LAV Filters - DirectShow Media Splitter and Decoders

[CruNcher]

@andyvt
How did you executed the HQV Color Up-sampling test i guess you compared the Cards on their native EVR results did you used a Player for that ?
Also the Noise test is this based on Luma and Chroma in the Intel Control Panel ?

I wish you would have compared AMDs 2nd Generation Video Research of the 7000 series here too

if i supply you with a test sequence could you test this on the Intel Deinterlacer on Ivy Bridge, though i guess the test system is not in your possession anymore ?

Seeing the old Generation scoring so good just let my jaw drop and i would say ATI is back though without a real competitive next Generation complete compare between Nvidia, Intel and AMD it's hard to say GT 430 is hardly comparable with the 680 and the improvements there on the Video part (though comparing Intels partly fixed function Hardware Deinterlacing with Nvidias and AMDs Shader based is not really fair so the highest Class of Cards needs to be compared to get a General idea of where everyone stands

andyvt i guess you know that but there are too much frame variations in the Deinterlace and Upsampling Screenshots frame accuracy is key here else this is pretty useless to compare (though even better would be full Video Compares no still frames anymore), but anyway overall on the right track

[Midzuki]

Originally Posted by andyvt

......
When I can buy a 75-80" 4K PDP for $5K, then it will be a major, blocking, earth stopping problem

I'd recommend a 120'' device at least

[andyvt]

Originally Posted by Midzuki

I'd recommend a 120'' device at least

That would be nice, but the logistics behind a display/projector that large are unworkable in my current location

[andyvt]

Originally Posted by CruNcher

@andyvt
How did you executed the HQV Color Up-sampling test i guess you compared the Cards on their native EVR results did you used a Player for that (screenshots) ?
Also the Noise test is this based on Luma and Chroma in the Intel Control Panel ?

I use PowerDVD for HQV-HD testing, with the settings at default levels. IMO it's not that useful because GPU vendors design for the test. For e.g. AMD has historically performed well in it, but in practice their default VPP settings are a hot mess of overdone NR, skin tone correction and other tweaks.

Originally Posted by CruNcher

I whish you would have compared AMDs 2nd Generation Video Research of the 7000 series with this

If AMD sends me one I am more than happy to run it through the same set of tests

Originally Posted by CruNcher

Btw this flickering of the Adaptive Contrast is a general issue

"First, it should be made clear that the issue observed in the video above is not unique to Intel HD Graphics"

- pretty sure I said something similar in the video

This issue is a major source of frustration for WMC users, and no one wants to take ownership (IMO it's MS's issue to fix). The idea behind including it in this review (and future ones) is 1) to provide an easy documented solution to end users who encounter it and 2) raise awareness around the problem so hopefully someone does something, even if they don't really own it.

Originally Posted by CruNcher

if i supply you with a test sequence could you test this on the Intel Deinterlacer on Ivy Bridge

Sure.

[CruNcher]

Look here http://forum.doom9.org/showpost.php?...&postcount=619

the source is on top results bellow it but i guess not much will have changed here

[andyvt]

Originally Posted by CruNcher

Look here http://forum.doom9.org/showpost.php?...&postcount=619

the source is on top results bellow it but i guess not much will have changed here

This is MS Decoder + EVR in GraphStudioNext @native size. LMK if you wanted something different.

[nevcairiel]

Originally Posted by andyvt

This issue has been raised and an RCR created. They haven't provided a timeline for changing the behavior (the current implementation is by design), but it is in the system.

If its by design, its a bad design.
Why offer an option that doesn't work? Even more so, one that resets on its own all the time?

Anyway, its sad such a trivial issue will force me to install a GPU in this box. Well, to be honest the iGPU wasn't fast enough for madVR to be 100% reliable anyway, 720p60 content wasn't handled fast enough.
Sadly the AnandTech HTPC review didn't use 720p content, because with madVR 1080p actually has less load then 720p (1080p does not need Luma scaling)

[wanezhiling]

Originally Posted by nevcairiel

Well, to be honest the iGPU wasn't fast enough for madVR to be 100% reliable anyway, 720p60 content wasn't handled fast enough.

sigh..

[jakmal]

Originally Posted by nevcairiel

Anyway, its sad such a trivial issue will force me to install a GPU in this box. Well, to be honest the iGPU wasn't fast enough for madVR to be 100% reliable anyway, 720p60 content wasn't handled fast enough.
Sadly the AnandTech HTPC review didn't use 720p content, because with madVR 1080p actually has less load then 720p (1080p does not need Luma scaling)

Oho :| I thought I had taken care of luma scaling by putting in two SD streams.. So, what is the GPU loading like with 720p60 ? Any difference between software decoding and QS / DXVA fall-back ? I assume you are using the 2133 MHz DRAM...

