madVR - high quality video renderer (GPU assisted)

[cyberbeing]

My reasons are essentially opposite of yours.

I use ED1 mono static explicitly since it's not completely transparent and adds a bit of texture (similar to static grain) to the image, which I find benefits clean HD sources the most visually, especially animation. ED1 also has the lowest tendency of producing undesired dithering patterns and artifacts, at the expense of a slightly higher noise floor. This noise floor in turn makes Dynamic mono a poor fit for ED1, since it adds too much energy to motionless and flat content which can be distracting.

ED2 on the other hand has a lower noise floor, yet has a tendency to produce minor patterns and artifacts. Dynamic is needed to counteract this. With SD and below, I'd rather have a transparent dither without any visible texture. While quality of SD is usually low enough, that I wouldn't notice ED2 mono dynamic even if I were looking for it.

[luk008]

I'm using Error Diffusion Option 1 in madvr, but acessing the LAV Video Decoder Menu there's also an option for dithering mode (in this case I'm using ordered dithering). Is there a way to disable the LAV dithering and use only the madvr dithering?

[tobindac]

Can't those users just turn off motion interpolation on their tv? Seems to me there are more people who dislike things like SVP than favor it.

In my experience internet comments tend to be "trollish". That means if someone doesn't have a PC that can handle SVP, will likely post negative comments about it anyway. A good 1080p source will not be able to be played properly on a common 5 year old PC (let alone on an average HTPC that various users in this forum have), especially if madvr and svp quality settings must be set to the better algorithms.

e.g. I noticed that on overclock.net, where most people have insanely good GPUs and CPUs, they are often in favor of SVP+madvr.

Also sometimes it's only purism. I know there are rare artifacts, but the slideshow of 24 fps is multiple times more awful.

[sneaker_ger]

I'm using Error Diffusion Option 1 in madvr, but acessing the LAV Video Decoder Menu there's also an option for dithering mode (in this case I'm using ordered dithering). Is there a way to disable the LAV dithering and use only the madvr dithering?
LAV only uses dithering when it does a colorspace conversion itself. It lets madVR do it unless you turn off the respective output format in the LAV Video settings.

[luk008]

LAV only uses dithering when it does a colorspace conversion itself. It lets madVR do it unless you turn off the respective output format in the LAV Video settings.

In the YUV -> RGB conversion I leaved it as "untouched" in the LAV options. MadVR is set to use PC levels. Is this the best choice?

[sneaker_ger]

Just leave everything at the default value. Do not uncheck any of the output formats.

[jkauff]

Now that madshi has given us 64-bits in madVR, I wonder how close he's getting to a 1.0 release where we can finally give him some money for all his years of hard work.

[Arm3nian]

Goddammit, Arm3nian... give up already. You just don't get it.
Go to that link and read it again. All of it!
Then, if you still think that huhn is wrong, read it again. And again, until you realise what he's talking about.
Stop dragging this, it's pointless and off-topic, and it just wastes everyones time, especially madshi's.

Apologies to everyone, I just couldn't stand him anymore... I'll shut up now.

Why are you so angry? And why is this off topic?

Huhn thinks that there are pitch issues when watching pal content with SM. I'm telling him that there aren't, and there isn't a need for reclock when using SM. Why is this concept so hard for you to grasp?

In my experience internet comments tend to be "trollish". That means if someone doesn't have a PC that can handle SVP, will likely post negative comments about it anyway. A good 1080p source will not be able to be played properly on a common 5 year old PC (let alone on an average HTPC that various users in this forum have), especially if madvr and svp quality settings must be set to the better algorithms.

e.g. I noticed that on overclock.net, where most people have insanely good GPUs and CPUs, they are often in favor of SVP+madvr.

Also sometimes it's only purism. I know there are rare artifacts, but the slideshow of 24 fps is multiple times more awful.

I'm not talking about SVP, I'm talking about things like SVP, motion interpolation, used in televisions. Sometimes they aren't well implemented, and people don't like it. SVP works fine with SM. It was mentioned that there is ghosting with SM on some TVs, first with no proof, and second with disregarding the fact that plasmas flicker at 24hz, so reclock isn't that much of a benefit compared to running at 60hz with SM.

