Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[raul]

Hi guys, i am a complete newbie & would appreciate all help from all u videophiles.
here goes:

when using red october HQ, should the "hardware accelerate video decoding when possible" be checked or unchecked?
what are the benefits?

in J River, when clicking "control+O", there are many options.
in the "general" section of the many options, there is a tab for 'video', listing hardware accelerated graphics.
there are four options ranging from low quality to highest quality. there is also a custom option where u can choose specific anti-aliasing & anisotropic filtering options.
when using madVR, which option is best? or should this be turned off as madVR already uses it's own anti-aliasing & anisotropic filtering settings?
also, does this come into play ONLY when the "hardware accelerate video decoding when possible" box is checked?

thanks for ur help!

[mykillk]

I believe the hardware accelerated section under "General" has to do with the UI elements of the program itself and has nothing to do with video playback image quality.

As far as I can tell, at this point in time the "hardware accelerate video decoding when possible" in the "Video" options section has absolutely no effect when using Red October HQ. Even if it did, there would be no difference in image quality between using software video decoding and hardware decoding. So, if your computer is fast enough to use Red October HQ I wouldn't worry about it. Hardware accelerated video decoding only really helps out people with very slow CPUs. The MadVR settings themselves are the ones that impact video image quality. What kind of CPU and graphics card do you have and I can give you some suggestions on which scaling options to use to get the most out of it.

[struct]

I found this thread helpful for getting the gpu to decode with AMD/ATI...

http://yabb.jriver.com/interact/index.php?topic=76449.0

It requires you to get your own LAV files but is easy enough.  I am very much a novice in these things and would love to hear why the MC button doesn't engage the dxva2 hardware decoding available in ROHQ?  Just a temporary oversight or is there a reason?

Thanks
Craig

[thezone]

I came across these settings on the slysoft forum (where I found out about MC18!). I have these exact settings running on my i5-3570 not OC, Gigabyte z77x-ud5h, 16g 1600, msi 660ti, intel 120g ssd. My set up scores a 5354 in JRiver's Benchmark Tool. PQ is amazing however I do get occasional and very subtle skips but that could be a bunch of other things that I haven't had time to diagnose yet. Maybe the Nvida card or my panasonic plasma don't properly sync with 23.9etc fps? Who knows.

Any advice regarding these settings would be greatly appreciated.
 
"Tools, options, Video

Video mode Advanced - Custom
Select Custom video mode settings...

In the drop down select Red October HQ

Click add...

For Type, select Video renderer
For Filter, select madVR

Click add...

For Type, select Video decoder
For Filter, select LAV Video Decoder

Click properties

Under Hardware Acceleration, select the option that makes sense for your hardware. For AMD you may want to try DXVA2 (native). For nVidia I recommend DXVA2 (copy-back). For Intel, QuickSync is a good choice. Click ok.

Now the tricky part. MadVR settings aren't available until a video is playing. What I do is start a video, immediately pause it, and then right click, select DirectShow Filters, then madVR.

Under processing, decoding, uncheck all decoders
Under scaling algorithms, this is where it gets very tricky...You have to find the balance of madVR options that give you the best PQ but don't overtax your hardware. For my hardware I choose:

chroma: jinc 3 with anti-ringing
image upscaling: jinc 3 with anti-ringing
image downscaling: lanczos 4 with anti-ringing and scaled in linear light

That could be a bit taxing for lower end systems. You could try chroma: Bicubic 75 and see if that helps. Basically spend some time adjusting, testing, readjusting, testing again until you find the combination that works best for your system."

[raul]

I believe the hardware accelerated section under "General" has to do with the UI elements of the program itself and has nothing to do with video playback image quality.

As far as I can tell, at this point in time the "hardware accelerate video decoding when possible" in the "Video" options section has absolutely no effect when using Red October HQ. Even if it did, there would be no difference in image quality between using software video decoding and hardware decoding. So, if your computer is fast enough to use Red October HQ I wouldn't worry about it. Hardware accelerated video decoding only really helps out people with very slow CPUs. The MadVR settings themselves are the ones that impact video image quality. What kind of CPU and graphics card do you have and I can give you some suggestions on which scaling options to use to get the most out of it.

