Archived from groups: rec.video.desktop (
More info?)
On Thu, 27 May 2004 10:38:40 -0400, "Morrmar" <morrmar@myway.com-no
spam> wrote:
>> >> Can someone explain why the common solution uses hardware encoding
>but
>> >> decoding is left to the CPU?
>>
>> There you go. The demands for good software encoding are much higher.
>> In fact, most of the PC-based realtime software encoders, in order to
>> run at 1GHz or whatever, compromise the MPEG-2 quality considerably.
>> Hardware is the best solution for this -- it produces a better result.
>From my mp3 days, I've learned to stay out of the "my encoder is better
>than your encoder" arguments, since almost no one can tell the
>difference between encoders in a true double-blind study. But... do you
>have any real world tests/comparisons for your "hardware is best"
>statement? Realtime may be the key word, as most quality s/w encoders
>use VBR and multiple passes but with today's fast processors, I would
>think the "quality" difference would be insignificant.
"Realtime" is the keyword. I agree that the best of the software
encoders are all pretty damn fine these days. I can tell the
difference between, say, MainConcept (though I haven't tried their
2-pass version yet, I'm waiting for Vegas 5 to arrive) vs. TMPGenc,
but only on certain bits of video.
But I have yet to find a realtime software encoder I'd even consider
acceptable, much less "decent". I have tried the GoMotion encoder
bundled with the ATi AiW cards, several other built-in on PVR
applications, etc. Most are pretty hideous -- you'd swear the GoMotion
encoder, for example, pulls a silk stocking over your "lens" (well,
that's the effect).
>I've compared mpg
>files from my AIW realtime captures with files that were encoded with
>ProCoder, a pretty high rated encoder, and I can't tell the difference.
That difference has repeatedly hit me over the head like a canonball.
It's not just the "through a stocking" effect, but dramatically worse
motion artifaces, other things, versus a proper, professional-class
encoding with a stand alone encoder.
There's only so much you can do in realtime, in software. For example,
the motion search cannot be done as well. So they're either running a
very limited search, or a heurestic search, neither of which will
always work as well as the in-depth, very time consuming searches you
can perform with the better encoders.
>For some reason or another, I have a tough time believing that an mpg
>file from a $200 USD Hauppage card is going to be "better" quality than
>an avi file that is then encoded with Cinema Craft Encoder, costing
>$2000.
Well, how about TMPGenc, costing $45 or whatever I paid for it, years
back, which is by most accounts nearly as good as the $2000
CinemaCraft. But no, I'm not making that claim, either. Hardware can
be very good, though, in most aspects, it's untweakable -- you cast
much of the algorithm (if it's a pure HW encoder) in silicon. So there
hasn't been the competition we've seen for software encoders on the
PC. But there are some advantages to hardware. For example, you
_could_ do a motion search on every pixel at once, without spending a
great deal of silicon, by today's standards. CPUs are still doing this
one by one, more or less (maybe MMX/SSE could speed up some of this,
but not by the same factors).
Or, you could look at an existence proof -- commercial DVDs are made
using hardware encoding. Of course, the real reason they're better
than ours, most of the time, isn't that, but the fact they're
hand-tweaked, frame by frame, by an MPEG-2 mastering engineer. The
hardware speed just makes that possible...
>I have to believe there's a reason why one costs ten times more
>than the other and I'd bet "quality" is somewhere in the mix.
There is precisely one reason: "because they can get it". It could be
the actual quality, the preception of quality, the reputation of
quality, etc. I have a program for DVD creation from Pinnacle, called
Impact Pro DVD. When this was owned by Minerva, they charged $6,000 or
some-such for it. Pinnacle charged me about $300, for a very slightly
improved version of the program. Did the program somehow stop being
20x as good? Nope.. but others came out that were better, or easier,
or more full featured, etc.
The only
>encoder comparison that I've ever seen, that used truly objective
>methods, is here:
>
>http://videosystems.com/ar/video_mpeg_encoder_shootout/index.htm
>
>and it judged CCE the "best", even over a $25,000 Sonic Solutions
>SD-2000 hardware encoder.
If it's intended for professonal use (which I would assume it is, at
that price), probably so. I don't think anyone doing a commercial DVD
(well, aside from self-published stuff like mine) would consider using
an encoder that's simply run on a "set 'em and he'll do his best"
basis. You can't master a CD simply by dialing in a few EQ settings
and streaming your production through it -- you need to listen
carefully to each part of each song, and tweak accordingly. Not
everyone has the ears for this. Video mastering (which, after all, is
precisely what DVD encoding is) need similar human interaction to
produce the best result. That's what you do with a high-end hardware
encoder (I'm not suggesting that's what a $200 PVR card is for, of
course). As you argue in CCE-SE's favor, "if they're selling it for
that, someone must see the value".
>Have any links to objective tests that demonstrates h/w encoding is
>"best"?
Nothing offhand. I worked in digital video, almost daily, from the
early 90s to the present (not The Day Job for the last three years).
You get used to what's what... at least until technology flips it
around, but I don't think that's quite happened yet. Just that there
are different tools for different kinds of encoding.. we've discussed
three very different cases here (realtime PVR, prosumer software
encoding, professional MPEG-2 mastering), there may be others.
Dave Haynie | Chief Toady, Frog Pond Media Consulting
dhaynie@jersey.net| Take Back Freedom! Bush no more in 2004!
"Deathbed Vigil" now on DVD! See
http://www.frogpondmedia.com
Facebook
Twitter
Instagram