On 03/08/12 08:38, Dieter BSD wrote:
Their conclusion: "For the time being, the best option for quick,
high-quality video transcoding is unfortunately to buckle down,
get yourself a fast CPU, and run the best software encoder you
can find (which may be Handbrake)."
...
Still, does anyone disagree with their conclusion? Will using the
CPU always be the best choice for quality, or does the GPU
software just need improvement? Implications for OGP?
My interpretation is that this study just confirms the ancient
engineering rule "fast, cheap, good: pick two."
For a long time CPUs weren't fast enough to decode MPEG-4 in
real time, but GPUs didn't have the right instruction sets to
do decoding either. So a block of dedicated hardware was
essential for any kind of video decoding.
The main drivers behind the "black-box" encoders are mobile
devices and wireless streaming.
Since around 2000 desktop and laptop x86 CPUs have been fast
enough to decode in real time. It's a bit tougher for the
ultra low power CPUs in phones and tablets though. A modern
GPU could be used, but the problem is that you're switching
on a lot of hardware which drains the battery. Much better
just to activate a small dedicated logic unit.
Until recently *encoding* was something you did offline with
a render farm, whether for DVD production or YouTube. Now
though we have Apple AirPlay and similar requirements to do
streaming *from* a laptop/desktop/tablet to a remote display
over wireless.
A GPU doesn't work for this because you have to compress
the frame buffer in GPU memory then copy back to the CPU
to send them out over the network. The copy time cancels
out the efficiency of the GPU. (I remember seeing a write
up of a specialised GPU board which had it's own Ethernet
port to get around this.)
The latest Intel CPUs have a dedicated MPEG-4 black box on
the CPU. Apple just announced AirPlay for Macintoshes, and
surprise surprise it only runs on those models with just
such a CPU.
Implications for OGP: I'd first make sure the GPU can run
an MPEG-4 decoder, and maybe leave space on the board/chip
for a dedicated hardware block.
cheers,
Hugh Fisher
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
