[mythtv-users] Trouble with HD playback

Anthony Vito anthony.vito at gmail.com
Fri Nov 12 15:52:52 UTC 2004


> very interesting paper.  I looked at the conclusions and it seems to
> indicate it's not memory or CPU bandwidth that improves the HDTV
> performance, it's ordering of instructions and data retrieval to
> optimize cache usage plus the use of MMX type instructions which causes
> a 4 fold increase in performance.  You need to write the software
> correctly so it can access the memory and instructions in the right
> order so to speak to improve cache usage.  If memory bandwidth and clock
> speed were the whole thing then a P4 3GHz wouldn't get it's ass kicked
> by an Opteron at 1.8GHz, clocked at over 1GHz slower would it.

Exactly right my friend! The problem is that the mythtv mpeg2 decoder
isn't as advanced as the one put forth in that paper. ( Maybe we
should send the guys an email and see if they will donate their code
for the advancement of all man kind! ) So without optimizing cache
usage, memory bandwidth becomes a big issue. The Opteron 1.8Ghz "kicks
ass" in mpeg decoding because a few key ehancements. First is the on
die memory controller, which lowers main memory access latency, next
is the extra hidden registers, and solid improvements to the TBLs.
Again, all cutting down on that all important amount of main memory
access. That and lets face it... The Athlon 64 cores are frickin
amazing!

> After all they got a P3 933 to decode 1024x768 HDTV video just fine...
> It's not the size of your boat, it's the motion of your ocean :D

I feel inadequate

( it was 720p (1280x720) they got running on a P3 933 at better then 60fps!! )

> I don't have HDTV sadly yet.  As soon as I do I will try with what's
> available, but I would probably use an AMD 64 1.8GHz @ 2700.

Sounds good to me.... Anyone want to buy a custom liquid cooled dual
XP1800... I want to play with x86-64 :(


-- 
Anthony Vito
anthony.vito at gmail.com


More information about the mythtv-users mailing list