[e2e] TOEs and DVDs

[Charles M. Hannum]

On Thursday 08 April 2004 06:29 pm, Craig Partridge wrote:
> Hi folks:
>
> Now, today, I suspect that DVD bandwidth doesn't stress a memory system.
> But it surely did a few years ago.  What did the video crowd interested
> in making general puropse computers into competent DVD players do in
> response?  Something like a TOE -- something else?

As someone who's worked on multiple DVD player apps, I think I can answer 
that.

There are three important changes that have occured in video cards:

1) Hardware scaling and colorspace conversion.  Most people want to play their 
DVDs "full screen," and with ever-increasing display sizes, this was taking 
not only a lot of processor time, but a lot of additional bandwidth to the 
video card to get the scaled data out.  Now it's all done on the card.  This 
is probably the most important change.

2) PCI/AGP DMA support.  This allows the card to pull data out of host memory 
in large burst cycles.  Pre-PCI video cards required us to blit the data in 
with programmed loops on the CPU, which was typically quite slow.  There were 
some earlier hacks (like USWC) that helped, but not nearly as much as DMA 
support.

3) Support for MPEG-type motion compensation in hardware.  This enables the 
video card to do part of the frame-to-frame blitting operations needed for 
motion compensation without touching host memory.  (This actually doesn't 
help as much as you might expect it to.  Most vendors include it because it's 
actually quite cheap relative to the complexity of a modern video chip.)

I'm not sure anything of these things is really equivalent to a TOE, though #3 
comes the closest.  #1 is about not pessimising what you're trying to send to 
the video hardware, and #2 is about making the transfer itself more efficient 
-- both things that don't seem to me to have much bearing on current 
networking peripherals.