[mythtv-users] Myth on WD HDTV Media Player for $99

[Brian Wood]

On Sunday 11 October 2009 07:33:02 Jean-Yves Avenard wrote:
> 2009/10/12 Brian Wood <beww at beww.org>:
> > By a "full" OS I was comparing standard Linux distributions to embedded
> > systems thast require a lot less in terms of hardware, can run with very
> > little RAM and little power. The Sigma chips are used with a lot of
> > embedded systems, I'm not aware of any way to utilize VDPAU with an
> > embedded system, though I'd love to hear of one.
>
> Are sigma available for Linux and MythTV, how is that relevant to the
> context of this discussion?

There are Sigma-based units that run Linux, a lot of STBs run Linux, and 
MythTV can be run on PPC systems.

So it may not be relevant to this particular point in time, but in the future 
a MythTV system should be possible using a Sigma type chip with full hardware 
decoding, requiring a lot less in terms of hardware than present-day 
Linux/Myth systems. The proverbial $99 frontend seems to be technically 
possible in the near future.

>
> Most of the embedded system I've seen, especially set topbox, use
> powerpc based core with full hardware decoding

A lot do, others use MIPS, ARM is also popular, though I haven't seen those in 
video systems, but they are very popular with embedded systems (NSLU2). 

>
> > I thought it required the proprietary nVidia video drivers (ie: the
> > binary blob) in order to use VDPAU, at least today.
>
> currently the only vdpau drivers available are from nvidia.
> but the vdpau libraries themselves are open source, and any
> manufacturers is free (and encouraged) to provide drivers for them

Are you saying I can build a VDPAU machine and not use *any* non-free code at 
all? Including the nVidia driver module? Can I utilize the VDPAU capabilities 
of an nVidia card without any proprietary code at all? If this is so, I was 
not aware of it.

I'm not a FOSS purist, but many people are. It is annoying to see 
that "tainted kernel" message, it means we are still dependent on something 
we can't control, and that might go away.

>
> > Do you mean ATI (owned by AMD) and Intel will build VDPAU-capable
> > hardware? Maybe, but by no means certain.
>
> they already are "vdpau" capable, that is can do hardware decoding of
> most content already.
> Would just need to provide a vdpau API, this is well documented, the
> ball is in their court.

I understood that some of the decoding is done by the VDPAU hardware, and some 
by the (proprietary) drivers, are you saying VDPAU is just an API? I thought 
it was more or less like XvMC on steroids, an API, some hardware, some 
libraries and an interface.

I conceed I am not very familiar with precisely how VDPAU works, but I 
understand it requires a lot more from the OS and the drivers than Sigma-type 
systems do.


>
> Both Intel and ATI have introduced their own vdpau-like interface, but
> with the advance made by nvidia being the first, I think it would be
> wiser for them to drop what they are doing and join the vdpau
> bandwagon

Maybe, but commercial companies are usually very reluctant to utilize 
technology "not invented here". They all seem to think they can make *their* 
version of a widget become the universal standard and make them rich.

>
> In the mean time VDPAU isn't half-baked as you implied, it does what
> it does very well, and provided it's comes on video cards designed for
> PCs, of course it's going to require a full fledged-OS and video
> drivers...
> How else could it work differently

My use of the term "full-fledged" was a poor choice, I meant to make a 
distinction between a "full" Linux system with on the order of 1GB of RAM 
(including video subsystem memory) and embedded systems, sometimes running 
with well under 512MB of RAM and a minimal OS.

When VDPAU first appeared it was said you needed at least 512MB of RAM just on 
the video card (or perhaps stolen from the system RAM, in the case of 
portables). I think this requirement has come down recently, but 512MB plus 
enough RAM to run an OS is a LOT of memory.

I guess I'd like to see a Myth frontend become a $99 appliance and not 
something requiring a full-blown PC, VDPAU is "half way" towards that goal, 
hence my comment. Perhaps my vision of the future is not that of others.

I certainly agree that VDPAU does what it does quite well, and fairly cheaply, 
it's certainly been adopted quickly as its advantages become apparent, but it 
has not really reduced the cost or complexity of Myth systems significantly.

nVidia's main customer base is still gamers and high-end CAD systems, neither 
of which is trending towards low-power embedded systems, they both need 
significant computing power, which video display should not require.

On the road to a $99 embedded Myth frontend both Sigma and Broadcom seem to be 
further along. VDPAU is a way to re-purpose hardware built for CAD/gaming to 
do video display, the goal is different.

-- 
Brian Wood
beww at beww.org