted enough here a OSD from 1060 it's based on default madVR settings no calibration nothing on this system all setting should be visible in the OSD: https://abload.de/img/osd2ejwd.png
Thanks I'll take a look.
Again, my point isn't to say that you can't be happy with a 1060. Clearly a lot of people are happy with a 1060, and I'm very happy they are. But if someone buys a 1060 today, I'd rather be sure that they realise that they are limiting the number of features and the PQ they can get. That's all. I didn't upgrade to a 2080ti because I don't really need to, but I'm glad I have the 1080ti, especially since madshi is doing the dynamic HDR Tonemapping, it's VERY taxing.
xabregas
1st July 2019, 21:51
I need to sacrifice something on my new computer for madvr 1080p->4k.
gtx 1650 + 16GB RAm + 250GB NVme SSD M2 + fancy case with lots of 3.5/2.5 slots
or
gtx 1660 + 8GB RAM + normal ssd 250GBG + not so fancy case with enough but limited 3.5/2.5 slots.
Please help!!!
before i answer in detail. never use d3d11 copyback if you don't have to do that. use d3d9 copyback.
Manni
1st July 2019, 21:57
before i answer in detail. never use d3d11 copyback if you don't have to do that. use d3d9 copyback.
You need D3D11 for 10bits in Full screen windowed mode. D3D9 isn't an option for 10bits unless you're happy with FSE (I'm not).
brazen1
1st July 2019, 22:06
Maybe just some simplicity for noobs?
There are two camps:
1080p PJ's with and without fake 4k or Displays that are 4k.
You are one or the other, perhaps both.
1080p needs a way to play 4k titles especially using HDR. Look into all the testing that is being created for this camp and the hoops you need to jump through in order to produce HDR looking results. Expect to use the high quality expensive video cards to process something watchable. A GTX 1070 is the minimum from what I understand and more expensive higher cards will produce even better results. Can't speak for AMD cards.
2160p displays can use low to mediocre cards producing the same if not better results than the algo camp above ^. GTX 960 4GB is a good example. With a good UHD HDR rip, nothing more is needed imo. You are not going to increase any picture quality by using a higher quality video card to use more algo processing because there isn't anything to improve on a native UHD rip on a native 4k display. That's assuming you have a decent display that doesn't need its deficiencies improved which is a whole nuther story. If that's the case, then you belong in the PJ camp.
You can expect better results upscaling 540, 720, and 1080p titles on your 4k display but tbh, the results will more than likely appear (small) and I doubt YMMV. This includes junk rips downloaded from someplace. You can only improve this crap so much before the return is negligible and you don't need 6/8 GB of memory to improve them nor do you need new cards costing as much as a mortgage payment. The prices are and have been outrageous forever not because of the tech and production cost but to appease over exuberant stock market investors and nothing more.
You need D3D11 for 10bits in Full screen windowed mode. D3D9 isn't an option for 10bits unless you're happy with FSE (I'm not).
totally unrelated they just have the same name because it is the same API.
d3d9 presentation can't do 10 bit presentation what so ever and decoding has nothing todo with that.
edit: i should just use the correct name DXVA2 copyback which uses d3d9.
Alexkral
1st July 2019, 22:52
Honestly, I don't think dynamic tone mapping in the way it's being developed deserves even a bit of extra processing. If you want brighter highlights there are other ways to get them, just use the curve the way it was formulated. Probably most people don't know that madVR uses an alternative roll-off based on the DisplayCAL implementation, which has the property of better preserving the detail near the peak at the expense of luminance. This is also the default option in DisplayCAL, but at least there you have the option to use the original BT.2390 roll-off. Now one may think that the latter is less correct because it has the risk of producing overexposed highlights, but in reality this is simply an aesthetic choice. It's a trade off between detail and luminance, some filmmakers prefer the former and others the latter, and neither is more correct than the other. We are sick of seeing blown out windows in SDR and nobody is shocked, probably only because they are not so bright, so I don't understand this obsession to preserve even the smallest detail of the highlights. Using a DisplayCAL 3DLUT you can even adjust that trade off by setting the Mastering display peak luminance, I do this with my shaders and for me this increase in luminance really makes a difference.
