Am I the only one or doesn't this issue bother other people?
From the first moment I installed my HD2600XT it suffered from something that looked like the notorious Chroma Upsampling Error when hardware acceleration enabled and using either motion adaptive or vector adaptive deinterlacing. A google search reveals some other users bothered by the error.
Some two months ago Arfster posted a few screenshots in another forum, with different deinterlacing methods with either an ATI 2400 or nVidia 8500 (if I'm correctly). In these shots, the colour displacement was clearly visible with the ATI adapter while not present with the nVidia.
I have submitted a ticket with ATI some time ago, where the support representative acknowledged the problem anbd sent data to the catalyst crew. On my query about proceedings yesterday I got this answer:
"What will be fixed in a driver update is never published before the driver release.
Typically the next 3-6 drivers are at different stages of development at any one time.
I have forwarded the feedback to the Catalyst Crew team. The more they see the issue the better.
If you can get other users to submit tickets as feedback on the same issue they will go directly to the Catalyst Crew Team."
This was a response to this ticket:
"As an addition to the screenshots I previously sent, it is quite simple to view the effect in catalyst control center. Here, while choosing a deintelace method, the red ATI-banner is correctly put on screen with BOB and adaptive modes, but shows an awfull shift of the red color upwards when using motion or vector adaptive mode.
Obviously, for HTPC use, the vector adaptive mode yields absolutely the best picture quality, by using extensive post-processing.
Could we (myself and the HTPC community using the HD2400/2600 display cards) get an estimate on the CCC version that corrects the color-shift problem in the high performance deinterlace modes?
In other words, the more people complain, the faster the response might be. Which is propably the answer to why ATI fixes problems in games faster than video problems, my guess would be that there are a lot more gamers than there are HTPC users.
The attached files show the colour offset in a real-live situation, the software.jpg is a photograph of a part of the screen with bitcontrol decoder in software mode and hardware deinterlacing (which then defaults to adaptive mode), the hardware is with the avivo decoder with forced vector adaptive mode in CCC.
I was about to replace the 2600XT with my good old X1950, when I decided to play around a bit with ffdshow. To my surprise, ffdshow can use NV12 mode (enabling full hardware acceleration, as shown by deinterlacing mode in vmr9 [the 3C-thingy] and enabled detail-enhancement) while shifting the colour down by one line. Thus getting rid of the colour displacement, but still having the advantage of VA deinterlacing.