24fps looks fine for most content. If you want higher fps content, blame the content creators. Motion interpolation is like upscaling, it helps sometimes, but its tied to the quality of the source, you can't create information that isn't there.

[pankov]

Arm3nian,
did you listen to the examples from the PAL speedup page (http://sandbox.slysoft.com/palspeedup/index.html)?
Do you hear the differences?
If not then there is nothing we can do to explain this better to you but if you do then you'll understand what we have to live with in our PAL countries.
The speed up is not something that's done by us or by our gear - it's done in the studios and reclock is the only software solution to undo it so we can listen to the tracks as it was intended by the authors.

I hope you'll finally understand it and stop this discussion in this thread.
SM has a lot of benefits and I do use it but for totally different reasons and you have to accept that there are situations/persons that either don't like it (visually) or it doesn't fit their needs ... and that's life.

[Arm3nian]

So why can I watch 25fps content on a 60hz screen with SM and get no pitch issues?

[Qaq]

there isn't a need for reclock when using SM. Why is this concept so hard for you to grasp?Because you're wrong?
madVR v0.86.0 released

* added smooth motion frame rate conversion algorithm
...
...Consequently the motion smoothness depends on proper timestamps. If the timestamps (or audio clock) contain jitter, the playback will contain jitter, too. So even if Reclock might not be needed to avoid frame drops/repeats, anymore, when using madVR's new FRC algorithm, you might still want to use Reclock, because it provides a stable and reliable audio clock with very low jitter...

[Arm3nian]

LOL. Because reclock magically generates a clock in software right?

Reclock gets the clcok the from the crystal on your board, like other renderers. That was a problem in windows XP.

[Qaq]

I can't speak for madshi. You better tell him he was wrong.

[Asmodian]

So why can I watch 25fps content on a 60hz screen with SM and get no pitch issues?

With PAL speedup you get the wrong pitch watching 25 fps on a 50 Hz screen or even only listening to the audio. The studio did the speed up and Reclock offers a speed down option, which madVR also supports, to play 25 fps video at 24 fps. The display's refresh rate does not come into it.

[Schwartz]

I noticed that using 64-bit madVR with the appropriate MPC-HC and internal LAV Filters, NNEDI3 upscaling is not working at all. No matter which sub-settings I pick. Milisecond count however (rarely) spikes up in the renderer as if it were active. Graphics card is a R9 270X.

Also: Thanks for making a 64-bit version finally! That should be huge for scaler performance.

[vivan]

Also: Thanks for making a 64-bit version finally! That should be huge for scaler performance.All processing is done on GPU, there's no benefit for madVR (but there's for playback chain - currently x64 HEVC decoder is much faster).

[Thunderbolt8]

this is my last word about the pal speed issue

http://sandbox.slysoft.com/palspeedup/index.html

madVR SM can't fix this.cant people just whine at slysoft forums for them to put out a x64 version of reclock? I guess thats all that is needed, right?

[Arm3nian]

I can't speak for madshi. You better tell him he was wrong.

Ugh madshi was correct in his statement, just like everything he says. That post is just old, and doesn't apply anymore. There is a clock for your cpu, gpu, and audio. Other renderers can access the same clock that reclock can. Madshi has stated often that reclock is not needed for SM. If you don't believe me, try SM with reclock and without, there is no difference in IQ...

With PAL speedup you get the wrong pitch watching 25 fps on a 50 Hz screen or even only listening to the audio. The studio did the speed up and Reclock offers a speed down option, which madVR also supports, to play 25 fps video at 24 fps. The display's refresh rate does not come into it. What content actually suffers from this? 24fps or 25fps works fine on 50hz or 60hz. Those with Pal televisions can get the NTSC content and watch it with SM. Same with those with NTSC monitors who want to watch pal content. Not all content is sped up or slowed down. What is actually released today that is only available in a pitch alretered version?