[nevcairiel]

Originally Posted by jakmal

Oho :| I thought I had taken care of luma scaling by putting in two SD streams.. So, what is the GPU loading like with 720p60 ? Any difference between software decoding and QS / DXVA fall-back ? I assume you are using the 2133 MHz DRAM...

Well, 720p is more pixels to process then SD content, so it'll take more power. madVR reports rendering time of 19-20ms. At 1080p its 13-14ms for comparison.
I didn't check actual GPU loads (yet), but i did see it drop quite some frames.

This is with M-N for Chroma and Lanczos3 for Luma (my preferred settings)

I also only have a rather old driver (2598 or something, came with the board), but i couldn't find a better one on the Intel page, so stuff may improve.

[Qaq]

Funny thing is that with my 5450 (yeah, already ordered 7750) I only got issues with 1080p and only with present queue. I mean with prev. madVR versions and only sometimes. So I wonder what is the difference for present queue between 1080 and 720 scaled to 1080 (which plays fine)? This question is related to comments above and not to madVR itself

[jakmal]

Originally Posted by nevcairiel

Well, 720p is more pixels to process then SD content, so it'll take more power. madVR reports rendering time of 19-20ms. At 1080p its 13-14ms for comparison.
I didn't check actual GPU loads (yet), but i did see it drop quite some frames.

So, no difference between fullscreen windowed and fullscreen exclusive? For Andrew, FSE worked well, but in my system it was FSW..

This is with M-N for Chroma and Lanczos3 for Luma (my preferred settings)

Which of the algorithms / configurations load the GPU the most? In addition to making sure that none of the 'tradeoff quality for performance' options are checked, perhaps we should just benchmark with those algorithms.

I also only have a rather old driver (2598 or something, came with the board), but i couldn't find a better one on the Intel page, so stuff may improve.

I think both Andrew and me reviewed it with v2696, and Asus provided v2712 just after I finished the benchmarking...

[nevcairiel]

Originally Posted by jakmal

Which of the algorithms / configurations load the GPU the most? In addition to making sure that none of the 'tradeoff quality for performance' options are checked, perhaps we should just benchmark with those algorithms.

The "most" is Lanczos 8-tap, but its unrealistic because it has awful image quality. Just go with the 4-tap variants of either Lanczos or Spline.

Originally Posted by jakmal

I think both Andrew and me reviewed it with v2696, and Asus provided v2712 just after I finished the benchmarking...

With provided i assume you mean to you directly? Sadly their download page doesn't have anything above the 2598, and Intels 2712 download is still passworded.

[egur]

Is the RGB-HDMI black level issue only in IvyBridge?

My HTPC has a SandyBridge connected the same way and the black levels are perfect. I have Panasonic plasma (50" V20) with excellent black levels - makes it very easy to spot wrong settings.

I personally don't use MadVR, I use EVR.
EVR results were missing from the Anadtech review. EVR-CP provides worse results at worse performance. I don't see a reason to use it in SandyBridge or newer platforms.

8 tap Lanczos exhibits too much ringing (in upscaling). For MadVR, probably better (quality wise) to use 6 taps or even Bi-cubic.

In order to achieve Lanczos4 (8 taps) sharpness without (or little ) ringing, one needs a context adaptive algorithm like EVR provides. I hope MadVR will add this.

BTW, latest drivers on Intel's site is v2656.

[jakmal]

Originally Posted by nevcairiel

Well, 720p is more pixels to process then SD content, so it'll take more power. madVR reports rendering time of 19-20ms. At 1080p its 13-14ms for comparison.
I didn't check actual GPU loads (yet), but i did see it drop quite some frames.

I tested with a 720p60 camcorder clip, and the GPU utilization is between 80 and 90%, but frames are still getting dropped Running DDR3-1600 at 9-9-9-24 for this experiment. Software config is LAV Filters with QS Decoder, and madVR settings as quoted in the review. Driver version is v2712.

Originally Posted by egur

I personally don't use MadVR, I use EVR.
EVR results were missing from the Anadtech review. EVR-CP provides worse results at worse performance. I don't see a reason to use it in SandyBridge or newer platforms.

Interesting! I used EVR-CP because it is the default out of the box for MPC-HC, and it has always served me in good stead so far By the way, what exactly are the differences between EVR and EVR-CP ?

In order to achieve Lanczos4 (8 taps) sharpness without (or little ) ringing, one needs a context adaptive algorithm like EVR provides. I hope MadVR will add this.

The one provided by EVR is dependent on the drivers, I assume? Do you have any insights on whether context adaptive algorithms are being used by other GPU vendors also?

