madVR - high quality video renderer (GPU assisted)

[pankov]

madVR lets the decoder run faster than it normally does. So the decoder might be confused. Not sure exactly if that influences the graph's "playback position". If so, there's not much I can do about it, or else the whole queue logic would not work.

How do you know that ZP does not create a 2nd EVR instance, as well?

So the problem only occurs when using seamless playback with playlists?
I found what's causing the problem. It's an option "seamless playback" which is supposed to work only in VMR9 but obviously is used on madVR too ... but not on EVR and that made me think it's madVR's fault.
Sorry!
I knew that ZP didn't create a second EVR instance because it didn't open the second file while playing the first when using EVR. I used Porcess Explorer's lower pane view and watched the opened file handles. With EVR it closes the handle to the first file and then opens a handle to the second one. With madVR ... and VMR9 ... it is opening the second handle a few seconds before closing the first one.
I've reported the problem at the ZP's forum and I hope it will be fixed on their side ... at least they should make it clear that it's not only VMR9 that get's this behavior. In the meantime it will be nice if you can fix it on your side even though it's not a big issue right now that I know how to turn it off.


It's only for informational purposes. madVR strictly only uses the self-calculated display refresh rate information (first line in the OSD).
OK, that's a relieve.

Now I found another incompatibility/issue with ZoomPlayer:
As you can see in the attached screenshots there are a few missing values in the Media info window when madVR is used. If I use EVR, VMR9, Haali or Overlay mixer everything is OK. I'm sure madVR has the values for these fields but it simply doesn't expose them in the way other renderers do.
It's not a big issue but I think it needs only a small fix and we'll have all the info in ZP.
http://img818.imageshack.us/img818/7466/mediainfoevr.th.png (http://img818.imageshack.us/i/mediainfoevr.png/)http://img171.imageshack.us/img171/115/mediainfomadvr.th.png (http://img171.imageshack.us/i/mediainfomadvr.png/)

[janos666]

There are more demanding algorithms I'm planning for future versions (don't ask me what kind of algorithms and when)

This would be a suggestion (or more like an idea), not a question. :rolleyes:

Did you ever consider to implement some kind of motion analysis based frame interpolation for 60Hz displays with 23,976 fps movies?

I have played with MVTools and I found it useless but I have to use the simplest methods with a Q6600@3.6Ghz and a GPU is ~10 times faster than a CPU (with parallel operations). And I want to do some smarter thing than making a lot of interpolated frames.
I would like to make six interpolated frames only, spaced equally between the original frames. As much as it can be equable when you have 6 frames to place them in 23 gaps but the choice gaps can float.

It is only an idea. I could never test it. I can't predict the positive or negative effects either...

[pankov]

madshi,
this night I've been watching some episodes of 24 (I know I'm late, but I was saving it for good weekend marathon ;) ) and I had a lot of stutter which was not indicated as dropped or delayed frames in the OSD. Just as when I was about to give up on madVR for the night I noticed something strange in the OSD - my present times were through the roof while I was sure I've seen them in <1ms range. As you can see from the attached screenshots I get totally different results for the same file with the same madVR settings and the one with >20ms present times is the one that stutters every 10 or so seconds.
This made me do some testing and I found a way to reproduce it and a way to fix it. If start the playback and continue watching I get bad times but if I quickly reload the same file everything is fine
:confused:
I made some log files and I hope you'll be able to find the reason in them. If not I'm pretty sure I can reproduce it so if you want me I can do more tests.

two log files:
http://www.mediafire.com/?nmejgjajkjo

screnshots:
http://img171.imageshack.us/img171/559/settingsn.th.png (http://img171.imageshack.us/i/settingsn.png/) http://img101.imageshack.us/img101/1711/badtimes.th.png (http://img101.imageshack.us/i/badtimes.png/) http://img641.imageshack.us/img641/4427/goodtimesj.th.png (http://img641.imageshack.us/i/goodtimesj.png/)

Edit:
After I posted the problem I remembered that someone here in the forum said that those <1ms times are fake reports from Aero which made me think what if I stop using "aero" timing mode and go back to "normal". I'm happy to report that this fixed it ... or at least I'm not able to reproduce those high present time values. Now I get ~6ms present times which lead to a perfectly smooth 40 minutes of playback.
So the question now is: what does "aero" mode do different from "normal" mode and that's not working every time?

[namaiki]

Re:I'm not quite sure what to say. I think that for MadVR 0.18 and newer, ~24fps on ~60Hz seems a lot more juddery than before (0.17).

Weird. Then I don't know for sure why v0.17 would work better for you than newer builds. Maybe it's the frame drop/delay logic. I plan to improve that a bit in a future version. So there is still some hope to get playback smoothness back to v0.17 levels for you. Seems that you are the only one having best results with v0.17, though. Most other people seem to prefer v0.18 or v0.21.

Yeah, I noticed that I was the only one who seems to prefer v0.17, but v0.18 is certainly more juddery for me. This is only for ~24fps@60Hz. ~30fps@60Hz is perfectly smooth on the newer releases on Aero timing for Intel card (screen is ~61.8Hz) and normal/high for the nvidia card (screen is ~59.6Hz). Both report Aero as 60Hz, as far as I can tell.

[Porsche_fan]

Madshi, I have been wanting to use your renderer for a while but my HTPC wouldn’t support it. I was hoping maybe to get yours and possibly other forum member’s advice on how to best allocate my limited budget for a new HTPC.

I have been upscaling dvd’s using MPC HC and ffdshow through VGA (1280x1024) to my 50” plasma. I want to upgrade my current HP core 2 duo with integrated graphics to either a i5 or i7 with a dedicated video card, allowing 1920x1080 resolutions through HDMI.

I tend to use 3 ffdshow filters, resize lanczos2 to naïve screen resolution, unsharp mask on the resized image, and occasionally avisynth filters in realtime like SeeSaw too. I will be using these filters on DVD’s, 720p/1080p website based VC-1/h.264 content and Blu-ray eventually…I won’t use the computer for gaming btw.

As far as I can tell ffdshow filters and decoding aren’t done on the graphics card unless through DXVA. If I understand correctly, DXVA allows the video card to decode but wouldn’t allow my other ffdshow to be used. Therefore, I wouldn’t be interested in using DXVA.

I am interested in obtaining the highest quality video image. However, I’m not really sure how a dedicated graphics card would function with my video decoding and filters by ffdshow or how a dedicated graphics card would affect the video’s image quality.

It seems to me, most of the work is done by the CPU and very little by a GPU. Therefore, I see little benefit spending a lot of my budget on an expensive graphics card and should instead spend it on a better CPU.

I realize however, I may be missing some important information about how all these components work together. So I was wondering if I understand this correctly and have come to the right decision…to buy a basic graphics card with HDMI out and put the rest towards a better CPU and memory.

One other thing...the latest ATI cards support DirectX 11. How important would this be to renderering in the future and to my HTPC's functions?

