.m2ts

Theoretically it is possible edit .m2ts with FCP without transcoding to AIC. Practically however it is either too hard to create the workflow (mainly because software is lacking on OSX) or may be even impossible - FCP errors out ("Not enough memory") and crash on .h264 muxed to .mp4 with both ffmpeg and mp4box (thus it gotta be the stream de-muxed with xport that is bad)...

On Windows, CS3 doesn't recognize AC3 audio (even with ac3filter.net installed) and 2ch AC3 is what AVCHD camcoders capture, CS3 have no problem with PCM audio (FCP could import .AC3 as well as PCM)... So it is possible to edit .m2ts even with CS3, yet frameserving is slow, disk space is cheap and on 720p->480p transcode with video filters in post, saving to lossless Logarithm is twice as fast as frameserving... pretty much means that CS4 could be better using "saving to Logarithm" workflow. Logarithm takes about 2GB per 1.5 minute of 720p@30.

AVCHD cameras lay down h264 at 12-17Mbps. Those streams could be transcoded to 2-4Mbps (720p24) at about 5+xT or 4-6Mbps (1080p24) at ???xT. Transcoding may or may not make sense for Aiptek's 4Mbps 720p30, but it makes a lot of sense for AVCHD, so CS3 or CS4 is the editor of choice for AVCHD (especially considering that AVCHD is not supported natively in FCP and workflow on Mac sucks and need Windows for post anyways)

Aipteks clips that don't need color correction and post could still be edited in FCP, but once I am done playing with what could and can not be done in post, better workflow would still be CS4 keeping originals and may be figuring out how to interchange projects between CS4 and FCP.

So overall it looks like CS4 is the editor of choice that is when it would work

• 4.0.0 cannot export to Encoder on either XP or Vista
• 4.0.1 update - broke Encoder for good - infamous runtime error on Vista (MBP) after "failed to update encoder" on both Vista and XP
• playback is choppy on both Vista and XP (EEE is at 100% CPU, verify on MBP and emachine)

So overall even with CS4 there is no AVCHD workflow yet... Once FCP would support AVCHD natively the best workflow would be - do cuts in FCP, export XML and do post in CS4 (since CS4 is so much easier), but where we stand today, compared to CS4 FCP lacks

1. Native .m2ts support
2. applying effects without re-rendering
3. Soundbooth like tools for post on sound (Soundtrack doesn't have spectral view)
4. Mastering of .m2ts - don't really need it, but hey
5. 30xT for rendering 720p30 to not very efficient h264... Apple native needs more Mbps then x264 and is slower??? (hmm, my 720p workflow (without post) on Q takes about 5xT, but Q is about 2-3 times faster.... still would make workflow run 10xT-15xT vs. 30xT for Apple, so it is slower at least two times).
6. Freesoft muxer, encoders, etc are lacking on Mac.

FCE lacks compared to FCP

• ability to edit in anything but AIC - this is the only Sequence setting apart from DV
• export XML - no way to export to do post in CS4 without re-encoding to AIC

Pinnacle is a funny one (SONY is even worse, but hey SONY is known for not being capable of writing software) and lacks everything when it comes to editing, so it cannot be considered as an editor, only as non-transcoding mastering app, but it lacks it too - always transcoding to 1440x1080@30 when mastering either AVCHD or Blueray. Recently added to Pinnacle support for QuickTime .mov is useless - "enables exporting a project created in Pinnacle Studio to a QuickTime .mov file using DV, MPEG4 or MJPEG codecs and up to a resolution of 1024x576" (MPEG4 means non-AVC MPEG4 and since resolutions are lower it means transcode anyhow)

New portable

Both Asus and MSI are about to bring 1.6GHz Atom based 10" notebooks paired to 6-cell battery for 5+ hours of battery life. Both Wind and EEE1000 are about the same with Wind being less advanced (80GB 2.5" HD vs. 40GB SSD, G vs N) and hardware details less known. Both would have Intel 945GSE (GMA950) chipset and thus Leo would have QE and depending on wireless would make working Leo laptop.

Wind should come out at $549 any day now and Amazon already list them. With Wind I should make sure that it has 6-cell battery (2.6lbs vs 2.3lbs listed on Amazon) and Wind Linux version is a rip-off not worth $400 due to 3-cell battery and lack of bluetooth.