[kalston]

There are plenty of DVDs sped up to 25fps (from 24) that require Reclock or JRiver's equivalent to watch them properly. Otherwise you're just watching a sped up film.

Maybe recent releases don't suffer from this, I have no clue, I moved to blu-ray long ago and rarely watch DVDs (just for those few titles that aren't released on blu-ray) but I've had to deal with more than a few sped up ones.

[Arm3nian]

Fair enough, that's a valid reason to use reclock, even though my original argument was about not needing reclock if using SM for non altered content. Most of those DVDs are remastered or available on bluray anyway as you said.

[e-t172]

Ugh madshi was correct in his statement, just like everything he says. That post is just old, and doesn't apply anymore.

I agree, and in addition, I wonder if it ever applied to anything, even on XP. I think madshi was just being overly cautious by making clear that the clock needs to be high quality for SM to work correctly, but there's no evidence that people encountered bad DirectShow clocks. I certainly don't remember anyone on this thread reporting an issue with SM that could be traced back to a bad clock.

[XMonarchY]

nope reclock is not nneded for this but not this tapic again...



rendertimes? FSE shouldn't add noticeable rendertimes.
and there are some questionable settings but this is not a madVR setup guide.

Render time without FSE is about 24.5ms and the same for Max Stats, but with FSE turned on, I also get avg. 25ms, but Max Stats show 260ms and of course constant frame drops...

[dansrfe]

I would use MPC-HC Renderer for WASAPI but I get crackling noises randomly while using it for some reason. Sometimes the nosies go away when skipping around the file.

Using DirectShow for now.

[bcec]

I noticed that using 64-bit madVR with the appropriate MPC-HC and internal LAV Filters, NNEDI3 upscaling is not working at all. No matter which sub-settings I pick. It does however sometimes show a high milisecond count in the renderer as if it was active. Graphics card is a R9 270X.


I don't have an ATI card, but NNEDI3 is certainly working for me for the x64 stack (I have a particular video where NNEDI3 on/off makes a big difference). How did you verify that NNEDI3 was not working?

[heiseikiseki]

HI，I have very high present time in my optimus laptop.
about 8~10ms

That make my gtx860m can only handle jinc.

Sorry for my bad English.

[huhn]

Render time without FSE is about 24.5ms and the same for Max Stats, but with FSE turned on, I also get avg. 25ms, but Max Stats show 260ms and of course constant frame drops...

there is something not working as intended. hard to say what but this shouldn't happen.
can you try the old FSE mode?

you get the old FSE mode by disabling present several frames in advance under exclusive mode settings.

[huhn]

HI，I have very high present time in my optimus laptop.
about 8~10ms

That make my gtx860m can only handle jinc.

Sorry for my bad English.

and the PRESENT times rise with settings like jinc 3 ar or other settings?
is aero active?
do you get the same problem in fullscreen exclusive mode?

[heiseikiseki]

and the PRESENT times rise with settings like jinc 3 ar or other settings?
is aero active?
do you get the same problem in fullscreen exclusive mode?

Setting seems has no relationship with the PRESENT times.
Also in FSE mode I get the same PRESENT time.
I’m using windows 8.1 so no areo active.

This problem is also mentioned before:
http://forum.doom9.org/showthread.php?t=170937

[huhn]

Setting seems has no relationship with the PRESENT times.
Also in FSE mode I get the same PRESENT time.
I’m using windows 8.1 so no areo active.

This problem is also mentioned before:
http://forum.doom9.org/showthread.php?t=170937

on windows 8 and newer aero is "always" active.(there are ways to still disable it but let's be real...)
present times can rise when aero is disabled that's why i asked.

so yes this looks like a yet another optimus issue. sorry i'm pretty bad on this topic. i hope someone with an optimus laptop knows a work around.

[Anime Viewer]

HI，I have very high present time in my optimus laptop.
about 8~10ms

That make my gtx860m can only handle jinc.

Sorry for my bad English.

Setting seems has no relationship with the PRESENT times.
Also in FSE mode I get the same PRESENT time.
I’m using windows 8.1 so no areo active.

