@Butt I completely agree...His 2x the defaults approach seems to be the magic solution. I am not much of a tester...but my eyes tell me the 720p 60 IS VASTLY improved and 44MB looks much cleaner than the Un-hacked.

@sohus not true, the max sustainable at 24P is over 1,000,000, I've had that often shooting with the 66M AQ2 for 2 minute long shots. I've even had shots where its held at 1,160,000 consistently with my 88M AQ2 setting. The upper maximum is actually 1,180,000. If you look at the second deathchart test bkmcwd posted he was getting around 960,000 consistently (actually 1,010,000 in the last deathchart post he made). I'd expect the last test to be much lower as it's a night time shot with less detail. The bitrate drop in the bkmcwd patch is a result of the T1 T4 change, which if I'm right he says he's removed in the last post.

Edit : There is a lot more to stability than just frame sizes though.

Oh, unless you mean a max framesize at low GOP, which would be lower obviously, but not as low as 800,000 at GOP3 I think.

@all
I tested all the settings lowGOP
Result:still the best patch is with 44MBit by CBrandin's GOP 12
- so far are all lowGOP settings for me in practice not applicable
Stutter, stuttering, smearing, and more are the results of LowGOP
The best and cleanest image = CBrandin 44MBit GOP12 - all other settings were lost time

@Stray The B frames or P frames wont be like that in most real-world conditions unless filming trees of death on a static. However, if you look at a normal recording using those two new patches - notice how the P and B frames sit into place!
Its a test to see how well each performs against each other and to see how much I could push P/B frames close to an i frame with possibly getting better performance than GOP1 under similar conditions.
And yes, bkmcwd's patch won't get better than those two tests in terms of bitrate once he gets it stable and tuned.

@sohus yep, I've always had problems with the 1080i on those 66M settings, copying the settings for 1080i from the 44M AQ4 works though.

@sohus Brilliant and thanks mate. Yeah, the one from last night GOP3ILLA - the P frames versus B frames battle. It will be interesting to see how you find B frames comparing to P frames performance in motion and static.

@Stray Of course you can get higher i frames on a longer GOP. The fact that we did GOP1 was an achievement in itself because its currently very difficult to achieve that look without using high bitrate and limiting. The Q values in the GOP3 latest tests have been altered and I would love for you to run a few of your tests against your present GOP3 (or longer) and look at QP. If you havent time then fair enough. A new GOP1 is currently being prepared.
The fact of the matter is for a lot of filming scenarious the current GOP1 and the latest GOP3 patches are more than adequate.

@Driftwood, dear Nick like what promise i take a look of your last 720p50 GOP1 and is not fluid in orizzontal panning situations not only in cam lcd but also on mac in .mts or after converting in Prores.
Now one old theory is that if display are 60hz., like Apple, can't reproduce fluid movements for 50fps. materials, or its wrong???
PLS confirm me what's about this issue, tomorrow i'm in studio and can also test with HD 50hz. projector.
Thx one more time.

For shure I will give a look on your last GOP3 patch too.

@sohus I completely concur on that 176M setting, but thats an odd result from the 132M. Has the 132M been changed recently, as in is there more than one version of it uploaded ? I had no problems with it, but then again I haven't used it for weeks now and my testing regime has got a bit more serious since then. Actually IIRC @bkmcwd said he saw the 132M GOP3 behaving badly recently.

@wigginjs Not really as the 24H and 24L share a lot of the same setting parameters, so that could (probably would) result in instability in the 24L. So the 24L has to be quite close to the bitrate of the 24H setting. You can pretty much combine it with much lower 1080i settings, as they have their own complete set of parameters.

@sohus I do have a nagging doubt though that it may be possible for the 66M max var on AQ3 to break the bitrate maximum resulting in frame rate oddness, if not cadence problems. Although I know I really meant to test that to make absolutely sure I can't remember if I have or not right now (not had much sleep)

@Stray
Thanks a lot! :-)
On the problem of my language, I was always sketchy, sorry!

BTW,I also stick FLx1 Version for a sense.
Although this is the patch set to T4=T1, I do not know to put a check into T4.
I have seldom understood the difference.

And although it is completely satisfactory at the time of the usual recording, I do not know whether it is that its form of cadence at the time of this highest stress is satisfactory.

NOTE:
The camera position is also moving and changing SS and F value with the 3rd chart that recorded many leaves.
Of course, keeping ISO3200.

EDIT:
2nd patch was added.
This is a setting of the same contents as 1st patch except "T4=T1 UNCHECKED".

@driftwood With your new GOP3 I would have thought that having B-frames of the same (and in some cases according to that graph greater) size than I-Frames was pointless. What you can get out of a lower GOP is larger I-Frames than GOP1 is capable of, and therefore better image quality within the same bitrate settings. At the moment according to the graphs you've posted bkmcwds current 154M GOP3 is outperforming this 176M GOP3 considerably. However, this could possibly be because @bmcwkd ran tests with the denser colour chart, which is going to push the codec to create larger I-Frames. But I don't think that of itself will make the I-Frames climb the additional 160000 to come to what his 154M GOP3 is achieving. It could also be due to him running the deathchart test at a high ISO and shutter speed. The other thing I'm not sure of is if removing the rogue P-frame forces a lowering of frame size.

As I showed with a test I ran a few days ago, there is very little difference in terms of image quality between the 176M GOP1 quant me baby and Chris's 66M AQ2 GOP12 (in fact the 66M is slightly better sometimes). I've not compared any of the 176M GOP1s since that one though, but from the graph outputs posted I don't see it being much different. Obviously GOP1 motion is going to look incredibly different to that of a longer GOP though, and maybe the noise would be better too (as in easier to clean). Saying that though, with some small tests I've done, Neat Video and Nukes denoiser seem to clean up the image just as well anyway long GOP or not, but I'd really need to do a lot more tests there. My own subjective view is that GOP3 creates comparable looking motion to the GOP1, not the same but very close. Couple that with greater compression efficiency, and therefore image quality, and it makes more sense to me. But yes, definitely, noise looks a lot better out of the GOP1 before any denoising is done at all.