EEE1000 would come out probably before fall at $650+??? and depending on its price we may see Wind price drop (I want $50 Amazon rebate) and confirmed price drop on EEE900 when EEE901 comes out. EEE900 being a 900MHz Celeron and GMA900 is not worth it even at $400. If EEE901 Linux version (20GB) would ship for bellow $500 (and I mean $450) it may be worth it for N-wireless, but otherwise (say at $500) Wind's extra inch of keyboard for extra $50 makes all the difference, unless MSI totally don't screw up with hardware details.

So apparently Wind it is, but then again in Q3 Intel should start making 2 core 1.6GHz Atoms. If there would be an announcement of Wind or EEE1000 with dual-core for bellow $700 this is an Atom based computer I want and should wait for looking on how Leo would install on Wind...

AVCHD camcoders

Since "the first real' AVCHD camcoder become available sometimes in May 08 I looked at what would it take to edit and play AVCHD. All Windows software that claims to have AVCHD support is laughable. PowerDirector comes as the winner but it is laughable still - it would cut AVCHD only to export it by transcoding (same with Nero only PowerDirector has at least some timeline), SONY product... well SONY just don't know how to write software period. Since on Windows they still don't get it that transcoding HD is unacceptable (and not so much because generation is lost but more because it would take forever - even 720p to 720p transcode could be done on Q6600 at 12fps = 2.5xT at best and with 1080p we are talking 6xT)... Well since they still don't get it, until Adobe release CS4, the only working solution is still Final Cut on Mac.

Final Cut would chop AVCHD in no time, however any effects are not visible until rendering and that takes forever (25-30 min on MBP for 2 minute material), and since effects would require a transcode Premiere beats FCP for material that requires adjustments. The workflow is to chop AVCHD in FCP and do post in Premiere or otherwise on Windows transcoding with meGUI since Premiere would display effects rightaway and has better picture adjustments.

FCP will open MPEG-1 from SONY, but would be unable to decode audio and ffmpeg would error out on video when I try to rewrap.

On Mac the only way to transcode arbitrary video is ffmpeg. (Handbrake does better job but works only with MPEG-2 MPG wrapped). Instructions are here, but apparently quality suck (still need to further verify). ffmpeg could crop and resize with "-croptop 60 -cropbottom 88 -cropleft 252 -cropright 252 -s 720x480 -aspect 720:480" and could cut with -ss -t .It is slower then Handbrake 21/20 vs 40/25 on 480p even with -threads 2 and needs to be installed with "sudo port install ffmpeg +lame +xvid +x264 +a52 +faac +faad +dts"

The following would crop and resize "Globe Trekker" ffmpeg -i /Users/zander/Globe\ Trekker_Extra_South_England.mpg -an -croptop 60 -cropbottom 88 -cropleft 252 -cropright 252 -s 720x480 -vcodec libx264 -deinterlace -b 900k -partitions all -me umh -subq 6 -trellis 1 -refs 1 -coder 0 -me_range 16 -level 30 -bt 900k -maxrate 10M - threads 2 England.mp4 ffmpeg -i /Users/zander/Globe\ Trekker_Extra_South_England.mpg -acodec libfaac -ab 96k -croptop 60 -cropbottom 88 -cropleft 252 -cropright 252 -s 720x480 -vcodec libx264 -deinterlace -pass 1 -b 900k -partitions 0 -me epzs -subq 1 -trellis 0 -refs 1 -coder 0 -me_range 16 -level 30 -bt 900k -maxrate 10M -threads 2 England.mp4 and the following would do 720p->480p(wide) ffmpeg -i /Users/zander/Movies/720p/Christmas\ Party.mov -an -aspect 1280:720 -s 720x480 -vcodec libx264 -pass 1 -b 900k -partitions 0 -me epzs -subq 1 -trellis 0 -refs 1 -coder 0 -me_range 16 -level 30 -bt 900k -maxrate 10M -threads 2 Christmas-orig.mp4 ffmpeg -i /Users/zander/Movies/720p/Christmas\ Party.mov -acodec libfaac -ab 64k -pass 2 -aspect 1280:720 -s 720x480 -vcodec libx264 -b 900k -partitions all -me umh -subq 6 -trellis 1 -refs 1 -coder 0 -me_range 16 -level 30 -bt 900k -maxrate 10M -threads 2 Christmas-orig.mp4