hhmmmm, r u 100% sure on this 'mykillk?'
anywayz, here are my specs:
Dell XPS L502x:
2nd generation Intel® Core™ i7-2670QM processor (2.20 GHz with Turbo Boost 2.0 up to 3.10 GHz)
Intel HD Graphics 3000 (with latest drivers)
1GB NVIDIA® GeForce® GT 525M graphics with Optimus (with latest drivers)
15.6" (39.6cm) HD (1366x768) WLED display with TrueLife™

(i honestly wish i had went with 2GB NVIDIA® GeForce® GT 540M graphics with Optimus. it may not have made too much of a difference for a gamer, but it whens comes to video playback with madVR, i think better scaling options could be used than with my current nvidia card)

btw, i have noticed in the madVR options under general settings that one can choose the CPU & GPU queue settings. the CPU ranges from 4 to 32. the GPU ranges from 4 to 24. what combinations are best?
does this have any effect on image quality or the smoothness/flow of a video?
could one even use the bare minimum of 4 on both the CPU & GPU?

thanks!

[mykillk]

chroma: jinc 3 with anti-ringing
image upscaling: jinc 3 with anti-ringing
image downscaling: lanczos 4 with anti-ringing and scaled in linear light

That could be a bit taxing for lower end systems.
[/i]

A "bit" taxing??

Jinc and anti-ringing will slaughter all but fairly high end GPUs. Just as a test I tried going from softcubic image upscaling to jinc 3 with anti-ringing and it increased my rendering time by a factor of 60!!

[mykillk]

hhmmmm, r u 100% sure on this 'mykillk?'
anywayz, here are my specs:
Dell XPS L502x:
2nd generation Intel® Core™ i7-2670QM processor (2.20 GHz with Turbo Boost 2.0 up to 3.10 GHz)
Intel HD Graphics 3000 (with latest drivers)
1GB NVIDIA® GeForce® GT 525M graphics with Optimus (with latest drivers)
15.6" (39.6cm) HD (1366x768) WLED display with TrueLife™

(i honestly wish i had went with 2GB NVIDIA® GeForce® GT 540M graphics with Optimus. it may not have made too much of a difference for a gamer, but it whens comes to video playback with madVR, i think better scaling options could be used than with my current nvidia card)

btw, i have noticed in the madVR options under general settings that one can choose the CPU & GPU queue settings. the CPU ranges from 4 to 32. the GPU ranges from 4 to 24. what combinations are best?
does this have any effect on image quality or the smoothness/flow of a video?
could one even use the bare minimum of 4 on both the CPU & GPU?

thanks!

Yea I'm fairly sure about the advice I gave you before.

From your specs, I'd say you'd probably get the best performance continuing to allow the CPU to do the video decoding, which is more than fast enough. This will leave more of your GPU power available for the MadVR scaling settings which, by far, give you the greatest control over performance and video quality. From your specs I'd recommended:

Chroma upscaling: Bicubic 75 with anti-ringing filter
Image upscaling: Lanczos 4-tap with anti-ringing filter enabled
image downscaling: Catmull-Rom

Now go to "Trade quality for performance" and ensure nothing is checked.

As for the queues, I would leave them at default 12 for CPU, 8 for GPU. They have no impact on video quality, they are just there to prevent dropped frames if you are right on the border of pushing your hardware too far to render the frames fast enough. If you press Ctrl+J while a video is playing it will bring up the MadVR On-Screen Display and you can see the status of the buffers. Essentially, they should stay mostly full. If they aren't, and you're getting dropped frames, increasing the CPU/GPU queues will only slightly delay the onset, not fix the problem. In that situation you'd be better served using less intensive scaling settings to reduce load on your GPU. Such as using 3-tap lanczos instead of 4-tap. But given your setup, I think you should have no problem handling my suggested values. Just pay close attention to the information in the MadVR OSD and you should be able to nail down some quality settings that will run smooth as butter on your system. Just make sure to do your testing on non-1080p files because a 1080p video won't need any scaling.

Enjoy!

[mykillk]

Doh did it again lol

[dean70]

DXVA2 (native) does not work too well on AMD APU based setups -DXVA2 (copy back) works correctly. Is Lanczos 3 tap better for Chroma upscaling than Bicubic 75?

[mykillk]

DXVA2 (native) does not work too well on AMD APU based setups -DXVA2 (copy back) works correctly. Is Lanczos 3 tap better for Chroma upscaling than Bicubic 75?