This problem is also mentioned before:
http://forum.doom9.org/showthread.php?t=170937

If you're using MPC-HC did you rename the .exe file to something like mpc-hcnv.exe, mpc-hc64.exe, mpc-hc2.exe, etc, then open Nvidia control panel and associate that file to using the Nvidia GPU, then launch a file open the registry and confirm that that it made an nvidia related directory inside the OpenCL key? (ex: HKEY_CURRENT_USER\Software\madshi\madVR\OpenCL\GeForce GTX 680M). In your case it should read 860m.

What your settings are should definitely effect your present times. Besides telling us all the upscaling/downscaling settings you have set we would also need to know :
What dithering option do you have selected (none, random, ordered, ED1, ED2)? Are you trying to use NNEDI3 for resolution doubling and/or quadrupling? Are you using any hardware acceleration options in LAV Video? (If so, what?)

As a test try setting image upscaling and downscaling both to DXVA2 in madVR settings. That will effectively disable any software processing and NNEDI3 settings you have selected and use the GPU. With that setting you should have very low present times.

Are there any other video/audio altering programs you're trying to run with it (SVP, Reclock, etc)?

While a video is running if you right click on it, and select filters what does it show running? (Ex:
Filters currently loaded:
- Default DirectSound Device
- XySubFilter (=> madVR)
- madVR
- LAV Audio Decoder
- LAV Video Decoder
- LAV Splitter Source
)

[Schwartz]

I don't have an ATI card, but NNEDI3 is certainly working for me for the x64 stack (I have a particular video where NNEDI3 on/off makes a big difference). How did you verify that NNEDI3 was not working?

Comparing render ms, looking at the Ctrl-J chroma > and image > lines and zooming in to compare the pictures while the regular upscaler was set to nearest neighbor. I have separate scaler profiles for SD and HD respectively (only SD uses image doubling), and switching between them showed no discernable difference.

There's no error message and otherwise it works flawlessly. Image doubling is just flat-out ignored.

[heiseikiseki]

If you're using MPC-HC did you rename the .exe file to something like mpc-hcnv.exe, mpc-hc64.exe, mpc-hc2.exe, etc, then open Nvidia control panel and associate that file to using the Nvidia GPU, then launch a file open the registry and confirm that that it made an nvidia related directory inside the OpenCL key? (ex: HKEY_CURRENT_USER\Software\madshi\madVR\OpenCL\GeForce GTX 680M). In your case it should read 860m.

What your settings are should definitely effect your present times. Besides telling us all the upscaling/downscaling settings you have set we would also need to know :
What dithering option do you have selected (none, random, ordered, ED1, ED2)? Are you trying to use NNEDI3 for resolution doubling and/or quadrupling? Are you using any hardware acceleration options in LAV Video? (If so, what?)

As a test try setting image upscaling and downscaling both to DXVA2 in madVR settings. That will effectively disable any software processing and NNEDI3 settings you have selected and use the GPU. With that setting you should have very low present times.

Are there any other video/audio altering programs you're trying to run with it (SVP, Reclock, etc)?

While a video is running if you right click on it, and select filters what does it show running? (Ex:
Filters currently loaded:
- Default DirectSound Device
- XySubFilter (=> madVR)
- madVR
- LAV Audio Decoder
- LAV Video Decoder
- LAV Splitter Source
)


I'm using MPC-BE and madvr only.
Decode only use CPU,
(becasue if i use DXVA2 on 720p->1080p(jic3AR/jinc3AR) it will cause dropped frames too.)
And I'm sure that I did open the Nvidia control panel to that the MPC-BE to use the Nvidia GPU.
I've checked the registry,that's right.

I am using Ordered dithering .


I’ve tried NEDDI3 16 with
chroma upscaling:Bicubic75
image upscaling:Bicubic75
on sd->1080p often no promblems
but on 720p->1080p
Even the rendering time is only 21ms,
but the PRESENT time is about 10ms
frame drops (render and present queue not getting filled fast enough)





when 720P->1080P
http://i.imgur.com/pl7dTmL.jpg


Sorry for my bad English.

[agustin9]

I would try lowering the queues...