I'm not against dynamic tone mapping, but I think it would have made more sense that its development would have been based on the original roll-off. I was more interested at the beginning in dynamic target nits, which I thought could be very useful to raise the reference white level in a way that made sense, but lately I'm thinking it's probably more correct to set a static target nits based on the measurements. Anyway, to say that this is the best PQ that can be achieved, seems to me at least quite exaggerated. The HDR10+ curve is much more sophisticated, and unlike the BT.2390 EETF, has been expressly designed to do dynamic tone mapping. Furthermore, there are currently spatially varying tone mapping algorithms that are able to deal with problems such as noise visibility or contrast distortions, and which should not be very difficult to implement.
mclingo
1st July 2019, 23:06
some people on here have AMD cards, but you'd never guess it :)
xabregas
1st July 2019, 23:46
some people on here have AMD cards, but you'd never guess it :)
In that case, between RTX 570, RTX 580 and GTX 1650? pricewise
I know about the NGU thing on amd polaris but...
Wait until the 7th... GPU prices may shift a lot with the upcoming Navi releases
QB
el Filou
2nd July 2019, 00:18
gtx 1650 + 16GB RAm + 250GB NVme SSD M2 + fancy case with lots of 3.5/2.5 slots
or
gtx 1660 + 8GB RAM + normal ssd 250GBG + not so fancy case with enough but limited 3.5/2.5 slots.Is this for a dedicated HTPC, or a general purpose PC that will also be used for madVR?
For dedicated HTPC the second option is better. For a general usage PC I wouldn't get less than 16 GB of RAM.
xabregas
2nd July 2019, 01:04
Is this for a dedicated HTPC, or a general purpose PC that will also be used for madVR?
For dedicated HTPC the second option is better. For a general usage PC I wouldn't get less than 16 GB of RAM.
Mainly general purpose PC that will also be used for madVR!!
I can replace gtx 1650 with RX 580 4GB or RX 570 8GB. The price is the same.
Is the only problem of amd polaris on NGU algos??
Warner306
2nd July 2019, 01:29
Honestly, I don't think dynamic tone mapping in the way it's being developed deserves even a bit of extra processing. If you want brighter highlights there are other ways to get them, just use the curve the way it was formulated. Probably most people don't know that madVR uses an alternative roll-off based on the DisplayCAL implementation, which has the property of better preserving the detail near the peak at the expense of luminance. This is also the default option in DisplayCAL, but at least there you have the option to use the original BT.2390 roll-off. Now one may think that the latter is less correct because it has the risk of producing overexposed highlights, but in reality this is simply an aesthetic choice.
I think the aesthetic choice comes from designing tone mapping for SDR devices that don't produce overly bright highlights anyways. Getting extra pop from 100 nit highlights on a projector has negligible returns in its own right. User feedback drove some of the decision making. I'd have to see images of the two curve options to see if the difference is notable.
I'm not against dynamic tone mapping, but I think it would have made more sense that its development would have been based on the original roll-off. I was more interested at the beginning in dynamic target nits, which I thought could be very useful to raise the reference white level in a way that made sense, but lately I'm thinking it's probably more correct to set a static target nits based on the measurements.
The static target nits sacrifices too much brightness or too much contrast, depending on the scene. For SDR output, it can sometimes seem necessary when you are working with a display with a limited range of contrast. It also prevents some of the issues on low brightness displays such as lowering the black floor too much with a static curve when the source has a lot of content near black and a fair bit of content at a high peak white.
Anyway, to say that this is the best PQ that can be achieved, seems to me at least quite exaggerated. The HDR10+ curve is much more sophisticated, and unlike the BT.2390 EETF, has been expressly designed to do dynamic tone mapping. Furthermore, there are currently spatially varying tone mapping algorithms that are able to deal with problems such as noise visibility or contrast distortions, and which should not be very difficult to implement.