How so? I've been using it on my E-450 APU for weeks now and it seems to work flawlessly. The only issue was that I couldn't get DXVA2 Native enabled unless I added LAV Video to the HQ Red October filter chain manually and enabled it there.

And I just did some checking on doom9 forums about chroma upscaling, and I think you might be onto something. Lanczos doesn't seem to be recommended for the chroma. Either bicubic 75 or Mitchell-Net. is best for that.

[gvanbrunt]

I didn't really have time to read through all this, so if someone already mentioned it, my appologys.

The "hardware accelerate video decoding when possible" refers to decoding of video which in Red October HD (might be SD too now I think) is done by LAV. LAV will offload CPU intensive things to the Video card which can handle it much more efficiently. At one point LAV only supported Nvidia CUDA, but I believe that has been expanded to others now as well. I think it now also has some effect on MAD VR if you are using Red October HD. And yes in that case there are (or were?) some issues with AMD DXVA2. I think you can just shut that off in MAD VR though and still accelerate LAV.

The short answer is turn it on unless you have problems. There is no down side.

[dean70]

How so? I've been using it on my E-450 APU for weeks now and it seems to work flawlessly. The only issue was that I couldn't get DXVA2 Native enabled unless I added LAV Video to the HQ Red October filter chain manually and enabled it there.

And I just did some checking on doom9 forums about chroma upscaling, and I think you might be onto something. Lanczos doesn't seem to be recommended for the chroma. Either bicubic 75 or Mitchell-Net. is best for that.

I have tried DXVA (Native) and get stutter and lots of dropped frames, switch back to DXVA (copy-back) and it is smooth with no dropped frames (this is without changing anything else). Running on a A8-3850.

Lanczos shows higher sharpness with a fraction more ringing on the graph than Bicubic 75 (with AR enabled) - seems to run without dropped frames, so I will leave it as Lanczos .

[raul]

@mykillk: my sincere apology for this very late reply, been busy but procrastinating as well unfortunately.

hhhhmmm, my monitor is 15.6" (39.6cm) HD (1366x768) WLED display with TrueLife™, so playing a 1080p file on my XPS L502x would have to be downscaled.
now let me explain a bit of how i watch videos on my monitor. the only type of video that utilizes the whole screen is 1080p files. (downscaled obviously)
the rest of the videos (1280x720/1280x544/720x576[DVD PAL]/720x480[DVD NTSC]) are played in their NATIVE resolution. IMO, i don't see the point in adding extra pixels to those videos (ie. upscaling them to 1366x768) as i'm looking for the highest quality possible. so really, with these type of videos, the only thing i'm worried about is the chroma upscaling.

hhhmmm, u seem to undermine the full potential of the 1GB NVIDIA® GeForce® GT 525M graphics with Optimus (with latest drivers).
with the videos that i play in their NATIVE resolution, i am able to use Jinc 3 with AR for the chroma upscaling, with NO frames being dropped or delayed.
playback seems to be smooth. btw, this is with the CPU queue level at 32 and 4 for GPU.
u stated: "Essentially, they should stay mostly full. If they aren't, and you're getting dropped frames, increasing the CPU/GPU queues will only slightly delay the onset"
 - i think u made a mistake there, when one is getting DROPPED/DELAYED frames, it is best to DECREASE the GPU queue. this is one of the reasons i am only using 4 for the GPU queue. any higher & Jinc 3 for chroma upscaling would start dropping frames.
looking at the stats after pressing "ctr-J", what rendering times indicate the GPU is handling a video fine? i read somewhere that one of the figures should b under 5ms, and that's how u would know everything is smooth.

btw, my display rate is 59.974Hz & i notice that the composition rate of most of my video files is 60Hz. is it normal for the composition rate to be higher than the display rate?

with 1080p files, however, image downscaling comes into play, and i use lanczos 3 with AR for chroma upscaling - & lanczos 3 with AR for image downscaling. i haven't tried "scale in linear light" as yet... btw, what is this? what purpose does it serve?

u mentioned: "get the best performance by continuing to allow the CPU to do the video decoding, which is more than fast enough."
please tell me, what process do i do to make sure that ALL of the decoding is done by the CPU & that none of it is being done by the GPU?
what settings do i need to change/alter in JRiver or MadVR?


thanks a lot, mykillk!