[mythtv-users] Temporal 2x vs Temporal-Spatial 2x

[Mark Lord]

On 12-02-23 10:13 PM, Steven Adeff wrote:
> On Thu, Feb 23, 2012 at 9:19 PM, Phil Bridges <gravityhammer at gmail.com> wrote:
>> On Thu, Feb 23, 2012 at 9:13 PM, Mark Lord <mythtv at rtr.ca> wrote:
>>> On 12-02-08 12:26 PM, Mark Lord wrote:
>>>> On 12-02-08 09:31 AM, William Powers wrote:
>>>>>> I didn't think the 520 could do Advanced 2x on 1080i? I would like to know
>>>>>> if you are indeed using it, and if you don't get skipped frames.
>>>>>>
>>>>> In a half-height frontend I have tried a passive G210, a passive GT520 and a
>>>>> GT220 with a fan. At least I think it's a fan. It sounds like a coffee
>>>>> grinder and the card eventually overheats anyway. But it's supposed to be a
>>>>> fan. In a full-height frontend I have tried an 8400 GS, a 9500 GT, and a GT
>>>>> 430, all passive. Only the 220, 430 and 9500 would reliably do Advanced,2X
>>>>> and, according to qvdpautest, the 430 and 9500 didn't have a lot of headroom.
>>>> ...
>>>>> Fermi stream processors appear to be about half as capable as pre-Fermi
>>>>> units. Whether at is due to drivers or the design, I cannot say.
>>>>
>>>>
>>>> That's interesting.  So I just now went and ran the only objective benchmark
>>>> that seems to exist -- qvdpautest -- on my shiny new ASUS fanless GT430,
>>>> and compared results with an older run on the previous
>>>> (2-slot) Zotac fanless GT240 card.
>>>>
>>>> The GT430 is faster at decoding than the GT240, twice as fast for VC-1,
>>>> and somewhat faster on all of the others.
>>>>
>>>> But the GT240 blows it away for de-interlacing modes, by a wide margin.
>>>> The GT430 still comes out with room to spare for Advanced/2X,
>>>> but not nearly as much as the GT240.
>>>>
>>>> Both cards have 96 stream processors, but the GT430 is Fermi, and the GT240 is not.
>>>
>>> Time for a followup:
>>>
>>> The GT430 card does indeed run into difficulties (jerky playback)
>>> when "vdpauhqscaling" is enabled and a high-resolution interlaced file
>>> is being played (and scaled).
>>>
>>> The older GT240 card did not have any issues with that, but the GT430 does.
>>>
>>
>>
>> Hmmm - looks like there's a trade-off between HD audio and HQ video.
> 
> Mark,
> 
> what is being scaled in this case? ie 720p->1080p? I wonder if
> hqscaling is really needed for that, and if not, perhaps a profile
> that enables it for converting SD content to HD but does not for 720p
> to 1080p?

The recordings that play back really poorly are 1920x1080i ATSC,
where the source material has been "double letterboxed",
and I'm having Mythtv zoom to expand the actual content portion
to full screen ("half" setting).

But even some straight 1920x1080i without zoom is stuttering.
MMMmm.. that is probably due to having Mythtv expand the image by 1 pixel
all around to eliminate the "data line" that otherwise flickers at the
top of the image.

So I've modified the playback profiles to not use "vdpauhqscaling"
on files larger than 1280 (horizontal), and playback is acceptable again.
Too bad one cannot (easily) qualify that to apply only on _interlaced_ content
though, because for non-interlaced there's plenty of leftover horsepower.

My Atom-2 machines also cannot use vdpauhqscaling for HD files,
but I didn't really expect that from a full 16X PCIe GT430 card.

It's a crying shame that Nvidia cripples the Fermi processors so much
(artificially in the drivers, to prevent use of cheap "consumer" hardware
in what are intended to be high-end expensive systems etc..).