Does HDR10+ have a tone curve? Do you have a link to information on this curve? HDR10 leaves the choice of curve to the display. The spatial tone mapping options have been discussed at AVS Forums, but they were deemed too slow, even for madVR.
Edit: I found the HDR10+ Bezier curve that can be embedded in HDR10+ content on Google. It looks like an interesting curve for rolling off the highlights, but it doesn't seem to dig deep enough to be a good curve for HDR to SDR tone mapping.
HDR10+ does apparently have the ability to read the APL of the image to avoid unwanted tone mapping for low APL scenes and has a method of avoiding tone mapping of small patches of bright pixels.
http://mile-high.video/files/mhv2018/pdf/day1/1_10_Mandel.pdf
the 1650 is a generally bad card avoid it.
polaris are freaking heater with missing hardware decoder like VP9 which youtube and netflix are pushing hard.
SamuriHL
2nd July 2019, 02:17
Not really jumping into the card debate too far, but, I will tell you that coming from the 1060 6gb going to a 2070, FOR ME, was a really good decision. The 1060 I did have to compromise on settings that I personally did not want to really compromise on, however, that being said, it was very capable. It's without a doubt the lowest card I'd recommend. These 16xx cards don't seem like a good bargain for madvr. You'd be better off with a 1070 or 1080 IMO.
As for AMD, I've no modern experience there. AMD burned a bridge with me in their 5xxx series of cards and I've not really bothered looking back as nVidia has served me well for the past many years. But, that being said, AMD is getting more aggressive so I wouldn't necessarily count them out. I do agree on hardware decoding and I believe, correct me if I'm wrong on this, that hardware encoding on AMD cards is also behind nVidia. That's not directly related to madvr necessarily, but, something to consider.
My $0.02
Alexkral
2nd July 2019, 04:07
Getting extra pop from 100 nit highlights on a projector has negligible returns in its own right. User feedback drove some of the decision making.
Of course if we accept that the SDR is equal to 100 nits then you are probably right. I don't think it makes sense to use the original roll-off on displays with limited luminance, in fact the result would be worse. I can also clearly see the potential benefits of using dynamic target nits in situations where the limited range needs to be exploited in the best possible way. What happens is that today most TVs can reach around 300 nits, so taking into account the above, and the name of the thread in which the tone mapping development is taking place, it may be time to assume that madVR tone mapping is being optimized for low nits projectors, with all the limitations that this implies for more brilliant displays.
And in the same way I can easily agree with you that unlike madVR, the HDR10+ curve has not been developed with SDR 100 nits in mind.
Warner306
2nd July 2019, 13:13
Some projectors can approach 300 nits, but they are getting far less when calibrated. Raising the brightness that high also raises the black floor, so it is better to get as much contrast as possible on a projector with HDR content rather than attempting to max out peak brightness.
Maybe a separate curve option would benefit HDR output.
Do you ever read what people write before writing nonsense? This is EXACTLY the point I was making. That for some people who don't use all the features and are ready to compromise on PQ, a 1060 is fine. And for others, who use all the features and don't want to compromise on PQ, a 1080ti can be just enough.
the discussion starts because people tent to claim that you need a high end GPU for people that ask for scaling not tonemapping.
it goes as far as people claim for X you need a 1080 or better even through the exact claimed option can be done with a 1060.
Here D3D11 copyback is at least 15-20% slower than D3D11 native. It's even more obvious when measuring files with MadMeasureHDR. It takes 10-15mn in D3D11 native with my 1080ti, at least 20-25mn in copyback, and more than 30mn with my iGPU on the MBP 13" 2018.
measuring files is a bad comparison there is only that much PCIe bandwidth if you run out of it madVR has to wait and that's what's happing here.
Of course it is possible to shift them without cropping them, I do it all the time. I tried cropping and it doesn't make much of a difference. In any case, you need to test with 1.78 pictures otherwise it looks like you're fine and then you play Pacific Rim in HDR and you don't understand why your frames are dropping. That's why I don't bother cropping the frames. Whatever settings I choose have to work with 1.78 content, so even if I could save a few ms with cropping wide ARs it wouldn't help my 1.78 titles.
at least i'm not alone on this anymore...