I was thinking of buying another HP but any recommendations on manufacturer, CPU, memory or video cards, I would greatly appreciate as well.

Btw, these are the HP graphic cards available listed from least to most expensive. If you have any advice as to which might be best, that would be extremely helpful…

512MB ATI Radeon HD 5450 [DVI, HDMI, VGA adapter]

512MB NVIDIA GeForce 315 [DVI, HDMI, VGA]

1GB ATI Radeon HD 5450 [DVI, HDMI, VGA]

1GB ATI Radeon HD 5570 [DVI,HDMI, DP, VGA adapter]

1GB NVIDIA GeForce GT 220 [DVI, HDMI, VGA]

1GB NVIDIA GeForce GT 320 [DVI, HDMI, VGA]


Thanks in advance for any info and help.

[namaiki]

Porsche_fan: part of the point of MadVR is to allow the GPU to do the resizing of the video. It has support for a number of resizing algorithms, ie softcubic, lanczos, spline.

Perhaps you could go for:
1GB NVIDIA GeForce GT 320 [DVI, HDMI, VGA]
or
1GB ATI Radeon HD 5570 [DVI,HDMI, DP, VGA adapter]

If you want to be able to drive 1080p with ease.

You will still need a half decent CPU if you want to do post processing, though honestly, postprocessing is something I try to avoid.

[ryrynz]

Colors are messed up on this vid, link to sample provided.
Colors are fine with EVR.

http://www.fileden.com/files/2006/10/26/319696//sample.wmv

[namaiki]

http://www.fileden.com/files/2006/10/26/319696/ff7053500k (3).wmv

Seems fine for me even using all MPC-HC's internal filters. What filters are you using, and what are the rest of your system's specs?

[cyberbeing]

I see the same problem as ryrynz with his sample when using the Windows 7 built-in WMVideo Decoder DMO.

Using FFDshow instead doesn't exhibit the problem.

It appears to be yet another madVR problem with aspect ratio signaling.

Crazy diagonal shifted chroma:
- Connection media type:

Video: YV12 852x480 29.97fps

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Video {73646976-0000-0010-8000-00AA00389B71}
subtype: MEDIASUBTYPE_YV12 {32315659-0000-0010-8000-00AA00389B71}
formattype: FORMAT_VideoInfo2 {F72A76A0-EB0A-11D0-ACE4-0000C0CC16BA}
bFixedSizeSamples: 1
bTemporalCompression: 0
lSampleSize: 613440
cbFormat: 112

VIDEOINFOHEADER:
rcSource: (0,0)-(852,480)
rcTarget: (0,0)-(852,480)
dwBitRate: 0
dwBitErrorRate: 0
AvgTimePerFrame: 333667

VIDEOINFOHEADER2:
dwInterlaceFlags: 0x00000000
dwCopyProtectFlags: 0x00000000
dwPictAspectRatioX: 852
dwPictAspectRatioY: 480
dwControlFlags: 0x00000000
dwReserved2: 0x00000000

BITMAPINFOHEADER:
biSize: 40
biWidth: 852
biHeight: 480
biPlanes: 1
biBitCount: 12
biCompression: YV12
biSizeImage: 613440
biXPelsPerMeter: 0
biYPelsPerMeter: 0
biClrUsed: 0
biClrImportant: 0

pbFormat:
0000: 00 00 00 00 00 00 00 00 54 03 00 00 e0 01 00 00 ........T...à...
0010: 00 00 00 00 00 00 00 00 54 03 00 00 e0 01 00 00 ........T...à...
0020: 00 00 00 00 00 00 00 00 63 17 05 00 00 00 00 00 ........c.......
0030: 00 00 00 00 00 00 00 00 54 03 00 00 e0 01 00 00 ........T...à...
0040: 00 00 00 00 00 00 00 00 28 00 00 00 54 03 00 00 ........(...T...
0050: e0 01 00 00 01 00 0c 00 59 56 31 32 40 5c 09 00 à.......YV12@\..
0060: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................



Correct chroma:
- Connection media type:

Video: YV12 864x480 (71:40) 29.97fps

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Video {73646976-0000-0010-8000-00AA00389B71}
subtype: MEDIASUBTYPE_YV12 {32315659-0000-0010-8000-00AA00389B71}
formattype: FORMAT_VideoInfo2 {F72A76A0-EB0A-11D0-ACE4-0000C0CC16BA}
bFixedSizeSamples: 1
bTemporalCompression: 0
lSampleSize: 622080
cbFormat: 112

VIDEOINFOHEADER:
rcSource: (0,0)-(852,480)
rcTarget: (0,0)-(852,480)
dwBitRate: 0
dwBitErrorRate: 0
AvgTimePerFrame: 333667

VIDEOINFOHEADER2:
dwInterlaceFlags: 0x00000000
dwCopyProtectFlags: 0x00000000
dwPictAspectRatioX: 71
dwPictAspectRatioY: 40
dwControlFlags: 0x00000000
dwReserved2: 0x00000000

BITMAPINFOHEADER:
biSize: 40
biWidth: 864
biHeight: 480
biPlanes: 1
biBitCount: 12
biCompression: YV12
biSizeImage: 622080
biXPelsPerMeter: 0
biYPelsPerMeter: 0
biClrUsed: 0
biClrImportant: 0

pbFormat:
0000: 00 00 00 00 00 00 00 00 54 03 00 00 e0 01 00 00 ........T...à...
0010: 00 00 00 00 00 00 00 00 54 03 00 00 e0 01 00 00 ........T...à...
0020: 00 00 00 00 00 00 00 00 63 17 05 00 00 00 00 00 ........c.......
0030: 00 00 00 00 00 00 00 00 47 00 00 00 28 00 00 00 ........G...(...
0040: 00 00 00 00 00 00 00 00 28 00 00 00 60 03 00 00 ........(...`...
0050: e0 01 00 00 01 00 0c 00 59 56 31 32 00 7e 09 00 à.......YV12.~..
0060: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................

[Mark_A_W]

I see the same problem as ryrynz with his sample when using the Windows 7 built-in WMVideo Decoder DMO.

Using FFDshow instead doesn't exhibit the problem.

It appears to be yet another madVR problem with aspect ratio signaling.

