[mythtv-users] FYI - New HW video accellAPI from NVIDIA

Nicolas Will nico at youplala.net
Sun Nov 16 22:17:43 UTC 2008


On Sun, 2008-11-16 at 13:13 -0800, Justin The Cynical wrote:
> Nicolas Will wrote:
> 
> > You are talking about PureVideo.
> > 
> > NVIDIA is really talking about PureVideo HD. It looks like the
> circuitry
> > on-chip is quite different, thus needs different code.
> 
> Ah, I wasn't aware that there was a PureVideo and a PureVideo HD
> available.
> 
> *looks a bit more*
> 
> Hm, I think my point still stands...
> 
> > If the marketing names are about the same, it doesn't mean that the
> > engineering is.
> 
> 
> Agreed.  However, the the difference between the two appears to be
> more 
> marketing than engineering.
> 
>  From what I can tell at the PureVideo HD page and associated PDF's,
> the 
> only difference between the two is the HD version of PureVideo is
> that 
> the HD version is HDCP Capable.  Heck, the product comparison PDF on
> the 
> PureVideo and PureVideo HD pages are the same file that I looked it 
> showing that every chip back to the Go 6600 support h.264
> acceleration 
> under PureVideo.
> 
> Even if a given video card can't move everything to the GPU, which is 
> one of the selling points in regards to the 8xxx series cards, being 
> able to move even parts of the h.264 pipeline to the video card would
> help.
> 
> Don't get me wrong, I'm glad that NVidia is finally bringing this to
> the 
> Linux community as it's past due,

"past due". Shooting at the ambulance, heh?

You see, a few days ago I would certainly not have placed any bets on
NVIDIA delivering anything regarding hw accel of video codecs on Linux.

My bet would have been on Intel with an open source driver.

I was wrong. Tough luck. In the end, I'm so glad anyone shot first.


>  but it does seem to be more of a 
> marketing move than a answer to something many end users have been 
> wanting for some time.

Ah.... end users...

Do you know who NVIDIA's end users are?

Let's say that a year ago I was in front of a tricky problem for a very
high end setup, involving 4k projectors, 2 Quadroplex units and passive
stereo. I ended up with NVIDIA's chief engineer on the phone. He said he
understood the problem. He said that it could be solved. He said that it
would require engineering time. He asked how many situations like mine
he would encounter world-wide. I answers that one hand should be enough
to count the situations, even if we are one of the 2 big guys in the
field.

I was basically told that my issue was not helping gamers at all, that
he would need to evaluate retasking engineering resources.

That was a polite "never".

I can't blame him.

How many Linux users willing to play HD does it take to retask
gaming-oriented driver developers at NVIDIA?


In the end my system worked, more out of luck than anything. Those guys
are not as bad as they are pictured most of the time.

> 
> Regardless, this is all speculation as I don't have access to the
> lower 
> level GPU design specs, and I really couldn't understand most of it
> if 
> it was available (I am not a hardware engineer).  I may be very wrong 
> and there may be some very large design differences that are keeping 
> them from implementing this newly announced API across more of their 
> chipsets.
> 
> However, what NVidia is stating with this announcement and PureVideo
> HD 
> seems to ignore what they have stated earlier with the 6xxx series
> chips 
> and PureVideo in regards to chipset support.


Well, I was surprised by the limitation on chips supporting VC-1, as
advertised in the announcement.

So I pulled my email client and emailed the NVIDIA guy. 

The answer came quickly and nicely. It is very much in line with
hardware differences, and it is very much in line with what I know from
professional contacts with NVIDIA (much different field, advanced Oil
and Gas visualization stuff): 

        The only currently launched NVIDIA GPUs with support for the
        entire
        VC-1 pipeline are the GPUs listed in the announcement snippet
        you quoted
        (i.e., G98-based products).  I believe most future GPUs will
        also have
        this support.
        
        GPUs such as G84 and G86 could accelerate part of the VC-1
        pipeline, so
        we may investigate trying to expose that at some point in the
        future,
        based on end user feedback.  But for now the priority is to
        stabilize
        the support in the cases that the GPU accelerates the entire
        pipeline.

There are hardware differences. It may not be what you like to hear, the
marketing is not helping, but I really trust that this is true. CPU/GPU
architecture is not trivial.

Nico



More information about the mythtv-users mailing list