In any case, I only mention the black bars detection because it's a common software feature that requires copyback. I need copyback for UHD Bluray Menus in jRiver, so black bars isn't really relevant.
Good for you. It's not meaningless here. That's why I suggest you test a 1080ti with my OS and then report back, instead of claiming that there is no difference.
And what difference do you make between "a couple of ms" and "2ms"? That's exactly what I said.
if a 960 can do it this fast a 1080 ti is supposed to be much faster. that why i say that.
A lot of people brag about their lower GPU, but they either don't use all the features or they compromise on quality. Or they don't notice frame drops.
doesn't that count for every GPU? if someone asked for a upscaling 1080p to 2160p with decent settings he didn't ask for a 2080 ti here.
Manni
2nd July 2019, 14:23
the discussion starts because people tent to claim that you need a high end GPU for people that ask for scaling not tonemapping.
it goes as far as people claim for X you need a 1080 or better even through the exact claimed option can be done with a 1060.
Except that this is NOT what happened in this instance. Someone asks for a recommendation for a GPU WITHOUT saying specifically what they intend to do with it (yet mention 4K content, which means HDR, so possibly HDR dynamic tonemapping). So lots of people reply saying that a 1060 is more than enough. That's not a helpful reply. They should *first* ask what the user plans to do with it, including which features (3D LUT? HDR dynamic tonemapping? Need for software decoding?), instead of starting to brag about their low-end card being more than enough to meet whatever limited use they have for it.
My point was not to say that *everyone* needs a 2080ti. I don't need one.
My point was to say that while some people can be perfectly happy with a 1060, they will quickly reach its limits if they want to do everything that madVR offers, especially if they need software decoding only features either from madVR or from other software or want to use HDR dynamic tonemapping.
I couldn't care less that many people are happy with their 960 or 1060 or whatever. I'm only saying that if you want the best PQ that madVR has to offer and want to use all the features, that's most likely not enough given that I meet some limitations with a 1080ti.
measuring files is a bad comparison there is only that much PCIe bandwidth if you run out of it madVR has to wait and that's what's happing here.
Please read my reply. The difference is 15-20% here, and it's EVEN MORE OBVIOUS when measuring files. It would really help if you could read each reply before writing back more nonsense.
if a 960 can do it this fast a 1080 ti is supposed to be much faster. that why i say that.
And my point is that it's not the case here. A 3D LUT takes a few ms here even with a 1080ti. This is why I suggested that you test a 1080ti, using all the options that I mentioned before coming up with all your nonsense.
doesn't that count for every GPU? if someone asked for a upscaling 1080p to 2160p with decent settings he didn't ask for a 2080 ti here.
Yes, and this has nothing to do with my point, I never said a 2080ti was necessary. The OP also mentioned playing 4K content, which means HDR as far as commercial content is concerned. Read above.
madjock
2nd July 2019, 15:16
Maybe you two can start a new thread where you constantly post each others comments over and over.
We get it you disagree. :devil:
mclingo
2nd July 2019, 15:41
they can disagree to their hearts content on the general thread.
https://forum.doom9.org/showthread.php?t=175769
mambans
2nd July 2019, 16:19
Hello, I would really appreciate some help getting my HDR working in Madvr if anyone don't mind helping.
Both Madvr OSD are in HDR but the video it self aren't. D3D11 are also only in 8 bit not 10 bit and it says HDR "Unknown properties". Also Fullscreen Exclusive doesn't turn on either with HDR atm.
When comparing the exact same scene from Youtube in HDR, the Youtube video are A lot brighter which is also why I don't think the HDR in working in my Potplayer+Madvr.
OSD (Tab) (https://i.imgur.com/y9HnPgm.jpg)
OSD (Ctrl+j) (https://i.imgur.com/nol04mM.jpg)
Settings:
Identification (https://imgur.com/pSOJ94H)
Properties (https://imgur.com/yS00uiY)
Calibration (https://imgur.com/Ju3867W)
Color&Gamma (https://imgur.com/HSFO482)
HDR
(https://imgur.com/nBGVel6)
Player: Potplayer
Side question: I've keybind zooming/blackbars Profiles but when when switching zooming profiles the whole window changes size not just the video.