Crazy diagonal shifted chroma:
- Connection media type:

Video: YV12 852x480 29.97fps

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Video {73646976-0000-0010-8000-00AA00389B71}
subtype: MEDIASUBTYPE_YV12 {32315659-0000-0010-8000-00AA00389B71}
formattype: FORMAT_VideoInfo2 {F72A76A0-EB0A-11D0-ACE4-0000C0CC16BA}
bFixedSizeSamples: 1
bTemporalCompression: 0
lSampleSize: 613440
cbFormat: 112

VIDEOINFOHEADER:
rcSource: (0,0)-(852,480)
rcTarget: (0,0)-(852,480)
dwBitRate: 0
dwBitErrorRate: 0
AvgTimePerFrame: 333667

VIDEOINFOHEADER2:
dwInterlaceFlags: 0x00000000
dwCopyProtectFlags: 0x00000000
dwPictAspectRatioX: 852
dwPictAspectRatioY: 480
dwControlFlags: 0x00000000
dwReserved2: 0x00000000

BITMAPINFOHEADER:
biSize: 40
biWidth: 852
biHeight: 480
biPlanes: 1
biBitCount: 12
biCompression: YV12
biSizeImage: 613440
biXPelsPerMeter: 0
biYPelsPerMeter: 0
biClrUsed: 0
biClrImportant: 0

pbFormat:
0000: 00 00 00 00 00 00 00 00 54 03 00 00 e0 01 00 00 ........T...à...
0010: 00 00 00 00 00 00 00 00 54 03 00 00 e0 01 00 00 ........T...à...
0020: 00 00 00 00 00 00 00 00 63 17 05 00 00 00 00 00 ........c.......
0030: 00 00 00 00 00 00 00 00 54 03 00 00 e0 01 00 00 ........T...à...
0040: 00 00 00 00 00 00 00 00 28 00 00 00 54 03 00 00 ........(...T...
0050: e0 01 00 00 01 00 0c 00 59 56 31 32 40 5c 09 00 à.......YV12@\..
0060: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................



Correct chroma:
- Connection media type:

Video: YV12 864x480 (71:40) 29.97fps

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Video {73646976-0000-0010-8000-00AA00389B71}
subtype: MEDIASUBTYPE_YV12 {32315659-0000-0010-8000-00AA00389B71}
formattype: FORMAT_VideoInfo2 {F72A76A0-EB0A-11D0-ACE4-0000C0CC16BA}
bFixedSizeSamples: 1
bTemporalCompression: 0
lSampleSize: 622080
cbFormat: 112

VIDEOINFOHEADER:
rcSource: (0,0)-(852,480)
rcTarget: (0,0)-(852,480)
dwBitRate: 0
dwBitErrorRate: 0
AvgTimePerFrame: 333667

VIDEOINFOHEADER2:
dwInterlaceFlags: 0x00000000
dwCopyProtectFlags: 0x00000000
dwPictAspectRatioX: 71
dwPictAspectRatioY: 40
dwControlFlags: 0x00000000
dwReserved2: 0x00000000

BITMAPINFOHEADER:
biSize: 40
biWidth: 864
biHeight: 480
biPlanes: 1
biBitCount: 12
biCompression: YV12
biSizeImage: 622080
biXPelsPerMeter: 0
biYPelsPerMeter: 0
biClrUsed: 0
biClrImportant: 0

pbFormat:
0000: 00 00 00 00 00 00 00 00 54 03 00 00 e0 01 00 00 ........T...à...
0010: 00 00 00 00 00 00 00 00 54 03 00 00 e0 01 00 00 ........T...à...
0020: 00 00 00 00 00 00 00 00 63 17 05 00 00 00 00 00 ........c.......
0030: 00 00 00 00 00 00 00 00 47 00 00 00 28 00 00 00 ........G...(...
0040: 00 00 00 00 00 00 00 00 28 00 00 00 60 03 00 00 ........(...`...
0050: e0 01 00 00 01 00 0c 00 59 56 31 32 00 7e 09 00 à.......YV12.~..
0060: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................


Does it happen when you use the wmv decoder called though the ffdshow codecs section?

[Porsche_fan]

Thanks namaiki for your response and the information regarding madVR...really appreciate it.


Perhaps you could go for:
1GB NVIDIA GeForce GT 320 [DVI, HDMI, VGA]
or
1GB ATI Radeon HD 5570 [DVI,HDMI, DP, VGA adapter]

If you want to be able to drive 1080p with ease.

You will still need a half decent CPU if you want to do post processing, though honestly, postprocessing is something I try to avoid.


I am a little confused about this…why would I need a beefier graphic’s card for 1080p content? If I understand correctly, ffdshow doesn’t use the gpu unless it offloads decoding through DXVA. So if I don’t use the gpu to decode, and I can’t use it to postprocess…then the only function I can see that uses the gpu is madVR to resize.

However, all most all 1080p content I've seen is already at 1920x1080 and therefore wouldn’t the graphics card just pass the image along untouched? The only exception I can think of is if the 1080p content was anamorphic 1440x1080.

It could be that most if not all 1080p Blu-ray content is anamorphic but even if it were, from a purely resizing issue (which is all madVR would be using the gpu for) wouldn’t it be more computationally taxing to upscale 480p dvd or 720p content? I realize that bitrates are higher for 1080p vs. 720/420p but isn’t that more a decoding issue?

Also does nVidia or ATI have superior technology when it comes to just video display and not 3D renderering...for example anti-aliasing? All the reviews I have seen seem to suggest ATI over nVidia but mainly they are talking to a gaming audience and don't really address their relative strengths from a video display prospective.

[Mark_A_W]

Thanks namaiki for your response and the information regarding madVR...really appreciate it.




I am a little confused about this…why would I need a beefier graphic’s card for 1080p content? If I understand correctly, ffdshow doesn’t use the gpu unless it offloads decoding through DXVA. So if I don’t use the gpu to decode, and I can’t use it to postprocess…then the only function I can see that uses the gpu is madVR to resize.

However, all most all 1080p content I've seen is already at 1920x1080 and therefore wouldn’t the graphics card just pass the image along untouched? The only exception I can think of is if the 1080p content was anamorphic 1440x1080.

It could be that most if not all 1080p Blu-ray content is anamorphic but even if it were, from a purely resizing issue (which is all madVR would be using the gpu for) wouldn’t it be more computationally taxing to upscale 480p dvd or 720p content? I realize that bitrates are higher for 1080p vs. 720/420p but isn’t that more a decoding issue?

Also does nVidia or ATI have superior technology when it comes to just video display and not 3D renderering...for example anti-aliasing? All the reviews I have seen seem to suggest ATI over nVidia but mainly they are talking to a gaming audience and don't really address their relative strengths from a video display prospective.


Can you start a new thread on this guys?


(But even with 1920x1080 at 1920x1080 you still need to upscale the chroma.)

[madshi]

I planned on double-checking at the end but it seems that CTRL+R doesn't work if you don't have the OSD enabled...could you please allow it to reset the rejected frames numbers even if the OSD isn't being currently shown?
Can do.

Try this: http://www.multiupload.com/HJ5JISVUJB

Are you saying that the sample displayed incorrect AR on your system, even with EVR/EVR-Custom?
And you do have the "Keep Aspect Ratio" option enabled? Perhaps this is specific to Win7 if you're not seeing the problem on XP?
Thanks for the detailed information. Downloads work now. I can't reproduce the problem on my XPSP3 machine, I think it's probably win7 specific. Will try that with win7 later...