[Anime Viewer]

I'm using MPC-BE and madvr only.
Decode only use CPU,
(becasue if i use DXVA2 on 720p->1080p(jic3AR/jinc3AR) it will cause dropped frames too.)

I am using Ordered dithering .


I’ve tried NEDDI3 16 with
chroma upscaling:Bicubic75
image upscaling:Bicubic75
on sd->1080p often no promblems
but on 720p->1080p
Even the rendering time is only 21ms,
but the PRESENT time is about 10ms
frame drops (render and present queue not getting filled fast enough)

Sorry for my bad English.

Things like NNEDI3 can be more taxing on 720p->1080p compared to sd->1080p. What scaling factor do you have set in the doubling screen? Keeping the BiCubic settings you mentioned try with only luma doubling checked, and with the "only if scaling factor is 2.0x (or bigger)" selected.
Does that make any difference?
Your target rectangle is larger that 1080p, so its actually doing more upscaling and downscaling that it would for 1080 resolution. (Is 1920x1140 the native resolution of your screen?)

When you're watching in a default 720p window size what type of numbers are you seeing (half the present ms of fullscreen, or more drastic difference)?
Are you noticing the same numbers in both Full Screen Exclusive and Full Screen Windowed mode, or are they different?
Under trade quality for performance in madVR settings what (if anything) do you have checked?
Are you watching something with subtitles, and (if you are) do you see a difference decoder (aka CPU) at 12, upload and render (aka GPU) at 6, and present at 8. (Usually if your present several frames in advance is too high you'll see presentation glitches, but that doesn't seem to be a problem given your screen shot. However you may want to uncheck the present frames in advance option as a troubleshooting test).

[JarrettH]

Back when we were testing the various dithering algorithms, it did seem like more people preferred color vs luma-only dithering noise. Though there was a group of us, including myself, who really disliked the idea of intentionally adding color noise to otherwise pristine chroma, and would rather have the light texture added by luma-only dithering. It's all a trade-off. The type of person who likes grainy movies and hates the appearance of grain-free denoised movies, would likely prefer luma-only dithering. The type of person who likes when their movies have (and keep) a smooth denoised appearance, would likely prefer color dithering.

Personally I like:
HD: Error Diffusion, Option 1, Luma, Static
SD: Error Diffusion, Option 2, Luma, Dynamic
Performance: Ordered Dithering, Luma, Dynamic
Other: Random Dithering, Color, Dynamic


The most popular option favored by others for color dithering I believe was:
Error Diffusion, Option 2, Color, Dynamic

The luma noise lends to smoother gradients IMO. :thanks:

[omarank]

Back when we were testing the various dithering algorithms, it did seem like more people preferred color vs luma-only dithering noise. Though there was a group of us, including myself, who really disliked the idea of intentionally adding color noise to otherwise pristine chroma, and would rather have the light texture added by luma-only dithering. It's all a trade-off.
I too prefer mono dithering.

Normally madVR does dithering at the last step when the video is converted to 8 bits (or lower bitdepth). However, when doing DXVA deinterlacing on a video that is more than 8 bits, madVR dithers the video to 8 bit before it is deinterlaced. I was wondering whether madVR uses the selected dithering algorithm for this dithering step or it uses Random dithering (Colored Static?).

[heiseikiseki]

Things like NNEDI3 can be more taxing on 720p->1080p compared to sd->1080p. What scaling factor do you have set in the doubling screen? Keeping the BiCubic settings you mentioned try with only luma doubling checked, and with the "only if scaling factor is 2.0x (or bigger)" selected.
Does that make any difference?
Your target rectangle is larger that 1080p, so its actually doing more upscaling and downscaling that it would for 1080 resolution. (Is 1920x1140 the native resolution of your screen?)

When you're watching in a default 720p window size what type of numbers are you seeing (half the present ms of fullscreen, or more drastic difference)?
Are you noticing the same numbers in both Full Screen Exclusive and Full Screen Windowed mode, or are they different?
Under trade quality for performance in madVR settings what (if anything) do you have checked?
Are you watching something with subtitles, and (if you are) do you see a difference decoder (aka CPU) at 12, upload and render (aka GPU) at 6, and present at 8. (Usually if your present several frames in advance is too high you'll see presentation glitches, but that doesn't seem to be a problem given your screen shot. However you may want to uncheck the present frames in advance option as a troubleshooting test).