Could I maybe run a Fullscreen command after zooming to restore it to fullscreen? But I can't find any Madvr commands.
Video example (https://youtu.be/Bqm0Ut0GSoQ)
(Sorry for spamming, but would really like to get the HDR working.)
Except that this is NOT what happened in this instance. Someone asks for a recommendation for a GPU WITHOUT saying specifically what they intend to do with it (yet mention 4K content, which means HDR, so possibly HDR dynamic tonemapping). So lots of people reply saying that a 1060 is more than enough. That's not a helpful reply. They should *first* ask what the user plans to do with it, including which features (3D LUT? HDR dynamic tonemapping? Need for software decoding?), instead of starting to brag about their low-end card being more than enough to meet whatever limited use they have for it.
What gpu should i buy to play 1080p -> 4k sources and some 4k playback on madvr with decent (medium) settings??
did i miss something?
Manni
2nd July 2019, 18:05
did i miss something?
Yes, you did. :rolleyes:
The OP in the part you quoted said that they also wanted to play 4K content.
99% of the time that means HDR content (UHD Bluray), unless they are only interested in demo files.
Before making suggestions that a 1060 was more than enough, no one asked the OP if they planned to send HDR passthrough to the display/PJ (we don't even know the screen size/sitting distance, which will impact how much processing is needed) or if they wanted to use the HDR dynamic tonemapping with pixel shaders to play their 4K content. And in connection with that, no one asked the OP if they wanted to use any feature that needed software decoding, or if they planned to use madVR for calibration as well, or if they planned to use 8bits or 12bits, or if 50/60p was important or if they only cared about 24p content. All these put together have an impact on performance that could make the difference between frame drops and no frame drops unless PQ is compromised.
The OP might have no idea that some of these features even exist, but my point is to simply bring to their attention that the question in itself doesn't provide enough information to allow anyone (included myself) to make an educated guess that won't lead to frustration or disappointment once they buy the recommended 1060. Will you refund them when they find out that they can only use half to the features offered by madVR or have to compromise PQ more than they intended to if they follow your advice? Or will you blame them for not having provided enough information when they asked the question?
For me, madVR is about getting the best possible PQ. When I bought my 1080ti a couple of years back, I had some headroom. Today, especially with HDR dynamic tonemapping for 4K content, I have none. I would just like to see people who make recommendations to read carefully what's written in a question, and ask for more information before making a recommendation.
Now it looks like some are not happy with this. I have no idea why, but that won't change the reason why I posted. It's because a 1060 is limited in many situations, and I'd rather be sure that a prospective buyer is aware of these limitations before they buy one. If they are and can live with them, fine, they save money and everyone is happy (including me).
I agree with those who say that this exchange has been long enough, so this will be my last post on the subject as I don't have the time nor the inclination to go on discussing this in another thread :).
but a 1060 can do tone mapping with medium setting it can do NGU with medium setting and it can do calibration all at the same time at 60 hz.
BTW. tone mapping with TVs is sadly not that great people are usually not eager to have an 500+ nits SDR image or change the TV settings all the time...
tp4tissue
2nd July 2019, 20:04
On the latest build, 1060 is the minimum for HDR->SDR Tonemapping. You can use 3DLUT, you can't use NGU.
Some users on here consider (no NGU chroma) -A Compromise-, As an enthusiast that is their right.
But 100% of the time actually WATCHING A MOVIE, you can not tell the difference between lanczos chroma and Ngu Chroma.
At more than 2 feet away, on a 50 inch screen, I can't even tell the difference between Nearest neighbor chroma and NGU VeryHigh.
For 1080p SDR movies, and 2160p HDR, 1060gtx is more than enough.
Enthusiasts like to pump the chroma setting because it makes them feel special. In reality it's a waste of electricity. This has been discussed to death since the NNEDI craze.