I'm not sure, but I think it's the decoder, not the splitter. What I mean is the filter that can be enabled/disabled in the options->internal filters setting dialog of MPC-HC. In the right pane of that window, there's a option named "VC1 (FFmpeg)", but if I enable it, MPC-HC just doesn't play any VC1 video with madVR, although it do with any other renderer. If I disable it, then it works with madVR too. Why?
Which OS are you on? Which splitter are you using?

I found what's causing the problem. It's an option "seamless playback" which is supposed to work only in VMR9 but obviously is used on madVR too ... but not on EVR and that made me think it's madVR's fault.
Sorry!
No problem, good to know.

Now I found another incompatibility/issue with ZoomPlayer:
As you can see in the attached screenshots there are a few missing values in the Media info window when madVR is used.
Probably a missing interface. That doesn't bother me for now. I plan to implement all VMR/EVR interfaces sooner or later, then this should be fixed, as well. Might take a while, though.

After I posted the problem I remembered that someone here in the forum said that those <1ms times are fake reports from Aero which made me think what if I stop using "aero" timing mode and go back to "normal". I'm happy to report that this fixed it ... or at least I'm not able to reproduce those high present time values. Now I get ~6ms present times which lead to a perfectly smooth 40 minutes of playback.
So the question now is: what does "aero" mode do different from "normal" mode and that's not working every time?
Hmmmmm... "aero" timing mode always presents at scanline 1, while the "normal" timing mode presents at a later scanline (varying, depending on the queue fullness). Maybe scanline 1 is too early? Maybe it should make it screenHeight / 10 or something like that...

Madshi, I have been wanting to use your renderer for a while but my HTPC wouldn’t support it. I was hoping maybe to get yours and possibly other forum member’s advice on how to best allocate my limited budget for a new HTPC.
That depends very much on which filter combination you ultimately want to end up with. If you plan to use madVR, you need a powerful GPU and a reasonably powerful CPU (powerful enough to do software h264 + VC-1 decoding). If you plan to use ffdshow video processing and not madVR, you need a very powerful CPU and almost any GPU will do. If you want to combine ffdshow video processing with madVR, you will need a reasonably powerful GPU and a very powerful CPU.

My general advice is that if you plan to use madVR, you should get the highest power GPU that you can afford and which fits your thermal requirements. Of course if you're on a budget and plan to use ffdshow's post processing (at least for a while), too, then you may need to compromise somewhere, to get a faster CPU.

I tend to use 3 ffdshow filters, resize lanczos2 to naïve screen resolution, unsharp mask on the resized image, and occasionally avisynth filters in realtime like SeeSaw too. I will be using these filters on DVD’s, 720p/1080p website based VC-1/h.264 content and Blu-ray eventually…
You need to decide...

(1) ... what you plan to use right now.
(2) ... what you plan to use in the future.

Currently madVR does not support unsharp masking, SeeSaw etc. So if you want such algorithms, you'll have to use ffdshow for video processing for now. Maybe madVR will some day do sharpening, too, in that case you can switch to using madVR only. That would lessen the requirements on the CPU power and increase the requirements on the GPU.

As far as I can tell ffdshow filters and decoding aren’t done on the graphics card unless through DXVA. If I understand correctly, DXVA allows the video card to decode but wouldn’t allow my other ffdshow to be used. Therefore, I wouldn’t be interested in using DXVA.
There is a way to use DXVA decoding and still use ffdshow's post processing - but it's a rather bad solution IMHO. So if you want to use ffdshow's post processing, I'd forget about DXVA.

One other thing...the latest ATI cards support DirectX 11. How important would this be to renderering in the future and to my HTPC's functions?
For normal HTPC use, DirectX 11 does not seem to have any use, as far as I can tell. Of course things can always change in the future. However, maybe you want a GPU which can bitstream HD audio formats (e.g. TrueHD)? In that case you can rule out quite a number of older NVidia cards.

I am a little confused about this…why would I need a beefier graphic’s card for 1080p content? If I understand correctly, ffdshow doesn’t use the gpu unless it offloads decoding through DXVA. So if I don’t use the gpu to decode, and I can’t use it to postprocess…then the only function I can see that uses the gpu is madVR to resize.
The higher the movie/display resolution, the more GPU resources madVR needs. And if you want to use potential future madVR algorithms (not implemented yet), you will need a high power GPU, anyway.

It could be that most if not all 1080p Blu-ray content is anamorphic but even if it were, from a purely resizing issue (which is all madVR would be using the gpu for) wouldn’t it be more computationally taxing to upscale 480p dvd or 720p content? I realize that bitrates are higher for 1080p vs. 720/420p but isn’t that more a decoding issue?
Bitrates don't matter for the GPU. But the higher the movie resolution is, the higher the scaling cost for the GPU. And the higher the display resolution is, the higher cost all madVR's processing has. Of course, if you don't scale at all, that reduces GPU power a lot. Blu-Ray HD is never anamorphic, IIRC.

Also does nVidia or ATI have superior technology when it comes to just video display and not 3D renderering...for example anti-aliasing? All the reviews I have seen seem to suggest ATI over nVidia but mainly they are talking to a gaming audience and don't really address their relative strengths from a video display prospective.
All ATI 5xxx cards can do HD audio bitstreaming, while most current NVidia cards can't.

Colors are messed up on this vid, link to sample provided.
Colors are fine with EVR.
Thanks, will have a look at that.

[cyberbeing]

Does it happen when you use the wmv decoder called though the ffdshow codecs section?

No, it doesn't. Hence why I suspected it was a problem with how madVR was handling the different aspect ratio signaling in the output pins.

__________

Before I forget, I was having a lot of trouble getting anything to playback even somewhat smooth with the ATI 5750 and NEC 20WMGX2 LCD @59.95Hz on Win7. I'm sure it is partly because of me being way to use to my CRT @96Hz and 120Hz, but it still seems like madVR could do much better in this area. Pretty much everything produces horrible stuttering without madVR showing dropped or delayed frames. EVR Cust, even without D3D Fullscreen is much smoother.

The only settings I found somewhat watchable, yet still with little stutters all the time, are the following:

Timing model: Aero
After render: Don't Flush
After last: Don't Flush
After backbuffer: Don't Flush
After present: Flush & Wait Loop

Core i5-750
ATI 5750
Single Monitor 59.95Hz
Win7 x64
Aero ON

[cyberlolo]

Which OS are you on? Which splitter are you using?

Windows 7 x64, and Haali Media Splitter.

That depends very much on which filter combination you ultimately want to end up with. If you plan to use madVR, you need a powerful GPU and a reasonably powerful CPU (powerful enough to do software h264 + VC-1 decoding). If you plan to use ffdshow video processing and not madVR, you need a very powerful CPU and almost any GPU will do. If you want to combine ffdshow video processing with madVR, you will need a reasonably powerful GPU and a very powerful CPU.

