>>"You are referring to Analog connection between the HTPC/Scaler and D-ILA aren't you? If so, that's where the problem lies - in the resampling - not in the scaling per se."
Yes, I'm referring to the analog connection.
It does depend -- sometimes it does not make a significant difference, depending on how the scaler does it.
If the resolution is really filtered (i.e. softened) horizontally in a manner so that the upconverted signal is more truly "analog" where there is no clear 1:1 pixel mapping, then it's more likely more flexible to variances in final A/D conversion (within the digital display) and 1:1 pixel matching is less important here.
For relatively unfiltered scalers that use fully clearly defined pixels horizontally (especially HTPC's and many PC-appliance based scalers like those I currently work with) it becomes very important to have 1:1 pixel mapping (dotclock matching between scaler output and digital display's native rate). In this latter case, there is more control over the final appearance of individual pixels when D/A (scaler output) and A/D (digital display) are much more perfectly matched. The results can look better or worse, since the burden is now potentially nearly fully on the scaler on the final appearance of each single pixel, depending on how good the scaler deals with these individual pixels (interpolation, deinterlacing, etc). There is, of course, the additional overhead of requiring the end-user to calibrate the scaler/projector to run with as flawless 1:1 pixel mapping as possible via tracking/phase/dotclock adjustments on the display itself, or via timings adjustments on the scaler/HTPC itself. But the results can be well worth it for power-tweakers like many of those found on AVSFORUM!
Of course, this more applies to 720x480p being upconverted -- since the benefits of 1:1 pixel mapping is far more noticeable during downconversion than during upconversion than downconversion (as in 1920x1080p down to 1280x1024p) ....
Of course, 1:1 pixel mapping on digital displays really only shows noticeable benefit for scalers with unfiltered or relatively unfiltered outputs (ie, scalers that output really clearly-defined pixels that don't blend gradually into horizontally adjacent pixels. So that's easy to A/D back to digital within the digital display electronics without moire artifacts) .... I'm talking about pixels at the destination resolution level (output resolution).
When I use the terminology "filter", I specifically mean how well-defined each output pixel is. Unfiltered would mean clearly defined pixels that has little analog blend between them. Filtered would mean lots of analog blending between horizontally adjacent pixels. PC computers, for example, typically use very unfiltered output (or comparatively minor filters, that often-picky videophiles sometimes still remove anyway) because graphics card makers want each computer pixel to be as razor sharp as possible. Now in the scaler market, that isn't always good (unless, obviously, you're outputting to a digital display AND you do very good scaling/deinterlace algorithms). Just need to make sure my use of the word "filter" is not confused for other terminology. Basically, when I say "unfiltered" here in this particular message - it means minimum analog blending between horizontally adjacent pixels in the output resolution. (Just making sure that techy types here doesn't confuse this with other types of filters like digital noise filters, digital sharpness filters, 3D comb filters, etc.)
Picture quality goes in this order, when we are dealing with digital displays.
1. Best: 1:1 pixel mapping, unfiltered output
2. Middle: 1:1 pixel mapping, filtered output
3. Middle: No 1:1 pixel mapping, filtered output
4. Worst: No 1:1 pixel mapping, unfiltered output
This is assuming same digital deinterlacing/scaling algorithms are being used in all cases, and the bandwidth of the scalers is high enough for good 1:1 pixel mapping at the specified output resolution. Also, of course, that the algorithms are at least halfway decent: Scaling algorithms that aren't worse than the horizontal interpolation advantage provided by filtering.
Those older scalers with bad scaling, it's possible that the filter provides better "interpolation" than the bad scaling in the old scalers. However, many new scalers provide scaling interpolation that can be superior to the horizontal interpolation caused by analog filtering.
When dealing with scalers with output that's unfiltered, you may see quite a dramatic difference when 1:1 pixel mapping is achieved, especially if you sit very close to the screen.
That's why it's not generally recommended to use an unfiltered scaler with a digital display (aliasing, for example!) - UNLESS 1:1 pixel mapping is achived AND the scaling/deinterlacing is good. Then it can yield superior results.
On the other hand, when dealing with scalers with output that's filtered, the picture quality difference is not quite as dramatic.
For plug-and-play (not needing to worry about horizontal resolution), it is often easier to go with a filtered scaler .... Since A/D converters are more forgiving of properly and fully filtered output (one that has a really analog-like output that doesn't appear to have a specific horizontal resolution at all)
Even the most dramatic differences, of course, may not be noticeable to average "Joe User" eyes. But it's most obviously noticeable to many well known videophiles who frequently post in this forum.
Of course, things are different when we are talking about CRT's since there's no such concept of 1:1 pixel mapping on such displays.
Although most experts here likely know, I also need to mention this little point for others who may need to know: Just doing a specific resolution doesn't automatically do 1:1 pixel mapping -- the resolution needs to be native scanrate matched with the digital display. With proper timings, including dotclock and horizontal and vertical retrace intervals, etc.
I believe it's quite possible that the amazing TeraNex scaler will probably be even more amazing inputting native 1365x1024p into a DILA projector (properly calibrated) rather than the already impressive 1920x1080p into the same DILA projector! Since TeraNex seems to have so much control over each individual output pixel, and do quite an awesome amount of processing already on each single output pixel, it would therefore follow (if the TeraNex output filtering is adjustable or can be turned off) that superior results are possible with a TeraNex outputting 1365x1024 with DILA-precise timings versus TeraNex outputting 1920x1080 when it's being connected to a DILA projector (which has a native 1365x1024 resolution).
(1360x1024 or 1368x1024 could theoretically be used, since many graphics chips are designed to output in horizontal resolutions divisible by 8. In this case, there's minor horizontal overscan or underscan, when 1:1 pixel mapping is achieved.)