[mythtv-users] Graphics card recomendation

Michael T. Dean mtdean at thirdcontact.com
Thu Aug 31 00:59:46 UTC 2006


On 08/28/06 08:35, Steven Adeff wrote:

>  On 8/28/06, Dylan Semler <dylan.semler at gmail.com> wrote:
>
> > On 8/27/06, Steven Adeff <adeffs.mythtv at gmail.com> wrote:
> >
> >> On 8/27/06, Dylan Semler <dylan.semler at gmail.com> wrote:
> >>
> >>> My highest ambitions would have me try to take the two monitors
> >>> apart, perhaps do a little machining, and try to rig it up so
> >>> it looks like one very big and very wide monitor. Though, I
> >>> guess the chances that this is even possible are very, very,
> >>> very small.
> >>>
> >> just get a bigger monitor/HDTV.
> >>
> > I haven't looked extensively, but it seems that any monitors that I
> > would find of the 2200x1200 genre would be prohibitively
> > expensive.
> >
>  But you don't need that resolution for Myth... at most you could use
>  1920x1080(1080p).
>

Or, really somewhere around 3840x2160 would be an appropriate output 
resolution for a 1080i/p input resolution...  In addition to the fact 
that (as previously mentioned by Daniel) 3840x2160 is an integral 
multiple of both 1280x720 and 1920x1080, sampling theory (specifically, 
the reconstruction theory part of it) says that an output resolution 
must be greater than an input resolution to fully represent the detail 
in the image.  As a general rule of thumb, the output resolution needs 
to be at least 2x the number of pixels on each axis (i.e. 4x the pixels 
of the input signal).

Yes, I know I took a lot of heat last time I said this, but last time I 
was on the road for a few weeks and didn't have the time to "prove it," 
so I eventually just let it drop.  In fact, the truth is, I'm not smart 
enough to prove it, but I do ask that before anyone writes back with 
messages saying, "How can you need more than 1920x1080 pixels to display 
a 1920x1080 image?" or "Well, actually, if you upscale the image, you're 
just 'inventing' new information, meaning it [adversely affects picture 
quality|displays a made-up image, not the captured image]," that they 
read some of the many, many good books on sampling theory and 
reconstruction theory.  Instead, if you feel you must disagree with me 
and don't feel like reading up on reconstruction theory, just write me 
off as some crazy guy who shouldn't be given e-mail access.  (Oh, and 
make sure you post all sorts of messages about how crazy the marketing 
guys must think we are when they start trying to sell 3840x2160 displays 
even though ATSC defines a maximum resolution of 1920x1080.  I always 
enjoy a good rant--especially one with an "inside joke.")

Boiled down to basics, there's a difference between image pixels (which 
are truly "picture elements"--samples of a picture at a point (of zero 
size)) and display "pixels" (which are really "dots"--that have a 
physical area).  The picture elements are created by sampling a 
continuously-defined image function, and, although you can display an 
image by painting pixels of the same value and at the same positions 
used to generate the picture elements (i.e. "1:1 pixel mapping"), you 
can create a much better image by recreating the image function and 
taking more samples at different positions and incorporating the 
additional information about the image function into the final display 
pixel values.

Oh, and, of course, dots aren't "just" dots.  Each display pixel's 
brightness can vary in intensity across its "surface" (i.e. as on a 
CRT).  And, display pixels can vary in intensity across the display 
(i.e. having a directionality as on an LCD).  And...  But, that's a 
whole different argument.

Unfortunately, there aren't many good sources on the 'net.  Why?  Who 
knows?  Perhaps computer programmers are too smart to be fooled by all 
those complex mathematical formulae.  Fortunately, though, the ideas 
have been incorporated into the algorithms we use everyday (in 
image-processing libraries, in printers/printer drivers, and even in 
graphics hardware).

Mike



More information about the mythtv-users mailing list