CPU usage for those with working HDTV tuners

[flashbacck]

jkoutrouba mentioned in the beta 3.0.11 thread that he was seeing between 7-15% CPU usage with his A180, p4 2.8 , using the nvidia decoders. I'm currently seeing fairly high usage, but I remember I've had it at 20-30% before and I'm trying to get it back to those levels. I'm interested in knowing what CPU usage other people are getting when watching live HDTV.

Please be sure to post your CPU, graphics card, HDTV tuner, what decoder you are using and what resolution you're outputing to your display.

Me:
CPU usage: 40-45%
Athlon 2400+
Radeon 9550 overclocked
Fusion Gold 3
Cyberlink PDVD6 video, overlay/Nvidia, Directsound
resolution: 1776x1000i

[Mike Young]

I'm between 5% and 8% with a Pentium 2.4Ghz, Radeon 9800SE.
Using A180, Sonic Decoder for video/audio and DirectX sound Rendering.
Also outputing directly to a VGA input on a HD display. Had higher usage when trying to use DVI output for some reason.

mike/

[gtd885a]

Originally Posted by Mike Young

I'm between 5% and 8% with a Pentium 2.4Ghz, Radeon 9800SE.
Using A180, Sonic Decoder for video/audio and DirectX sound Rendering.
Also outputing directly to a VGA input on a HD display. Had higher usage when trying to use DVI output for some reason.

mike/

Can you tell a big difference in video quality between the DVI and VGA for HD content? I haven't tried the VGA, just assummed the DVI would give better results.

Edit: sorry, didn't mean to hijack the discussion. I use firewire capture, so I'm not sure how it compares to HD capture cards. I have an A64 3400, 1GB pc3200, Radeon 9800 NP (clocked as a 9800 Pro). I'll update this when I've confirmed, but I'm pretty sure I'm around 20% during HD capture/playback.

[cornrow]

Wow... mine is at 80%

I have an Athlon 64 3200+ and a Radeon 9250. I'm using a Fusion 5 HDTV card. Any ideas why mine is so high? With VMR it's too choppy to watch. Overlay looks good, though. Interestingly, both bring the CPU to 80%.

Ty.

[Mike Young]

I just responed to another post about the DVI, I never got it to work properly, It seemed to be 'locked' at a very high resolution (1920 x 1080) which I think is what accounts for the higher CPU, there may also be some DirectX Accel. issues - I got some freezing and lockups. I want to experiment some more with it, because I DID notice a better picture with the DVI, it was too unstable. That said, the straight VGA output is pretty good as well.
I have never liked how VMR looks, it seems washed out to me no matter what, I thought that might change when I got my HD display, but it was the same.
The only reason I can guess why CPU is so high is due to the video card, maybe it is not utilizing DX Hdwr Acceleration. I am not an expert on HD, but from what I understand the capture card should take almost no cpu, since the stream is mpeg anyhow, it doesnt really have to encode (someone correct me if I have misunderstood)
What codec are you using and are you using Hardware Accel for it ?
What resolution are you using ?

mike/



mike/

[gtd885a]

I am at 15-20% CPU when capturing HD using Cyberlink video in Sage, but it jumps up to 60% when I use Intervideo. Unfortunately, Cyberlink usually picks the wrong aspect ratio so I have used Intervideo most often (although it gives me problems too).

Mike, it's interesting that you mention the VMR looking washed out, because I always thought it was a limitation of my graphics card. Playback in Sage has never seemed to have quite the depth of straight cable for me. However, yesterday I was playing around in graphedit and rendered an mpg file and it produced a graph I was not expecting - and when I played it, the picture was much better than it normally is in Sage. I switched back and forth between Sage and regular (live) cable on my TV and couldn't tell much difference. This tells me that my graphics card is not the issue with the picture depth / washed-out look. This is both encouraging and discouraging - because I can't reproduce the picture in Sage.

I don't use VMR9, though. I tried it, and got choppy video. So I wonder why the video is still washed-out looking (not a bad picture, just not as good as the original).

[stanger89]

Originally Posted by Mike Young

I have never liked how VMR looks, it seems washed out to me no matter what, I thought that might change when I got my HD display, but it was the same.

You realize you have to recalibrate for VMR9 since it (usually) maintains correct video levels, while Overlay tends to be "optomized" for PC levels. The result is if you're display is setup for Overlay and you switch, VMR9 will appear washed out due to it's different levels.

[mikejaner]

From what I understand, VMR9 also goes through a color conversion process to get it into the 3d engine of DirectX, or something to that nature, and this process, passes it through a filter which is 8 bit or so resulting in not as dynamic range in the colors. Is this correct?
Either way, I also notice that Overlay is sharper in the details on HD video. It's like the comparison between a PVR350 output VS a Standard video card at SD resolutions.