The resolution of my display is 1920*1200. (I also tried a FHD display. That would be the same result.）

And I set the scaling factor to 1.5X.

YES，I've tried the exclusive mode too. Nothing changed，the Present time is still about 8~9ms.（I don't know how to do a printscreen in exclusive mode so I use windows mode（new path）to attach a picture.）

I've tried to choose all the Trade quality for performance setting， seems no relationship of the PRESENT times.
So now I only checked "don‘t analyze gradient angles for debanding".


I've tried to change the numbers of present frames and lower the queues，too. Also tried old path. and other option like overlay.
Also nothing changed with the PRESENT time.

The full screen mode seems double the present time.

When default 720P window size the PRESENT time is about 4ms，and the 1080P is about 8~10ms.

I dont't use subtitles.

Here are some attched picture with the 720P.



720P in window size
http://i.imgur.com/7hrKmDI.jpg
720P in full screen（no scaling to 1080p）
seems strange. The PRESENT time also doubled.
http://i.imgur.com/eNrtTM2.jpg

[6ari8]

The resolution of my display is 1920*1200.

And I’ve set the scaling factor to 1.5X.

YES，I've tryed the exclusive mode too. Nothing changed，the Present time is still about 8~9ms.（I don't know how to do a printscreen in exclusive mode so I use windows mode（new path）to attach a picture.）

I've tride to choose all the Trade quality for performance setting， seems no relationship of the PRESENT times.
So now I only checked "don‘t analyze gradient angles for debanding".


I've tried to change the numbers of present frams and lower the queues，too. Also tried old path. and other option like overlay.
Also nothing changed with the PRESENT time.

The full screen mode seems double the present time.

When default 720P window size the PRESENT time is about 4ms，and the 1080P is about 8~10ms.

I dont't use subtitles.

Here are some attched picture with the 720P.



720P in window size
http://i.imgur.com/7hrKmDI.jpg
720P in full screen（no scaling to 1080p）
seems strange. The PRESENT time also doubled.
http://i.imgur.com/eNrtTM2.jpg

Just want to mention something that might help you.

I have an M14x R2 Optimus laptop with an HDMI and a miniDP port. I remember having a similar problem to yours when I was using the HDMI port but not when using the miniDP port. It turned out that the HDMI port is wired to the Intel GPU while the miniDP was wired to the Nvidia GPU. I'm not sure why it had a high present time only when connected to the iGPU but I solved my issue by using the miniDP.

[heiseikiseki]

Just want to mention something that might help you.

I have an M14x R2 Optimus laptop with an HDMI and a miniDP port. I remember having a similar problem to yours when I was using the HDMI port but not when using the miniDP port. It turned out that the HDMI port is wired to the Intel GPU while the miniDP was wired to the Nvidia GPU. I'm not sure why it had a high present time only when connected to the iGPU but I solved my issue by using the miniDP.
Thanks, but I'm using the DISPLAY port and I also tried the HDMI port.
Nothing changed.Seems In my computer(ASUS ROG GR8), Both of them are wired to IGP controller.

[Anime Viewer]

I'm using MPC-BE and madvr only.
Decode only use CPU,


Are you using the latest Nvidia drivers (350.12) with OpenCL 1.2, or an older driver with OpenCL 1.1?

You could try downloading, installing, and setting up (disabling internal and setting external to preferred) the latest LAV filters. Extending on that tell us exactly what your filters are named when you're running a video and right click it selecting filters. Instead of none for hardware decoding try with CUVID selected.

If you haven't already try putting a check in the use a separate device for presentations box in the general settings.

Do you have v-sync forced to enabled or disabled in Nvidia control panel settings general and mpc-be linked areas?

[heiseikiseki]

Are you using the latest Nvidia drivers (350.12) with OpenCL 1.2, or an older driver with OpenCL 1.1?