That confuses me a bit. I've got an ATI HD4850 and an i7. I don't use any ffdshow video processing (ffdshow video isn't listed in the filter list of MPC-HC when playing a video), but when watching h264 videos (720p and 1080p with Lanczos4), GPU usage is 0% (says GPU-Z) and CPU is 6-7%.

In the other hand, when I watch VC-1 videos (using Windows decoder, because as I told you, MPC-HC internal VC-1 decoder doesn't work), the GPU usage is about 30%.

Of course, I'm using madVR in both cases (and CoreAVC to decode h264 files). So why my GPU usage is 0% when using madVR without ffdshow video filter in h264 files?

[namaiki]

^On my 9600M GT, 1280x720x24@1280x800x60, GPU load/memory controller load is ~45%/20%. ~12%/5% GPU usage if no resizing. Aero is ~4/8%.

[leeperry]

Can do.
OK great! I always execute the following procedure when watching a movie:
-go FS
-seek back to catch the VSYNC fliptime properly(some old superstition, not sure if it's REALLY needed?)
-CTRL+J / CTRL+R / CTRL+J

and be amazed by the über-smoothness for 90/120 mins straight :eek:

I always check back the number of dropped frames at the end, and these days they're always 0/0. I guess polishing my timings to 96.000Hz and resetting Reclock's timings have really allowed to unleash mVR's true potential :cool:

I'm kinda scared to reboot at this point as this might mess w/ Reclock's timings...but it'll happen soon enough, and nothing a timings reset couldn't fix anyway. I don't have anymore excuse to not try 48Hz on the pj :p

[Mark_A_W]

I guess polishing my timings to 96.000Hz and resetting Reclock's timings have really allowed to unleash mVR's true potential


No, because you should be using 95.904hz. The framerate on the disc, for better or worse, is 23.976.

Or, as James says, it doesn't really matter with Reclock - you only have to be close enough, reclock takes care of the rest.

[leeperry]

James has always been clear that there's only one resampling pass taking place, but 2 adaptations happening in Reclock for 23.976 material: the first speeds up to 24.000fps(as the Reclock panel says "CINEMA adaptation: media speed changed to 24.000 fps", 24fps being the genuine "cinema" frame rate), and the second makes it match a perfect multiple of the refresh rate.

Anyway, maybe the GPU pixel clock frequency I was using was the problem, or maybe I should have reset Reclock's timings after rolling audio interfaces/players/mVR builds...all that matters is that mVR+PotP haven't dropped a single frame in the last half-dozen movies I've watched...and I really would love it to stay like that http://forum-images.hardware.fr/images/perso/ayuluna.gif
That's mVR's OSD after 100 mins: http://thumbnails32.imagebam.com/8742/a6a7c987415008.jpg (http://www.imagebam.com/image/a6a7c987415008) using those setttings: http://thumbnails16.imagebam.com/8742/ae2fa687415010.jpg (http://www.imagebam.com/image/ae2fa687415010)

[buletti]

Currently madVR does not support unsharp masking, SeeSaw etc. So if you want such algorithms, you'll have to use ffdshow for video processing for now. Maybe madVR will some day do sharpening, too, in that case you can switch to using madVR only. That would lessen the requirements on the CPU power and increase the requirements on the GPU.


I'm really looking forward to the day MadVR can do sharpening via Didee's avisynth functions (i.e. limitedsharpenfaster and/or seesaw).
Then after scaling, dithering and sharpening the only postporessing missing would be denoising. Maybe we can make Madshi squish Fizick's MvDegrain into MadVR too... :D

[leeperry]

Maybe we can make Madshi squish Fizick's MvDegrain into MadVR too
I'd rather have GrainFactory3() tbh, especially w/ the recent bugfix from foxyshadis that allows fixed grain sequences..but humm, having a GPU accelerated Avisynth compiler sounds a bit optimistic :D

but yes, LSF is the best sharpener you can find IMHO...having it in mVR would be really neat. Its nicest feature is that it sharpens up motion blur http://forum.slysoft.com/images/smilies/agreed.gif

[Mark_A_W]

""CINEMA adaptation: media speed changed to 24.000 fps""

That is purely cosmetic. Look at the actual audio clock number, mine says something very close to 192,000hz, not 192,192hz, which would be expected if it was forcing it to 24fps.

[DigitalLF]

MadShi for your future use you said that a 5750 might be the one.. is there a reason why you prefer ATI before NVIDIA? because i am in need of a new card and its mostly for future versions of MadVR.. and yes i know you said that buy the one that gives the most performance that my wallet can take..

sorry for my english..

[madshi]

I was having a lot of trouble getting anything to playback even somewhat smooth with the ATI 5750 and NEC 20WMGX2 LCD @59.95Hz on Win7. I'm sure it is partly because of me being way to use to my CRT @96Hz and 120Hz, but it still seems like madVR could do much better in this area. Pretty much everything produces horrible stuttering without madVR showing dropped or delayed frames. EVR Cust, even without D3D Fullscreen is much smoother.

The only settings I found somewhat watchable, yet still with little stutters all the time, are the following:

Timing model: Aero
After render: Don't Flush
After last: Don't Flush
After backbuffer: Don't Flush
After present: Flush & Wait Loop
Hmmmm... I guess we're talking about 24Hz content on 60Hz monitor, right? So you should see the typical 3:2 judder. But if EVR-Custom is smoother then it seems that madVR produces more judder than just that. What does the madVR OSD say about the composition rate and display refresh rate? Are both correct?

A log file may help.

BTW, again you're using funny flush settings. Nobody else is using settings like that. I mean it's possible that these settings really work best for you, but it's weird that nobody else is sharing your impressions in this area. So far I thought it would be due to your special setup with high refresh rates and all. But win7 Aero with 60Hz is pretty normal. Are you sure that the default flush settings don't work better? I've had several cases where people reported stuttering when using Aero, and got rid of it by using flushing (either with or without waiting) after the last render step.

That confuses me a bit. I've got an ATI HD4850 and an i7. I don't use any ffdshow video processing (ffdshow video isn't listed in the filter list of MPC-HC when playing a video), but when watching h264 videos (720p and 1080p with Lanczos4), GPU usage is 0% (says GPU-Z) and CPU is 6-7%.

In the other hand, when I watch VC-1 videos (using Windows decoder, because as I told you, MPC-HC internal VC-1 decoder doesn't work), the GPU usage is about 30%.

Of course, I'm using madVR in both cases (and CoreAVC to decode h264 files). So why my GPU usage is 0% when using madVR without ffdshow video filter in h264 files?
Don't know, sounds weird to me. GPU usage of 0% is not normal when using madVR. Maybe the measurement doesn't work in this situation for whatever weird reason? Or maybe madVR isn't used at all with that filter setup? Sometimes MPC HC changes renderers behind our back, e.g. if the decoder doesn't output YV12.