[stanger89]

Appologies for the long OT

Originally Posted by mikejaner

From what I understand, VMR9 also goes through a color conversion process to get it into the 3d engine of DirectX, or something to that nature, and this process, passes it through a filter which is 8 bit or so resulting in not as dynamic range in the colors. Is this correct?

Nope, nothing of the sort, well sort of, but that implies it's unique in that behavior. In fact, AFAIK, Overlay is 8-bit throughout the pipeline (video is only 8-bit to start with, and our video cards are only 8-bit output). It's been theorized that (at least on Geforce 6 series hardware) that maybe video is handled at 64 or 128-bit floating point precission in the pipeline, but I think that's mostly hope.

To your question though, VMR9 has no less dynamic range than overlay, as it perfectly preserves the ranges and levels found in the source material. It does the YUV -> RGB conversion correctly per BT.601. You can compare the output of VMR9 to that of a reference MPEG decoder (like DGIndex) and the outputs will be identical, at least in level. The same can not be said of overlay.

What overlay does (often) is one of two things, it either does the YUV -> RGB conversion incorrectly, or it "expands" the range (before or after). "Expand" is a misnomer though as what really happens is it clips the edges and then stretches the range out. It's been said (and can be justified to a degree) that by not maintaining correct video levels you're only getting 7.77 bits of the 8 bits of data encoded in the source (219 of 256 levels). This can result in banding and (at the edges of the range) color shifting. But it is incredibly subtle.

VMR9 is far closer to the output of a STB than is Overlay.

Either way, I also notice that Overlay is sharper in the details on HD video. It's like the comparison between a PVR350 output VS a Standard video card at SD resolutions.

ATI's (especially) overlay is known to run the video through a sharpenning filter, artificially "enhancing" the picture. So naturally when you switch to VMR9, which doesn't do the same, the picture will appear softer.

I suggest that before you give up on VMR9, you make a point of running through your calibration procedure with it first, as it will require different settings for most of your picture controls.

[mikejaner]

I suggest that before you give up on VMR9, you make a point of running through your calibration procedure with it first, as it will require different settings for most of your picture controls.
Reply With Quote

Will do, thanks for all the info, you have me convinced to check it out again.

[Mike Young]

I will work with it some more as well, I had tried some of the color calibration back when I was using Svideo out to a Standard TV but didnt really help much so I didnt spend a lot of time with it on my HD.

mike/

[flashbacck]

Has anyone else tried the Sonic decoders besides Mike Young?. 5-8% CPU usage is pretty damn low. I have a roughly similar system and I'm no where near that.

Mike, do you know what version of the sonic decoder's you're using?

[Mike Young]

Sorrry...you know, it just occured to me...I should have mentioned that my tuners are in the server that is never used to view TV. The CPU usage I stated was for my client while watching a recording or Live TV, that probably accounts for half the difference. Im not at home now but I will check the version tonight.
Interestingly, I played around with VMR, and various codecs last night, tried 4 or 5 decoders, VMR significantly increased my cpu use the sonics were the lowest CPU, about 35-40% with some of the others..up to 60%. I could never get stable VMR playback with any codec, there was alot of stuttering, I only tested HD.
putting it into full screen exclusive mode resulted in blank screen with audio and crashing the server ? ....maybe it was a coincidence.

[Mike Young]

sonic decoder ver. 2.5.03

[flashbacck]

Hmm... even if we tack on 20% usage for the recording part, you'd still only be seeing 25-28% CPU usage, which is great for your system.

I don't know if this matters, but what version of Java are you using?
edit: actually that probably doesn't matter.

A better question is, what are you using as a sound card and audio decoder? onboard audio or soundblaster audigy, etc.

[Mike Young]

Its just using the Shuttle onboard sound - I think RealTek and I am using directsound and the Sonic Audio. What kind of resolution are you guys running ?

mike/

[flashbacck]

I'm running 1776x1000i

[jkoutrouba]

Originally Posted by Mike Young

What kind of resolution are you guys running ?

848X480 via DVI

[Fuzzy]

In the wee hours of the night last night, i decided to to a complete reformat and reinstall on my HTPC. So after a clean install, with nothing but drivers sage, and the following decoders, these are the CPU usages I got. I may try some of the other decoders, but nvidia seems to work so well, I may just stick with it.
Overlay VMR9
Sage Decoder: 100%+ 100%+
FFDshow: 75%(bad combing) 75% (bad combing)
Nv PureVideo: 40% 65%
This is on an Athlon XP 3000+, with 1GB of RAM, and an integrated GeForce4 MX GPU. This was while playing live HD in sage from a DVico Fusion 5 Plus with the original drivers from the CD-ROM. 40% seems great and offers plenty of overhead for adding a second tuner, and some comskip processing. I'm pretty happy with the quality, and didn't see any advantage offered my VMR9, except that I can see the video in VNC with VMR9, while only a black screen goes over VNC from Overlay.

[flashbacck]

Thanks for the info Fuzzy. One question though,
what resolution are you outputing to your display?