Upgrades (all the way to Nehalem)

My GA-G31MX-S2 is dying. At better times it could be overclocked upto 3.0GHz (333FSB=1333) vs. 2.4GHz (266FSB=1068) stock or 25% boost yielding 15fps vs. 12fps on 2nd pass 1080i->720p. I wouldn't mind buying a new mobo (with native 1333FSB support) to be more stable at 3.0GHz, but they still don't make anything interesting. Cheapo starts at $40 (ESC P945 100Mbps is $49.99 - $10 = $39.99, 1 year warranty) with $65+ buying nForce 630i chipset with GEForce 7100 (EVGA or ECS) and toping out at $100 with same intel G33 (I have G31 but it is a minor upgrade to support more memory 8GB vs 4GB). GEForce 7 serries is still not DirectShow 10 accelerated, but may or may not bring QE support on OSX (video editing). Unfortunately, GMA3100 is different from X3100 (on Apple portables) so QE would not work and thus OSX would require video card. So to get back my "fastest computer ever" it is 50/50 to either get new mobo or RMA GIGABYTE (phone 626-854-9338 select option 4) and getting $50 small factor video card (from X1550 DX 9, 256MB, DVI-I $36 to 8840GS DX10, 256MB, Dual-DVI $55, better get 8400GS) to later reuse in HP.

45nm Q9450 (2.67GHz at 1333MT, 2x6MB L2) at $316 is not worth it unless I get Penryn (T8xxx, T9xxx) processor and it would show more that 25% performance increase. (Should be just 10-15% clock for clock). Since Q6600 is known to run all the way to 3.6GHz (400FSB) clearly I am hitting mobo FSB limitations and that calls for new mobo and pushing it past 3.0 GHz. So if QE would work on HP, new GeForce 7 mobo it is (while GIGABYTE is RMA to decide to keep it or not) like this one (Firewire, DVI, S/PDIF, RAID 5) for $65 after $20 rebate. Running at 3.6GHz would boost Q6600 performance 50% vs. stock or I should see 18fps on 2nd pass 1080i->720p.

Rumors have it that Nehalem is upto 50% faster then Penrym for h264 encodes. Yorkfield at 2.5GHz is 10% faster then Q6600 at 2.4GHz that would make Nehalem based Quad at 2.6+GHz 1.65 times faster then Q6600 or 720p encodes should go 13*1.65=22 fps or not quite real time yet. At 3.6GHz Nehalem Quad (being say 50% faster then Q6600 at the same clock) would make it 27 fps. so even with Nehalem (overclocked) we will not see RT for 1080i->720p, but would be getting closer. Nehalem would use LGA 1366 or LGA 1160 socket, so it would be new mobo etc. again.

According to this forum sticking a Phenom into HP Slimline will not work because they are 95-125W and slimline's mobo max out on 65W. So absolutely no upgrade path for Slimline apart from sticking a video-card with dual-DVI-out and hope it would work (it might not because of the same power issues).

The bottom line is I need to push Q6600 above 3GHz (with QE on OSX or probably GeForce integrated) to get 1080i->720p closer to theoretical 60/18=0.5+1.66=2.17xT at 3.6GHz and that would be the end of "desktop upgrades" story until Nehalem comes with 8 cores.... So may be new mobo and 8840GS/RMA and that's it...

OSX won't work on Toshiba because of Radeon X200 video and won't work on Lenovo because of Intel Pro 3945 wireless... Laptop they sell are not much different - same 667FSB top and still T2xxx (or may be T5xx) Core Duos. So when it comes to portables Penryn MacBook is the best value with $200 rebate for $1099 (X3100, 2.4GHz, 800FSB, 2GB, 160GB, 13.3-inch), unless they start selling 13" Penryn PC for bellow $700 with 802.11N and giggy weighting less then 5lbs.