BTW, latest drivers on Intel's site is v2656.

v2712 is available here

[egur]

Originally Posted by jakmal

I tested with a 720p60 camcorder clip, and the GPU utilization is between 80 and 90%, but frames are still getting dropped Running DDR3-1600 at 9-9-9-24 for this experiment. Software config is LAV Filters with QS Decoder, and madVR settings as quoted in the review. Driver version is v2712.

I'm afraid Lanczos4 is too much for IvyBridge - a mainstream and north dGPU is needed.

Originally Posted by jakmal

Interesting! I used EVR-CP because it is the default out of the box for MPC-HC, and it has always served me in good stead so far By the way, what exactly are the differences between EVR and EVR-CP ?

EVR uses what the driver recommends as the best algorithm. This is HW and driver specific. Different GPU can give completely different results.
EVR-CP is an open source project that implements EVR's interfaces.
Using EVR (in SNB/IVB) you'll get:
* Context adaptive scaling
* Deinterlacing and all video post processing algorithms found in the iGPU control panel.

If the display is connected to a dGPU, EVR will use that GPU's video processing algorithms including a scaler.

Originally Posted by jakmal

The one provided by EVR is dependent on the drivers, I assume? Do you have any insights on whether context adaptive algorithms are being used by other GPU vendors also?

As far as I've seen, other vendors do not have such technology. I believe they use text book algorithms like Lanczos or bi-cubic.
Context adaptive scaling was done by another company (not Nvidia/AMD) but I'm not aware of video related products.

Originally Posted by jakmal

v2712 is available here

This file is password protected. I'll forward the complaints about this issue.
OEMS/ISVs can get the latest drivers and tools from Intel's Validation Internet Portal.

[wanezhiling]

Originally Posted by jakmal

By the way, what exactly are the differences between EVR and EVR-CP ?

http://forum.doom9.org/showpost.php?...ostcount=10488

[jmone]

Great review jakmal...finally one focused on HTPC use with LAV/mad. Anyway after reading it, I'm now doubtful if a CPU upgrade on my Shuttle HTPC from a i7-2600K's 3000 iGPU (needs a BIOS upgrade) for a i7-3770 4000 iGPU is going to give that much of a bump. At present I have to use non-tap scalers (eg Mitchell-Netravali) to prevent dropping frames (render queue issues) on 1080/50i/p or 60i material, and that works perfectly. While I'm not interested in OC the mem on a HTPC to get a "just works" soln (esp a Shuttle), I did read you were using the 4 Tap Lanczos scaler, how was it with 3 Taps on 1080/60i marterial - does it still drop frames?

[jakmal]

Originally Posted by jmone

At present I have to use non-tap scalers (eg Mitchell-Netravali) to prevent dropping frames (render queue issues) on 1080/50i/p or 60i material, and that works perfectly. While I'm not interested in OC the mem on a HTPC to get a "just works" soln (esp a Shuttle), I did read you were using the 4 Tap Lanczos scaler, how was it with 3 Taps on 1080/60i marterial - does it still drop frames?

Default for chroma upsampling is SoftCubic. Luma algorithms are Lanczos (4-tap).

I went back to DDR3-1333 and the performance is quite bad even for the M-N algo. If you are going to use madVR, I don't suggest using DDR3-1333.

Also, 720p60 drops frames even with DDR3-1600, and Hendrik reports that it is the case even with DDR3-2133.

So, yes, lots of caveats apply to the LAV / madVR section of the review I posted.

[nevcairiel]

Originally Posted by egur

Is the RGB-HDMI black level issue only in IvyBridge?

No, it affects SNB as well, and i assume all others as well.

The config screen is just lying to you. It offers an option to switch between limited and full range, but when i do, it doesn't change the video at all, and next time i change resolution or reboot its reset to the old value. My TV is calibrated for full-range RGB, and i would like to keep it that way.
It doesn't crush blacks or anything, that part is under control, but it just doesn't send full range RGB like i want it to.

Its a simple driver problem.

Originally Posted by egur

I personally don't use MadVR, I use EVR.
EVR results were missing from the Anadtech review. EVR-CP provides worse results at worse performance. I don't see a reason to use it in SandyBridge or newer platforms.

EVR doesn't offer proper subtitle support, and its vsync accuracy is a mess.
This is why EVR-CP or madVR are much better, even if you lose the (limited) hardware scaler. I'm happy with my madVR configuration, imho it looks better then EVR. Not to mention that i can calibrate to fit my display 100% with madVR.

PS:
EVR-CP (and to some degree madVR) offer the same deinterlacing and other processing. Not that i would want any post-processing except deinterlacing.

Originally Posted by jakmal

Also, 720p60 drops frames even with DDR3-1600, and Hendrik reports that it is the case even with DDR3-2133.

Memory can only do so much. The sweet spot seems to be around 1600 or 1833, my 2133 is probably overkill, but its not like it was extra expensive.
Considering IVB is even native 1600 now, getting 1333 seems just dumb.