Stress testing Archos 5 leads to questions on bitrate handling

[Chengbin]

I was doing some stress testing on my Archos 5 to find the maximum bitrate supported.

I used CRF 1 (yes, CRF 1), for a 720x576 video, which comes to about 10Mbps to 25Mbps according to bitrate viewer.

First question, how come some videos will lag at over 15Mbps, but some videos (like the end credits for BBB, for some reason uses) 23Mbps is playable.

Then I went on to test 720p playback. The resolution is 1280x720, the maximum allowed. How come the maximum bitrate is only around 7-8Mbps?

Second question, I thought bitrate is the amount of data a video uses to store the image. Why is a bigger resolution, but processing the same amount of data, takes far more CPU power?

Also, is XviD not demanding to decode? I'm asking because I want to know if it is a coding problem and/or a lack of CPU power to decode high bitrate and high resolution H.264. H.264 will start to lag when the bitrate goes above 5Mbps at 704x400.

[Guest]

I thought bitrate is the amount of data How can a *rate* of data be an *amount* of data? That's like saying "I thought 10 miles per hour was equal to 10 miles."

[Chengbin]

If I encode both videos (different resolution) with say 10Mbps, they'll all end up having the same file size. So isn't that the same amount of data?

[Guest]

If they have the same length, yes. But I only wanted to point out that you are confusing basic concepts. Precision is important, especially when you seek understanding.

It's not just bitrate that is a factor. A larger frame size will have more macroblocks to process, etc. And I don't know what an Archos 5 is, but maybe the difference is related to the transfer/blitting/display/etc. of the uncompressed video frames, and so has nothing to do with decoding per se.

[Chengbin]

Same video source, encoded at the same bitrate, but using 2 different resolutions. The SD video is playable with a much higher bitrate (15Mbps). The HD encode is only playable to about 7-8Mbps. 15Mbps 1280x720 will lag terribly.

The 2 videos (at 15Mbps) have the same bitrate and same file size, just different resolutions. And that's my question, same amount of data to decode, why is the SD encode more playable?

[Guest]

Did you read my second paragraph above?

[Chengbin]

The Archos 5 is a portable media player.

I just read your second sentence. I'm not totally sure what you mean by it might not be a decoding problem.

[Guest]

The uncompressed frames have to be transferred to the display. That requires bandwidth on the memory bus.

Also, I don't know the native resolution of the display, but maybe downsampling is needed for showing HD.

[Chengbin]

The uncompressed frames have to be transferred to the display. That requires bandwidth on the memory bus.

Also, I don't know the native resolution of the display, but maybe downsampling is needed for showing HD.

Well I'm not doing uncompressed encoding (I'm using CRF 1). If you mean the decoded image has to be converted to uncompressed, and I don't have the memory bus to do it, then how come lower bitrate 720p works? They all have to be converted to uncompressed at the end of decoding.

The native resolution is 800x480, so it needs to be scaled to fit the screen.

[Guest]

You need to consider the net load on the memory bus and CPU.

[Chengbin]

OK, let me put it this way.

It seems like resolution affects the amount of CPU needed to decode even if the bitrate is the same. But why?

[Guest]

I already told you.

[Chengbin]

Let's leave it at that.

Can you answer the second question regarding the CPU power needed to decode XviD? Thanks.

[Guest]

AVC decoding is more complex than ASP. Is that what you are asking?

[Chengbin]

Well I wouldn't be surprised if AVC decoding is more complex, but I'm asking because if we assume that the decoder has no problem (which isn't true), H.264 main profile with 4 b-frames and 6 ref-frames and b-pyramid is 5 times more demanding than XviD with GMC and QPel to decode. Is XviD that much more efficient to decode?

[Adub]

It's not that it's more efficient, its that it's less complex. There is a difference. The features of H.264 that enable it to achieve better compression in the first place are also what make it harder to decode.

[Chengbin]

I used a bad choice of word. Complexity is what I meant.

XviD is 5x less complex to encode? I find that kind of hard to believe because H.264's compression efficiency isn't dramatically higher than XviD.

[Guest]

Maybe your Archos just has a crappy AVC decoder. There are so many possibilities to explain your results. You can't just focus on one idea and assume it must be true.

[Chengbin]

Maybe your Archos just has a crappy AVC decoder. There are so many possibilities to explain your results. You can't just focus on one idea and assume it must be true.

You're probably right. It does have a crappy AVC decoder.

[CruNcher]

And it's known :) the Davinci DSP working inside the Archos 5 does 480p fast 720p works but you shouldn't expect major power there for complex AVC, might be that TI can improve the performance but you have to wait :)
ASP i guess runs on the Cortex itself and if the Archos 5 uses libavcodec for playback there it's questionable how optimized it is yet for the Cortex the best is you look @ the http://beagleboard.org/ development to get a inside view of the current possibilities :) last time i asked a dev to test a sequence 1080p 3-8 mbit 25 fps High Profile it could play it at 7 fps that was end of last year.

[Chengbin]

I'm just frustrated that high profile isn't even supported for SD! Even though Archos dominates in video decoding on PMPs, they takes forever to update their decoder, it took 6 generations for them to get GMC and QPel working. They say they're working on it, and claim it is very complex to properly decode high profile, but isn't high profile just 8x8dct?

What DSP does the PSP use? There was a thread in the H.264 forum that said the PSP was capable of playing 20Mbps H.264 @ main profile with 16 b frames (but only 3 ref-frames). That's DAMN powerful. I also did some stress testing for H.264. The maximum bitrate for 704x400 is around 5000Kbps. The maximum b-frame supported is 5 (with b-pyramid). Over 5 will cause the frame to not decode correctly. The maximum ref-frame allowed is 10, but strangely it doesn't even allow me to play a video will 11 or more ref-frames.

On a computer, how much more CPU does H.264 need to decode compared to XviD at equivalent bitrates?

[CruNcher]

Sony uses their so called Media Engine DSP for the PSP not much known about it but looking @ the time when the PSP was released it was far far away of it's competitors, everyone who had the luck to read the UMD Specs was amazed by the power inside @ that time in terms of a PMP :) Next Nvidia is going to blow that away with Tegra but who knows what Sony is going to come up with next in the cycle but somehow i doub't they gonna beat Tegra on the other side im not sure of that seeing the time they had now available to prepare the next version. Archos Imho is as good as dead even being the inventor of PMP they won't survive with their current lowtech stuff vs a Tegra based solution but i guess they will adapt and use Tegra themselves to survive or look for something similar to integrate into their next PMP products they can be flexible in chosing what they need but it won't be unique anymore.