And still they don't sell 22" monitors with DVI for bellow $200 capable of 1080p... nor 802.11N routers with giggy for less then $110... nor giggy switches for $20 (D-Link is $30 after rebate together with N-access point for $40) makes it $70 vs. $110)

So I am stuck with

• for 1080i->720p on Q: 40/12=0.75+2.5=3.25xT after extra 1xT to transfer from A to Q at 10-25Mbps that sometimes (but rarely) go all the way to 40Mbps (when I am using samba) or on A 13/5=8xT.
• for 1080i->480p on Q: 75-80/48-54=0.4+0.6-0.37+0.55=0.92 to 1xT after 1xT to transfer vs. 40/21=0.75+1.42=2.17xT on A or 40/25=0.75+1.20=1.95xT on MacBook
• for DVD (with deint) 140/63=0.21+0.46=0.67xT on Q.

and above are the best times...

BD/AVCHD vs. AppleTV and NAS.

BD is crap as movie storage. It needs 1 sec. GOP and that implies higher bitrates, bigger files, etc. Inefficient. Plus ffdshow craps out playing files with GOP with less then 60-90 frames. AVCHD allows longer GOP, but needs interlaced SD with LPCM audio (HD needs AC3 and 720 is all progressive and 1080 could only be progressive at 24p). So out of the box I could only try 720p24 and 720p60 and that means that for both BD and AVCHD I would have to frame-reduce my 720p30 Aiptek mp4, i.e. would have to transcode which is bad.

For 720p I would have to frame-reduce 1080i downconverts (and probably 720p60 originals) to 23.976 fps. 60i (for 1080 and SD) means 29.97 frames which could be 24p pull-downed, 29.97 interlaced and 29.97 progressive with pulldown flag set on h.264 stream. Apparently x264 could set pulldown flag with "--pulldown 23" (might not be fully supported in hbcli) and all streams need to be encoded with --aud and --nal-hdr (the later again might not be supported in hbcli), but overall it looks like AVCHD/BD is still being in its early days after a year of hacking... Additionally, QT cannot play 60i streams but could play 23.976 with pull-down disregard pull-down flag. Thus overall AVCHD/BD is good only for 720p24 and in order to play on both BD standalone and Apple TV everything need to be frame-reduced which is bad. (Theoretically, I could store 720p60, but at 4+xT even on Q6600 and 5Mbps bitrates it ain't worth it).

So there you have it - there is absolutely no reason to encode BD compliant streams and thus BD brings nothing new to viewing HD at home and solution remains the same - encode for storage on NAS and view thru networked TV appliance. Welcome back Apple TV.

Before completely giving up on BD and AVCHD however still try couple things:

1. Try Nero Vision 8 to see if it would try to transcode my 720p24. If it would try another 720p24 encoded with --detelecine -r 23.976 -x pulldown=23:aud=1:nal-hdr=1
2. Try Nero with 720x480@24p with pull-down and see if motion jerks because of the frame-reduction on say Seige or 60_seconds.
3. Try using seek and frames with HB.

P.S. Both BD and AVCHD wrap streams into MPEG-2 transport, mpg program stream cannot store PCM, thus VHS MPEG-2 encodes have .mp2 audio and for SD I would have to unencode .aac from .mp4 while for HD AC3 need to be copied over from the original and AC3 cannot be stored in .mp4...

Vista and XP on the same partition

Vista and XP do not live happily on the same partition because they share Program Files directory. They could be installed on the same partition but WMP 11 would work in only one of them. The way to install is

1. Install Tiny Vista (no Media Center) first
2. Install XP. This would blow Vista bootloader and overwrite WMP in Program Files breaking it for Vista.
3. Using Vista DVD restore Vista bootloader. This would blow Vista validation thus Vista needs to be re-validated (by copying sources to HDD and running \sources\$OEM$\$$\INSTALL\startVista1.cmd (startVista2.cmd)).
4. Install Easy BCD and add XP boot.

Tiny Vista takes 4GB after update (2.82GB after install) vs. 10.6GB for plain Vista XP SP3 takes 1.75GB after update (1.3GB after install) vs. 2.76GB for plain SP2. Program Files would take 1.78GB min. Thus double install with minimal Office etc. would require at least 9GB (or 8GB for Tiny Vista alone).

Even with most everything stripped out Vista is hungry for RAM requiring ~300MB vs. 650MB for standard Vista vs. 160-180MB for XP.

Also, all Vistas but Ultimate do not allow Remote Desktop and log out whoever is logged onto console. RDS could be turned on (and log out fixed) with termsrv.zip on Business, Premium and on Ultimate it would fix log out problem.

