Is the H.264 world going Baseline/High?

[benwaggoner]

So, are the days numbered for Main Profile?

My take is that High offers a significant improvement in coding efficiency without any real drawback in decode complexity. Thus, there's no reason to make a new ASIC or decoder that does Main and not High.

Baseline, of course, is much simpler in terms of decode complexity, and so will probably the common denominator for battery-powered devices for some time to come.

While there are legacy devices like AppleTV that have Main but not High, is anyone introducing anything new like that? Are there any scenarios where it make sense to have Main but not High?

[Chengbin]

Well, the PSP, and my Archos 5 all only support main profile. It is quite sad that my 5 does not support high profile, when it supports 720p XviD up to over 6Mbps

I thought high profile is very demanding, and that's why a lot of PMP don't support it. An Archos engineer told me high profile is not supported yet because it is a huge amount of work to get high profile working. I assume that the huge amount of work is optimization on the DSP to decode it?

[Dark Shikari]

I thought high profile is very demanding?You thought wrong. The 8x8 transform is not "demanding" at all, if anything it reduces decoding complexity.

[benwaggoner]

You thought wrong. The 8x8 transform is not "demanding" at all, if anything it reduces decoding complexity.
Because of reducing the number of pixels the in-loop deblocking filter needs to touch?

And/or by just making the encoding more efficient?

[benwaggoner]

Well, the PSP, and my Archos 5 all only support main profile. It is quite sad that my 5 does not support high profile, when it supports 720p XviD up to over 6Mbps
There's definitely older ASICs and older device that are Baseline/Main only. Bear in mind that High was added after the initial Part 10 standard, so it lagged productization.

I'm more thinking about future design wins; are there going to be any new products that are Main-only.

The nice thing about the device market is that devices turn over pretty rapidly.

I thought high profile is very demanding, and that's why a lot of PMP don't support it. An Archos engineer told me high profile is not supported yet because it is a huge amount of work to get high profile working. I assume that the huge amount of work is optimization on the DSP to decode it?
Likely "a huge amount of work" in the sense they'd have to switch ASICs. I doubt they're writing their own decoder software for a DSP or anything like that.

[Chengbin]

Likely "a huge amount of work" in the sense they'd have to switch ASICs. I doubt they're writing their own decoder software for a DSP or anything like that.

I think they do write their own code. The HD support was just recently added, even though it was advertised at the product's release 7 months ago.

[Dark Shikari]

Because of reducing the number of pixels the in-loop deblocking filter needs to touch?

And/or by just making the encoding more efficient?Mainly just reducing the bitrate necessary for a given quality or quantizer.

[CruNcher]

Hopefully Main dies finally would also bring down a big interoperability problem in the future (on the road to SVC and H.265) if everything supports High :) i personally like todo my DVD backups in High (for PC watching) for the additional Detail Preservation but this High->Main transcoding for PMPs is just bad and unnecessary another thing the industry borked majorly to annoy consumers (instead of just copy it over i need to wait hours and invest power).
And yeah Mpeg really was way of the track here FreXt should have come much earlier they could have predicted much earlier that Studios needed it the same as Microsoft was able to deliver the quality early on that the Studios demanded, making pressure for FreXt was unnecessary and should have been avoided early on in the Standard then this mess would never have happened, but hey aren't we used to this since Part2 remember the Qpel disaster ;).
And for sure i would never compare vs VC-1 without High it is absolutely useless and that's also part of a major problem most inet content is being done in Main even HD :( and in 99% of what i saw till now WMV (VC-1) is always beating it in terms of Subjective Sharpness (also of course most content creators use commercial crap encoder, currently on the top are Apple and Mainconcept for Inet delivery), though it wouldn't have to be like this :( im sure it's gonna hurt H.264 over the long run in the eyes of content creators this way especialy when they stand in front of the decision in which direction to go @ the end of this year. Only Youtube for now really goes the right way delivering quality based on x264 and High Profile :) (but it's no surprise they actually know what they doing) most content creators though don't (or are forced by Device Interoperability for the lower quality). Of course content creator still have to learn how to get the best out of H.264 and most are used to Microsofts Tools so it's also in the hand of the Codec Developer and ISVs to deliver solutions that are easy to use and maintain quality all the time.

[Mr VacBob]

While there are legacy devices like AppleTV that have Main but not High, is anyone introducing anything new like that? Are there any scenarios where it make sense to have Main but not High?

AppleTV mostly plays Baseline+some extra restrictions content, I think. It's not hard to replace the decoder (an old version of QuickTimeH264) with a High Profile decoder using Perian, but lots of Main/High profile encodes are too complex to play well on its CPU anyway.

Since CABAC is nearly the slowest part of decoding (when SIMD is available), switching to CAVLC decreases complexity, but anything else that reduces bitrate should improve performance in every way.

[akupenguin]

Since CABAC is nearly the slowest part of decoding (when SIMD is available), switching to CAVLC decreases complexity, but anything else that reduces bitrate should improve performance in every way.
Yep. Which makes High+CAVLC faster to decode than Baseline.

[Ryokurin]

I'm more thinking about future design wins; are there going to be any new products that are Main-only.

The nice thing about the device market is that devices turn over pretty rapidly.


I would think it has more to do with device chip prices and availability. If the main-only chipset is considerably cheaper than the high-chipset (shouldn't be but you never know how a few cents can make a difference in costs) and the developer is sure that the main-only won't be discontinued during the life of the product, then why not?

It kind of reminds me of how SiriusXM stated that they won't consider going to a single broadcast standard for at least 10-15 years as they realized how chipset designs can stay in use for years in OEM equipment while everyone else is 4-5 generations ahead, just because the old gen is dirt cheap and still available.