The only thing about 1060 anyone needs to watch out for is mining cards with low VRM count, these cards wear out and blow up.
It is recommended to get 1070, because who knows what madshi might cook up in the future.. But the politics is of course, now that Madshi is switching to Envy, will Madvr-Fans receive the same updates as before.
xabregas
2nd July 2019, 20:32
did i miss something?
I want to upscale only. i have 4k HDR TV. So no need for tonemapping.
And im on a tight budget. So i can go 1650, rx 570 or 580. 1660 makes me go from 16GB to 8GB, M2 nvme to ssd and a worst case.
I think i can live without NGU performance or VP) hardware decoding. Given my cpu can software decoding a 4k vp9 youtube video at 4k resolution im fine. Ryzen 5 can do that i guess...
Now RX 570 4/8GB or 580 4GB?
And 1650 is that bad??
a 580 should destroy a 1650 if NGU isn't used. it's a really bad card for the price the 1660 is massively faster.
used 580 and 1060 should be relative cheap. madVR is fine with 4Gb "everything" else not so much.
software decoder for VP9 are very good but i never tested VP9 10 bit decoding.
xabregas
2nd July 2019, 21:47
a 580 should destroy a 1650 if NGU isn't used. it's a really bad card for the price the 1660 is massively faster.
used 580 and 1060 should be relative cheap. madVR is fine with 4Gb "everything" else not so much.
software decoder for VP9 are very good but i never tested VP9 10 bit decoding.
ok so rx 580 here i go. And VP9 10 bit 4k decoding even by software with youtube low video bitrates shouldnt be an issue for an hexa core with 12 threads and idk how many cache but almost 20'mb. i guess 20% of cpu??
Alexkral
2nd July 2019, 22:48
BTW. tone mapping with TVs is sadly not that great people are usually not eager to have an 500+ nits SDR image or change the TV settings all the time...
I think we've talked about this before, the main reason is to raise the diffuse white level. This is useful even for 500+ nits TVs, though in this case it's much more justified to follow the PQ to get the HDR as it was conceived. But below that the conversion to SDR works very well, and could work better in my opinion for 250+ nits with some settings that ATM madVR does not provide.
Look at this, unfortunately the video has been deleted:
https://www.liftgammagain.com/forum/index.php?threads/grading-a-bright-sunny-day-in-hdr.10305/
tp4tissue
3rd July 2019, 01:39
a 580 should destroy a 1650 if NGU isn't used. it's a really bad card for the price the 1660 is massively faster.
used 580 and 1060 should be relative cheap. madVR is fine with 4Gb "everything" else not so much.
software decoder for VP9 are very good but i never tested VP9 10 bit decoding.
I've had banding issues with the 580, and the HDR passthrough sometimes doesn't trigger properly.. Madshi may fix everything in the future, but it doesn't make any sense to buy a 580 today, when we know a 1060 WORKS for all the important features.
kitame
3rd July 2019, 08:23
I want to upscale only. i have 4k HDR TV. So no need for tonemapping.
And im on a tight budget. So i can go 1650, rx 570 or 580. 1660 makes me go from 16GB to 8GB, M2 nvme to ssd and a worst case.
I think i can live without NGU performance or VP) hardware decoding. Given my cpu can software decoding a 4k vp9 youtube video at 4k resolution im fine. Ryzen 5 can do that i guess...
Now RX 570 4/8GB or 580 4GB?
And 1650 is that bad??
you know, the other compromise is much better.
2x8GB ➔ 1x8GB would still allow you to add one more stick later on, plus 8GB ram doesn't really affect video quality and your PC would still run just fine.
NVMe ➔ SATA SSD would still allow you to add an NVMe later on, on this note a SATA SSD is just as capable, in practice it would only be a tiny bit slower for non-SSD-to-SSD-filetransfer use.
heck if push comes to shove you could even temporarily go without an SSD if this PC is purely for playing videos, its not like waiting a few more seconds is worse than your eyes rotting right?
as for the case, well thats one thing i wouldn't compromise on.
where as buying a 1650, 570 or 580... you could only sell them at a loss to upgrade later on, worse yet if you can't even sell them off.
so you should ask yourself about your own priorities, is it to save money at the expense of barely scrapping the minimum, or buying less compromising parts at the expense of putting hold on buying other less relevant parts.
chros
3rd July 2019, 09:40
I want to upscale only. i have 4k HDR TV. So no need for tonemapping.