Q6600 vs. A5000+

Encode	1080i->480p	720p->480p	720p->720p30	1080i(s)->720p
A menc single	22.15=1.35xT	55->1.1xT		~5fps=6xT
Q menc single	35fps=1.15xT*	77->0.7+xT*
A menc 2-pass	31/24=2.2xT	68/50=2.07xT
Q menc 2-pass	50/49=1.2xT*
Duo2 menc	33/25=2.1xT			4fps?=7xT

* denotes not fully loaded CPU (that could be fixed by adding thread=8)

Encode	1080i->480p	720p->480p	720p->720p60	1080i(s)->720p
A HB 2-pass	28/17.5=2.77xT	41.5/17.5=2.2xT		13/5=8xT
Q HB 2-pass	75/45=RT	80/27=1.19xT	45/25=1.85xT**	27/15=3xT
A HB single	11-15fps=2-3xT	15fps=2.0xT	6.11fps=9.8xT	4.7fps=6.3xT
Q HB single		50fps=0.6xT	17.93fps=3.3xT	13fps=2.3xT
Duo2 single	16fps=1.8xT	40fps=1.5xT*		5.94fps=5xT

* denotes 60p vs. 30p for the rest ** denotes 30p vs. 60p for the rest

As can be seen from the table it is 2xT HD on Q6600 and 2x SD on A5000+ and there is no need to distribute since for HD adding A5000+ to help out Q6600 would at best give just 25% gain, and for SD Q6600 would be faster RT anyway and distributing SD would greatly depend on how well new avisynth would support multithreading since Q is not fully loaded with SD with either mencoder or x264.

Commercial detection is a mess. comskip trips, then mencoder trips even more (cannot cut the first right sometimes and almost always ignores the rest of .edl. mplayer however plays those .edl just fine)... so anyway you look at it cutting commercials is manual and thus you have to convert .vproj to .mpg and with .mpg HB is a king (marginally faster at least on Q, fully load CPU and does DPL) unless mencoder could do better on framerate reduction. But HB cannot be run from DRVMSToolbox... So overall DRVMSToolbox way is a miss and I am better off having my own process watching for .vproj appearing somewhere to kick off transcoding workflow.

The biggest problem however is that 720p60 material cannot be framereduced without jerkiness (neither with old HB, and especially not with mencoder) and that means twice number of frames to encode which implies double the work and double the bandwidth.

HB would reduce frames (with slight loss of temporal-quality). The artifacts I am getting are due to keyframes every 30 frames or 1 sec that ffmpeg cannot handle (BD however does require 30frames GOP so I am screwed... What about AVCHD???). Considering that having a keyframe every 30 frames greatly reduce quality I should get back to encoding at 150-300 keyframe intervals and with that SD could be fit into 900kbps at 29.97fps and 720p30 into 2840kbps giving 1Mbps for SD (at 720x480) and 3Mbps for 720p. Still single pass encodes at those high rates suck and the only way to go is to do 2 pass at the same bitrates (going lower on 2 pass creates artifacts and defeat the purpose). HB --denoise (hq3dn) doesn't load CPU 100% on either A or Q, hence once again gotta do 2-pass without denoising, and that means 3xT for SD on A and 3xT for 720p on Q (from local). Then I have networking bottleneck where I cannot get A to pump more that 7Mbps... so copying files off A to do 2-pass encode is not an option until I spend $50 on a new router which I don't really need...

Anyway you slice it, it is too slow... from 3xT for SD on A (or RT on Q if I could pump 30+Mbps over the WiFi) to 3xT for 720p30 on Q from 1080i (at 8xT for 2pass and 6.5xT for single A is not up to the task) all the way to almost 10xT for 720p-720p60 on all dual cores (both 2.6GHz AMD and 2.33GHz Duo2). And it takes quite a bit of bandwidth 1Mbps for SD, 3Mbps for 720p30, over 4Mbps for 720p60 (2.5Mbps has artifacts but is the lowest for 720p60 and similar to 1.5Mbps for 720p30) and around 5Mbps for 1080p30.

720p60 encodes need similar to 1080p60 bandwidth and about as fast at 5xT-6xT (5fps) and is way to much even for 4 core Q6600.