MadShi for your future use you said that a 5750 might be the one.. is there a reason why you prefer ATI before NVIDIA? because i am in need of a new card and its mostly for future versions of MadVR.. and yes i know you said that buy the one that gives the most performance that my wallet can take..
I prefer ATI right now because:

(1) The "power per watt" ratio is clearly better, which is very important for me (because HTPCs are usually thermally limited).

(2) All 5xxx ATI cards can do HD audio bitstreaming. Only the high-end NVidia cards can do that, which are running very hot.

(3) Unlike most games madVR uses *unfiltered* texture reads. According to Anandtech, ATI cards are twice as fast with unfiltered texture reads compared to filtered texture reads. With older NVidia cards unfiltered texture reads are not any faster than filtered texture reads. Only with the new Fermi cards NVidia has cought up with ATI in this department. That should give ATI cards quite an advantage over NVidia cards, when using madVR.

I like the 5750 because it has a very good performance per watt ratio. Much better than anything NVidia currently has to offer.

[DigitalLF]

I prefer ATI right now because:

(1) The "power per watt" ratio is clearly better, which is very important for me (because HTPCs are usually thermally limited).

(2) All 5xxx ATI cards can do HD audio bitstreaming. Only the high-end NVidia cards can do that, which are running very hot.

(3) Unlike most games madVR uses *unfiltered* texture reads. According to Anandtech, ATI cards are twice as fast with unfiltered texture reads compared to filtered texture reads. With older NVidia cards unfiltered texture reads are not any faster than filtered texture reads. Only with the new Fermi cards NVidia has cought up with ATI in this department. That should give ATI cards quite an advantage over NVidia cards, when using madVR.

I like the 5750 because it has a very good performance per watt ratio. Much better than anything NVidia currently has to offer.

"power per watt" is importent to me to. so il think i will buy a 5750 tomorrow.. but hd audio bitstreaming does not interest me for at least a few years... (got a perfectly good high-end 2.2 (L, R, L_sub, R_sub) amplifyer.


is there any disadvantage with a ATI card?


thank you for your help.

[LloydA]

I like the 5750 because it has a very good performance per watt ratio.

You mentioned before that your GPU is heatpipe-cooled. How exactly did you accomplish that?

[cyberlolo]

Don't know, sounds weird to me. GPU usage of 0% is not normal when using madVR. Maybe the measurement doesn't work in this situation for whatever weird reason? Or maybe madVR isn't used at all with that filter setup? Sometimes MPC HC changes renderers behind our back, e.g. if the decoder doesn't output YV12.

Nope, I think madVR is running. If I press Ctrl+J, it shows madVR's OSD, and if I change TV Levels to PC Levels in madVR settings dialog, I can pefectly see the change in the video black level. This does mean that madVR is running, doesn't it?

Example:

http://www.gandario.com/cal_img/madVR-GPU0.jpg

[LloydA]

Same here, 0% even though madVR is running. But I fiddled with PowerPlay settings (via Radeon BIOS Editor), reverting to stock settings shows proper GPU utilization.

[cyberbeing]

Hmmmm... I guess we're talking about 24Hz content on 60Hz monitor, right? So you should see the typical 3:2 judder. But if EVR-Custom is smoother then it seems that madVR produces more judder than just that. What does the madVR OSD say about the composition rate and display refresh rate? Are both correct?

A log file may help.

BTW, again you're using funny flush settings. Nobody else is using settings like that. I mean it's possible that these settings really work best for you, but it's weird that nobody else is sharing your impressions in this area. So far I thought it would be due to your special setup with high refresh rates and all. But win7 Aero with 60Hz is pretty normal. Are you sure that the default flush settings don't work better? I've had several cases where people reported stuttering when using Aero, and got rid of it by using flushing (either with or without waiting) after the last render step.
Yes, 24000/1001 content on LCD monitor @59.95Hz.

With 'Reset to optimal renderer settings" set in MPC-HC, EVR-Custom (aero disabled) seemed to do better than madVR (aero enabled), while EVR-Sync 'present at nearest' (aero disabled) was about the same as madVR (aero enabled). As I mentioned previously, the ATI 5750 does better with Aero enabled with madVR. When Aero is disabled with madVR, stuttering gets much worse, with no combination of settings helping. It was the exact opposite with NVIDIA cards I tested with Aero disabled always being better.

Composition Rate: 60Hz
Refresh Rate: 59.95Hz

I'll try to get you a log file later.

As for the funny flush settings, I suspect some people aren't really testing smoothness by watching panning scenes very closely (I sit at 6 inches away when examining scenes with my computer monitor) over and over watching for stutters with various settings. Many settings feel smooth, until you get to panning scenes which show obvious stutters. You really have to focus on a single high contrast edge to detect any tiny stutters. Watching the entire frame at once only allows you to easily see large stutters.

The first 60 seconds of Avatar has a vertical (forward moving) panning scene which makes stuttering quite obvious with poor settings. The scene with the moving spaceship and slow pan in space from 00:02:10 to 00:02:40 is also helpful for detecting stuttering. I've also been using the Evangelion 2.22 horizontal pan from 00:05:52 to 00:06:11 and random pans from 11:20 to 12:10 to help detect stuttering.

As for GPU usage reporting, GPU-Z seems broken in detecting GPU usage with Cat 10.6. I fired up Everest Ultimate and it was still able to detect GPU usage just fine.

[cyberlolo]

Same here, 0% even though madVR is running. But I fiddled with PowerPlay settings (via Radeon BIOS Editor), reverting to stock settings shows proper GPU utilization.

I didn't do anything with PowerPlay, so why I get 0% with madVR and h264/AVC files, and about 25-30% GPU usage with VC1 files? Maybe is a decoder issue, and CoreAVC manages evrything on the CPU? :confused:

[dansrfe]

Slight n00b question but even though everything is processed and shown as in the original file, isn't it eventually converted to 8-bit RGB by the Graphics Card driver? Also I don't quite understand the use of 3DLUT. Thanks madshi! :)

[Mark_A_W]

As for the funny flush settings, I suspect some people aren't really testing smoothness by watching panning scenes very closely (I sit at 6 inches away when examining scenes with my computer monitor) over and over watching for stutters with various settings. Many settings feel smooth, until you get to panning scenes which show obvious stutters. You really have to focus on a single high contrast edge to detect any tiny stutters. Watching the entire frame at once only allows you to easily see large stutters.

The first 60 seconds of Avatar has a vertical (forward moving) panning scene which makes stuttering quite obvious with poor settings. The scene with the moving spaceship and slow pan in space from 00:02:10 to 00:02:40 is also helpful for detecting stuttering. I've also been using the Evangelion 2.22 horizontal pan from 00:05:52 to 00:06:11 and random pans from 11:20 to 12:10 to help detect stuttering.




This is basically what I do. Stare endlessly at panning scenes at close range with my 24" CRT monitor.

However I also have the reclock tearing test active, which really shows any little jumps. (Although background motion behind the tearing test can confuse your mind.)


And as an aside, I noticed two distinct glitches in the Avatar opening scene pan, on two separate viewings AT THE CINEMA (on digital 3D projectors). That scene is smoother at home for me.