You could try downloading, installing, and setting up (disabling internal and setting external to preferred) the latest LAV filters. Extending on that tell us exactly what your filters are named when you're running a video and right click it selecting filters. Instead of none for hardware decoding try with CUVID selected.

If you haven't already try putting a check in the use a separate device for presentations box in the general settings.

Do you have v-sync forced to enabled or disabled in Nvidia control panel settings general and mpc-be linked areas?


Yeah，I'm using 350.12 and the latest intel GPU driver.

Also，I've tried MPC-HC or MPC-BE+LAV with CUVID，
still nothing changed.



This is another attached picture.
(MPC-BE+the latest LAV FILTER） v-sync forced to enable.
use a separate device for presentations box checked.


http://i.imgur.com/E8XhDkX.jpg

[Kirk Lazarus]

Hi Guys,
in terms of performance the PCI-Express version is important with madVR?

I'm going to buy a new GTX970 (PCI-e v.3) but my motherboard is only PCI-e v.2 ... So my fear is that the old version of PCI-e of the motherboard may reduce performance with madVR....

Can someone clear my doubts?

thank you all

[huhn]

this is a huge problem with AMD not nvidia.

pcie 2.0 shouldn't really matter for madVR.

[Kirk Lazarus]

this is a huge problem with AMD not nvidia.

pcie 2.0 shouldn't really matter for madVR.

Can you explain me better why AMD cards suffers of PCI-e V.2 on motherboard when used with madVR?

many thanks

[huhn]

Can you explain me better why AMD cards suffers of PCI-e V.2 on motherboard when used with madVR?

many thanks


they have a copyback issue with nnedi3.

when doing a directx <-> openCL interop AMD is copying the data from the GPU to the system ram and back to the GPU this is using PCIe and this can be limited with an older PCIe version PCI 1.0 would work like a handbrake for this.

nvidia doesn't do that they keep it at the GPU so PCIe speed doesn't really matter only the first upload to the GPU needs PCIe speed but even PCIe 1.0 is more than enough for this.

[bcec]

Hi Guys,
in terms of performance the PCI-Express version is important with madVR?

I'm going to buy a new GTX970 (PCI-e v.3) but my motherboard is only PCI-e v.2 ... So my fear is that the old version of PCI-e of the motherboard may reduce performance with madVR....

Can someone clear my doubts?

thank you all

huhn explained it perfectly, but to clear your doubts further, I have a pcie 2.0 + 970, and madvr nnedi3 performance is excellent for 1080p->2160p upscaling, especially after the newest drivers.

Get a blower type 970.

[Jtacdf]

For any of you guys that has problem getting x64 nnedi3 working with AMD gpu, a temporary workaround (I hope) is to rename your media player exe to "ForceSingleGPU.exe".
I've gotten it working without issues that way.

Disregard the above, the issue persist again.

[Schwartz]

Just tried it, didn't work for me. But I'm fine with having image doubling off until such a time that it's fixed. I just wanted to bring the issue to the dev's attention and see if anyone else had it.

[khanmein]

Hi Guys,
in terms of performance the PCI-Express version is important with madVR?

I'm going to buy a new GTX970 (PCI-e v.3) but my motherboard is only PCI-e v.2 ... So my fear is that the old version of PCI-e of the motherboard may reduce performance with madVR....

Can someone clear my doubts?

thank you all

for gaming it will drop around 5-10% or even more depending games. i suggest u go for GTX960 or buy new rig.

[StinDaWg]

Folks is it me, or the latest nVidia drivers sped up madVR NNEDI3 processing quite a bit? I saw roughly a 5-6ms processing time reduction, and I can now do NNEDI3x64 (up from x32) image doubling with my 970 for 1080p->2160p while keeping everything else the same.
Can confirm. I have a GTX 960 and with 720p60->1080p60 scaling, I went from 15ms to 10ms render time with the same settings (16 neurons). GPU usage is about 10-15% lower. Before it was borderline but now it's a piece of cake. Seems it's equal to AMD now if not better.