[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [AG-TECH] Computational requirements for the future
Hi Jason,
We have been playing around with the new encoders for almost a year
now, and are using dual-core AMD-64 X2 4400s in ASUS A8N-SLI Premium
motherboards for our rooms. That motherboard has 3 PCI slots for
encoders, and 2 PCI-express slots. We use two GeForce 6200 twin
head cards for support of three projectors and one control screen.
A 720x576 MPEG4 stream at 20 fps will take about 35% of a core to
encode and decode, with computation split around 2/3 encode 1/3 decode,
so you can encode 3 cameras and decode about 8.
This is with Fedora Core 5 i386, which has problems with the nvidia
driver and has to use the X.org nv one. Under FC5 x86_64, the nvidia
driver works, and improves performance by about 5% per decode stream.
The 64 bit version does not support shared presentation, however.
We are using 2 cameras only, as the resolution is so much better and
you don't need that many cameras to show all the participants. That
would allow us to receive about 10 streams. (The third PCI slot is
used for a Phoenix SOHO echo canceller).
H.264 has prohibitively high compute requirements, which prevent it
from being used at frame rates above around 5 fps (40 % CPU per
encoder). We use it for personal nodes at low bandwidth for
document cameras at 1 fps, and it works brilliantly for that.
Decode CPU requirements are about the same as for MPEG4.
You could try a really fast CPU (AMD FX 60 or so) at full frame rates
for one H.264 encoder per box.
HDTV is not worth it as you cannot find a projector bright enough
for video conferencing in a non-darkened room for under around
$AU 400,000.
We should arrange a video conference.
Regards,
Claus.
Jason Bell wrote:
G'day All
I'm in the process of configuring up some equipment for a couple of new
room nodes.
The problem I face is deciding on what type of machine I will need to
power some of the new "Features" that we may be confronted with over the
next couple of years. I guess what I am really interested in knowing is
the computational requirements of some of the "newer" video codec's.
For example, has any done testing on the following???
* H264 and/or Mpeg 4
* DV
* HDTV
* Others that may be used in the future
What sort of CPU processing power have you tested with, have you found
that the encoding it the computational intensive part, or is the
decoding pretty computational intensive as well???
Knowing this will determine the types of machines required for video
capture and video display side of things for an Access Grid.
The reason for this question is that a single CPU, Dual Core computer
seem to currently have the best bang for buck value. I would be
interested to know if you personally think that (for example) a dual
core 2.8Ghz workstation would be capable enough to handle many of the
above situations.
I guess another thing to consider, is that if the encoding side of
things is the most computational intensive part, then would using 2
machines for capture be better than say using a dual CPU machine (Around
the same price).
I would love to hear people's thoughts, comments and various tests you
have performed.
Thanks for your time,
Jason.
--------------------------------------------
Jason Bell, B.I.T.
B. Info. Tech. (Honours) Student
Network Engineer
Information Technology Division
Central Queensland University
High Performance Computing Support Officer
Central Queensland University
Research Assistant to the Nonlinear
Intelligent Systems Research Group
Faculty of Informatics & Communication
Central Queensland University
E-mail : j.bell@xxxxxxxxxx
Phone : 07 4930 9229
--------------------------------------------
Patience is a virtue.
But if I wanted Patience,
I would have become a Doctor.
--------------------------------------------
--
--------------------------------------------------------
Claus Endres | Phone: +61-3-5998 2310
Endres Consulting Pty. Ltd. | Mobile: +61-418-595 136
10 Facey Road | Fax: +61-3-5998 2540
Devon Meadows, VIC 3977 | claus@xxxxxxxxxxxxxxxxxxxx