And you really need to get a multisync display!! :)

[Mark_A_W]

is there any disadvantage with a ATI card?



With a 5xxx series ATi card it is difficult to create a custom resolution.


With a 4xxx ATI card you can use Powerstrip.


Nvidia has custom res's in the drivers.


If you use a standard digital display (like my Plasma) this doesn't matter.

But if you have a non-standard display, like my CRT projector, it is critical (run 1080i at 95.904hz for smooth pans, no flicker, and the full 1080p frame is shown (twice actually)).

[foxyshadis]

Nvidia has custom res's in the drivers.

It can be difficult to keep the custom res from resetting if your monitor/TV claims it doesn't support it, though, especially over DVI. I went to ATI when I upgraded my HTPC after that experience, DVI and HDMI both worked out of the box on my off-brand TV.


Back on topic, madshi, I investigated that message about macrovision a bit. It seems to happen on certain DVD menus (always if you leave the FBI warning in), but if you use another renderer to skip it and get to the movie, it works. Something in MPC's DVD parsing includes code to detect macrovision that only executes for external renderers, like Haali and MadVR. I suspect that it's more of a MPC bug, but it's been around for years and no one is inclined to fix it.

[madshi]

is there any disadvantage with a ATI card?
I remember some things:

(1) Some people say ATI driver quality is lower compared to NVidia.
(2) On some OSs ATI doesn't always offer a 23.976 mode, but only 24.000. I'm not sure if that applies to new GPUs, too, or just to my old HD 3850.
(3) A year ago I found that when using ATI's 24.000 mode, my old Sony projector didn't resolve a line test pattern properly. It looked like ATI would output interlaced, but the projector did say 24p, IIRC. Switching the card to 1080p60 and then using PowerStrip to change that to 1080p24 fixed that problem for me. It's possible that this problem was fixed with a newer driver version, don't know.
(4) Some people (leeperry, IIRC) report that the image is sharper for them when using an NVidia card. Don't know why. Maybe it's the same as (3).

:(

You mentioned before that your GPU is heatpipe-cooled. How exactly did you accomplish that?
I'm using this (rather expensive) case which has a quite clever heatpipe system:

http://www.hfx.at/index.php?option=com_content&view=article&id=135:hfxr-mini-home-server-&catid=51:heatsink-cases&Itemid=1

Look at the monstruous external coolers at the sides. The CPU, northbrige and GPU are all connected to these external coolers via heatpipes. You need to do a lot of manual work to connect this all properly etc.

Nope, I think madVR is running. If I press Ctrl+J, it shows madVR's OSD
Yes, then madVR is definitely running.

With 'Reset to optimal renderer settings" set in MPC-HC, EVR-Custom (aero disabled) seemed to do better than madVR (aero enabled), while EVR-Sync 'present at nearest' (aero disabled) was about the same as madVR (aero enabled). As I mentioned previously, the ATI 5750 does better with Aero enabled with madVR. When Aero is disabled with madVR, stuttering gets much worse, with no combination of settings helping. It was the exact opposite with NVIDIA cards I tested with Aero disabled always being better.
That's really weird. I would love to check out the situation myself on your PC. Unfortunately it's not possible remotely... :( No delayed/dropped frames with Aero disabled, either?

Let's wait until I get fullscreen exclusive mode implemented. If the problem then still exists, I may have to revisit/improve the frame presentation logic. If the problem goes away when using fullscreen exclusive mode, then it's caused by problems with windowed rendering.

Slight n00b question but even though everything is processed and shown as in the original file, isn't it eventually converted to 8-bit RGB by the Graphics Card driver?
madVR itself converts everything to 8bit RGB. This way I can make sure that all processing is done properly. The GPU driver does not know that madVR is a video renderer. The GPU driver thinks a game is running, which renders in 8bit RGB. The GPU driver does not even know that some of the data is actually YCbCr 4:2:0.

Also I don't quite understand the use of 3DLUT.
The 3DLUT feature is mainly for people who have their own meter, so they can do a "perfect" display calibration (gamma and gamut correction).

[madshi]

I investigated that message about macrovision a bit. It seems to happen on certain DVD menus (always if you leave the FBI warning in), but if you use another renderer to skip it and get to the movie, it works. Something in MPC's DVD parsing includes code to detect macrovision that only executes for external renderers, like Haali and MadVR. I suspect that it's more of a MPC bug, but it's been around for years and no one is inclined to fix it.
Thanks.

I'm not sure if it's a bug in MPC HC, though. I rather think that the Microsoft DVD menu navigator filter is responsible. It checks some things. I've been able to make it work in XP (it didn't work there, initially, either). But in Vista and Windows 7 it still doesn't work. Once you started movie playback, the situation is different because then the Microsoft DVD menu navigator is probably not running, anymore.

[droc]

Hi, Madshi maybe stupid question.
Why MadVR not working with DXVA (Ati), like mpc-hc dxva with EVR?
Why Cyberlink decoder also not working with madvr(green screen)

[cyberlolo]

Yes, then madVR is definitely running.

If that's so, how is it possible that my GPU usage is 0%? Maybe some of the madVR's features just aren't working but others do? Is that possible? Maybe this is happening to more people but they don't know it because they aren't measuring GPU usage while using madVR... :confused::confused:

[cyberbeing]

Let's wait until I get fullscreen exclusive mode implemented. If the problem then still exists, I may have to revisit/improve the frame presentation logic. If the problem goes away when using fullscreen exclusive mode, then it's caused by problems with windowed rendering.
Yeah, I hope fullscreen exclusive mode fixes it as well. Though it does seems like you may need to eventually code something more exotic for non-multiple refresh rates. Just please keep in mind that your current timing and present logic works excellent with exact-multiple refresh rates, so I wouldn't want that get worse, just to improve non-multiple refresh rates.

Once you start working on fullscreen exclusive mode, how long do you expect it will take to complete? Do you think we'll have something by December this year or is that too soon?

Also, is all this free time you found to work on madVR recently going to last, or do you expect to once again become really busy in near future? Is your busy time a seasonal thing?

[Neeto]

Originally Posted by dansrfe
Slight n00b question but even though everything is processed and shown as in the original file, isn't it eventually converted to 8-bit RGB by the Graphics Card driver?
madVR itself converts everything to 8bit RGB. This way I can make sure that all processing is done properly. The GPU driver does not know that madVR is a video renderer. The GPU driver thinks a game is running, which renders in 8bit RGB. The GPU driver does not even know that some of the data is actually YCbCr 4:2:0.


The 3DLUT feature is mainly for people who have their own meter, so they can do a "perfect" display calibration (gamma and gamut correction).

The ATI 5000 "evergreen" series GPUs "Display pipeline supports xvYCC gamut and 12-bit per component output via HDMI" according to
http://en.wikipedia.org/wiki/Evergreen_(GPU_family)

Is this useful to us? most likely not as BluRay etc done't support the wider gamut - right?

