Does anyone have a suggestion to a patch that handles high speeds? I am doing some aerial work from a Cessna Air Plane. Today on a teat run, Driftwood's 176 GOP 1 did not quite handle the high speed. I am running a Sandisk Extreme Pro 45mbts. Could that be the issue?

@driftwood (or another expert around here)
Can you tell me what the FB1, FB2, FL values do? Have you (or someone else) determined their maximum values? I get a very nice stream that crashes after 49 seconds and when I calculate the avg frame, it is about equal to the frame buffer size that is set. Is there a way to enlarge the buffer? At what point is the buffer cleared? Is there a setting for that?

Thanks inhousegoods

@sohus Let me tell you where we're at. Frame buffers in theory can be enlarged or reduced to control the size of i, p or b frames. The problem is getting it right with the amount of bitrate to the video buffer size, and obviously to the quality setting as set by the AQ or manual IQ and Q settings. Its not so bad on zero AQ (or default - everything unchecked - its the same as AQ0 just about) and AQ1 then you have more freedom to adjust the FBs and frame limit. But it still takes time to get it right. 176M with a VBuffer on 36000000 at around a QP (quantiser parameter) of 14 seems to be the optimum point with my tireless testing. Going over AQ1 to AQ2/3 the QP changes quite largely;-
Quantiser Parameters
Default or AQ0 = AQ or IQ / Q left unchecked = 20 giving max/min QP+-2 from 20 (18 to 22) QP. average qlty, great for movement.
if you set AQ to;-
AQ1 = Initial Q will equal 20 when it is unchecked, and min QP of 14, max QP of 18 - good for movement, ok image quality.
AQ2 = Initial Q will equal 20 when it is unchecked, and min QP of 9, max QP of 22, great all round balance.
AQ3 or 4 will lower the min QP even more towards quality.
Of course, everything then uses more bitrate - the better the AQ or the lower you manually set IQ and Q the more bitrate is needed to handle the better quality macroblocks.
There are many variables to keep looking at but we are definitely very near peak stable control with the current 176M settings.
At the moment Im trying ways of lowering the bitrate and pushing quality even more, and of course, lowering frame buffers has a similar effect on lowering the limits. Small increments/decrements (even with the step rule of testing) with each test means its gonna take a lot of time to reach a result.
I am of course looking at other areas to adjust stabilisation. For example the Limit 63000 and the other limits can help adjust the initial cpb delay time - which always must be under 1 second (0.53 to 0.70 seems to be the stable area) - to help keep new settings stable.
There are some strange things happening with the encoder settings too - which I can't comment on till Ive finished going thru about a weeks worth of possibilities - perhaps more. Who knows we may discover new findings soon. :-)
At the moment many tests on GOP1 show that anything around 828,000 sized i frames will cause the encoder to stop recording after about 3 seconds on the various death charts. If you decide to use higher unstable settings in real scenarious you'll probably never get to crash the codec or get to that high i frame size unless youre filming massively high detailed trees of death, chandeliers :-) etc... Anyways, HenryO seems to love unstable settings (dig dig!) so push the frame limit back up to a higher size if you need to ;-)

Sohus, so in an effort to answer your question its a case of continual testing and keep trying to balance all the settings within reasonable parameters. HenryO gets his 'Holy Fail' patch to short bursts of good quality by multiplying the FBs and limit by 4. However, you dont need to do that, you can get similar results just by adding smaller bits to the settings.

Well put together driftwood, thanks

@driftwood
Great instruction!

Thanks @Driftwood. Another question:

Your GOP 1 has a constant 800k quality
If you make the iFrame larger, the P/B frames need to be reduced in size
What would be desirable P/B frames with an I frame of 1,170? How will the quality differ from your constant 800K. In which situations (i.e. bright outdoor, static shots) do you notice an increased quality because of higher iframe, and when is there a disadvantage (probably low-light/dark scenes).

The ultimate setting of course would be 1,170K constant iFrame. But as I mentioned before, I see all settings max out at an average setting of around 820-840k (avg framerate). It really looks like a buffer issue. I have had higher sustained bitrates but the stream crashes within a minute on those.

I want to determine if, at all, it makes sense to have a higher iFrame setting with lower p/b settings at a 154M bitrate.

@driftwood is it cards that cant handle it or the camera itself?

there really is no way to change 4:2:0 to 4:2:2 you cant add color detail in post if it wasnt created from the get go

@driftwood, do you think your 244mbit GOP1 held more detail than QuantMeBaby?

Just had a shocking moment. We shot this morning with driftwood's 176 gop 1, then rewrapped the AVCHD to quicktime H264. Now, with a little time to rough cut a bit, I started transcoding the AVCHD to ProRes (we're still on FCP7). I just looked to see the size, and the ProRes files are all SMALLER than the AVCHD. We're throwing bits away.

I'll now start transcoding to ProRes HQ... but man this was a shock to see. Viva driftwood!!!

Edit: Yup, the ProRes HQ files are matching the look of the AVCHD original, the ProRes is NOT. I need to sit down... wait, I am sitting down...

Captivating Hallvalla....