And im on a tight budget. So i can go 1650, rx 570 or 580.
it doesn't make any sense to buy a 580 today, when we know a 1060 WORKS for all the important features.
Agreed, just buy a used (!) 1060 6GB (MSI Gaming series are really low noise/silent as well).
chros
3rd July 2019, 10:33
Here D3D11 copyback is at least 15-20% slower than D3D11 native. It's even more obvious when measuring files with MadMeasureHDR. It takes 10-15mn in D3D11 native with my 1080ti, at least 20-25mn in copyback, and more than 30mn with my iGPU on the MBP 13" 2018.
...
Good for you. It's not meaningless here. That's why I suggest you test a 1080ti with my OS and then report back, instead of claiming that there is no difference.
I already told them the very same here (https://www.avsforum.com/forum/26-home-theater-computers/2364113-guide-building-4k-htpc-madvr-95.html#post58167358) and here (https://www.avsforum.com/forum/26-home-theater-computers/2364113-guide-building-4k-htpc-madvr-96.html#post58177072), asked them for similar graphs, never got a reply :)
the question is what exactly do you want to do with madVR
Almost, but the full (?) question is: what "environment" do you have, at least:
- projector, 720p/1080p TV, 4k HDR TV
-- is it capable of chroma 4:4:4 and will you use it (it can have drawbacks)
- screen size
- sitting distance
- resolution of content and the avg framerate of it
E.g. "only" a 65" 4k HDR TV from ~3 meters has completely different requirements than a 140" 4k PJ SDR screen from 2 meters :)
A lot of people brag about their lower GPU, but they either don't use all the features or they compromise on quality. Or they don't notice frame drops.
...
Clearly a lot of people are happy with a 1060, and I'm very happy they are. But if someone buys a 1060 today, I'd rather be sure that they realise that they are limiting the number of features and the PQ they can get. That's all.
We agree with with "limiting the number of features" but not the "compromise on PQ" part.
Limitation of features (that is not required) is not a problem (e.g. with the B8 I don't need the tonemapping of madvr, I use HDR passthrough, and the result looks waaay better than the other at e.g. 140 nits)
The PQ debate is a different thing: this is what a 1060 6GB can do on a 4k HDR TV with different contents (https://www.avsforum.com/forum/26-home-theater-computers/2364113-guide-building-4k-htpc-madvr-95.html#post58165252).
Now, if you take a look at the 2160p25 and 1080p25 profiles, we can see that the settings are pretty high (not maxed out, of course).
Other refresh rates and/or resolutions (e.g. 720p30) obviously have worse settings, but the question is: do we really care about those? :)
(Note, that I even modded those settings a bit but I haven't updated that post (yet)? due to lack of interest. :) )
It is recommended to get 1070, because who knows what madshi might cook up in the future.. But the politics is of course, now that Madshi is switching to Envy, will Madvr-Fans receive the same updates as before.
I only don't agree with this part: there's no such thing as future proofing in the world of PC, never was :) I just buy what (I think) I need at the moment.
ryrynz
3rd July 2019, 11:18
Sometimes only by seeing what you miss out on do you realize it's worth upgrading.
Anyway, this got out of hand quickly, just throw a model out there don't overthink it, nobody is going to sue you for recommending a 1060 when they eventually decide they need a 1080+.
But yes you'll need a Fosmon DP to HDMI cable. I did the former as 9 bucks was better than hundreds on a new receiver.
Yeah, I said screw it and bought an adapter from work, it's up and running now, cheers, getting DTS instead of HD MA was kinda annoying and obviously much lower quality.
I found a nice amp online I could've bought but it would've meant spending a ~$800 so this'll do for a while.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.