[fps]

Hi, Madshi maybe stupid question.
Why MadVR not working with DXVA (Ati), like mpc-hc dxva with EVR?
Why Cyberlink decoder also not working with madvr(green screen)
From the first post of the topic:
- hardware accelerated video decoding (DXVA) is currently not supported

IIRC it is also not a priority of madshi to support DXVA, if it is even possible. In addition to that you would need a pretty powerfull GPU, way faster than currently needed for madVR.

[Mark_A_W]

madshi

The "no MMCMMS" version seems pretty good with "v0.18" settings.

The "no present stats" version I'm not so sure about.


I checked again, and v0.18 = good, and v0.21 = not good (no matter what settings I think).

[droc]

From the first post of the topic:
- hardware accelerated video decoding (DXVA) is currently not supported

IIRC it is also not a priority of madshi to support DXVA, if it is even possible. In addition to that you would need a pretty powerfull GPU, way faster than currently needed for madVR.

Thanks :)

[Mark_A_W]

Arrgh, no, just had a crappy patch with "no MMSMMS".

[madshi]

If that's so, how is it possible that my GPU usage is 0%?
cyberbeing said: "As for GPU usage reporting, GPU-Z seems broken in detecting GPU usage with Cat 10.6. I fired up Everest Ultimate and it was still able to detect GPU usage just fine."

Once you start working on fullscreen exclusive mode, how long do you expect it will take to complete? Do you think we'll have something by December this year or is that too soon?
Should be sooner than that.

Also, is all this free time you found to work on madVR recently going to last, or do you expect to once again become really busy in near future? Is your busy time a seasonal thing?
It's not a seasonal thing. Can't say any more than that. I can't look into the future, sadly.

The ATI 5000 "evergreen" series GPUs "Display pipeline supports xvYCC gamut and 12-bit per component output via HDMI" according to
http://en.wikipedia.org/wiki/Evergreen_(GPU_family)

Is this useful to us? most likely not as BluRay etc done't support the wider gamut - right?
xvYCC is not useful, because there's no content available with such a wide gamut. 12-bit per component could be slightly beneficial, but Microsoft has this implement in a bad way. madVR is only ever going to use 8bit (per component) RGB output, or maybe 10bit RGB output, but not more. Unless Microsoft improves DirectX in a future Windows version.

The "no MMCMMS" version seems pretty good with "v0.18" settings.

The "no present stats" version I'm not so sure about.

I checked again, and v0.18 = good, and v0.21 = not good (no matter what settings I think).
Arrgh, no, just had a crappy patch with "no MMSMMS".
Well, you once had a crappy patch with v0.18, too. Could you please double check? This is really important: If the MM... version works as well as v0.18, then I can implement that change for v0.22. So here is your chance to get v0.22 back to v0.18 smoothness levels with your setup!

[Mark_A_W]

Hang on I'll test once more before I go to bed...

[leeperry]

Unlike most games madVR uses *unfiltered* texture reads. According to Anandtech, ATI cards are twice as fast with unfiltered texture reads compared to filtered texture reads. With older NVidia cards unfiltered texture reads are not any faster than filtered texture reads. Only with the new Fermi cards NVidia has caught up with ATI in this department. That should give ATI cards quite an advantage over NVidia cards, when using madVR.
interesting! but we're talking about extremely large figures here...I think I read that in a game like Crysis every single pixel would get through at least 100 PS scripts...a far cry(pun intended) from mVR?

my 8800GS can provide 25.2 GTexel/s, apparently the more TMU the higher the texture fillrate: http://www.google.com/search?hl=en&source=hp&q=GTexel%2Fs
HD4850 has 20.8 GTexel/s [..] the 8800GTS G92 has 41.6 GTexel/s
if you don't plan on adding CSI super-resolution in realtime(as discussed here: http://www.nvidia.com/object/io_1237892916378.html), I think my 8800GS is already way overkill for video playback :o
Some people (leeperry, IIRC) report that the image is sharper for them when using an NVidia card.
It's a complicated matter...when I got my HC3100 pj, the picture was so blurry w/ my 8600GT that I thought the pj was faulty?! I upgraded to an HD2600XT, and the blacks seemed much deeper(levels conversion were fine at all times) and the picture had a much more pronounced "pop effect" on the pj, also the picture on my CRT was much clearer than on the 8600GT :eek:

I then upgraded to an HD3850, and the PQ remained unchanged..but at some point I got so sick and tired of their worthless drivers and buggy pstrip that I decided to run a test on a GF9500 and wow, oh wow! the picture on my CRT was far clearer and the projector was also much sharper and vivid(I'm using an 8ft THX Ultra800 Monster DVI/HDMI cable)...maybe due to their TMDS transmitters? there's been a long review here, all TMDS transmitters are not born equal: http://translate.google.com/translate?u=http%3A%2F%2Fwww.erenumerique.fr%2Fqualite_des_connexions_dvi_nvidia_et_ati_assurent_ils_-art-820-9.html&sl=fr&tl=en&hl=&ie=UTF-8

I then advised friends who were using HD3850/70 cards on HC1100/HC3100 pj's and DLP RPTV to get a Gainward 9600GSO(rebadged 8800GS) that was available locally for cheap, and they all confirmed that the PQ cleared up by quite a bit....as usual, there's no hard rule but nvidia and ati cards don't output the same PQ IMHO, depending on their generations as well apparently.

One thing I can definitely confirm is that the jitter in HR was much tighter on the 9600GSO than it's ever been on the 2600XT/3850...I'd even dare saying twice better, but it might have been due to pstrip not using use the proper PLL coeffs(being a dirty hack at heart).
It can be difficult to keep the custom res from resetting if your monitor/TV claims it doesn't support it, though, especially over DVI.
It's a bug in the nvidia drivers I think, after each new drivers update the custom res will vanish after the first reboot...but afterwards they're always remembered IME.

The problem I see w/ pstrip is that it's a dirty hack that's been reverse engineered and adds more bloat onto your system, nvidia provides native support(w/o having any resident apps running)...and it can be as accurate as pstrip if you spend enough time on it, I currently use 96.000Hz and 89.910Hz.

[Mark_A_W]

madshi

"madVR (no MMCSS).ax" with v0.18 just played for 10 mins with no delayed or dropped frames.

Seems good. I'll do a longer test tomorrow night.


Do you want me to test "no present stats" more thoroughly as well?

[Mark_A_W]

I'm sure the Monster DVI cable is what really makes the difference leeperry.

[madshi]

"madVR (no MMCSS).ax" with v0.18 just played for 10 mins with no delayed or dropped frames.

Seems good. I'll do a longer test tomorrow night.

Do you want me to test "no present stats" more thoroughly as well?
If the MMCSS version works for you as well as v0.18 does, and if the normal v0.21 definitely runs worse, then there's no need to test the "no present stats" build.