My Rambling Thoughts

A careless year

Somehow, I was more careless than usual this year, if measured by preventable mistakes that caused me time, money, stress and data loss.

Credit card late charge and interest. Happened twice for OCBC credit card. OCBC is very strict on interest charge, and the amount involved was so significant that, after the second time, I opted for auto-deduction from my bank account to prevent future mistakes. I forgot to pay two of my other credit cards at least once each too. Not a good record.

Shared car repair. I made the fatal mistake of not reporting the damage when I took the car. Guess what, the previous user played dumb, so the repair costs fell on me. Cost of repair: S$508.25. Price to find out a person's integrity and honesty: priceless.

Non-payment of ERP charge. Thrice. On my YBR125 bike, after a non-related repair, the IU no longer powers unless the bike is on. I like to insert the cardcard into the IU first. The problem is, the chance of error is quite high, say, 20%. This happened twice, even though the second time I was quite sure I have checked it. The third time, I was leaving the CBD area and forgot some exit routes have ERP gantry — and I purposely chose that route too!

Ferry tickets expired. We could not find time to go despite a 6-month validity period, until the last two weekends. And then we could not make it after all. I forgot why — I think I fell sick?

S3 phone dead. Somehow, it overheated during charging — the phone was hot — and could not be powered up anymore. The battery itself is still working (tested on another S3). I was not so upset with the loss of the phone (it was on its last legs anyway) than the loss of fuel mileage data. I had always put off copying them out, since it had to be done manually and it was very low priority. All other data had been copied out.

Tab 10 screen crack. To my surprise, it did not survive a sub-1m drop and the glass cracked at an edge. From there, it grew longer and longer until it interfered with the touch operation. Once that happened, it was game over. I knew this was going to happen — even the most careful toddler drops things from time to time — but I just could not find a protective cover for it. I was a bit upset because I was planning to use it. I had planned to get a smaller and lighter tablet for 吉吉. Just not in time.

Notebook burnt. Due to short-circuit. You know it's bad when smoke comes out of the notebook and the circuit breaker trips. I was not upset with the loss of the notebook — it was long past retirement and there were no data on it — but Missus and I argued over it because she had warned me this would happen and I didn't pay heed.

Missus

Missus had a normal year, except for one unlucky day, and it was a day she especially took leave to rest.

Lost contact lens. She was dreaming when she removed her contacts. It was washed away in the basin. Replacement: $380.

Fit glasses. As a result, she had to fit a new set of glasses (frame only) and that costed $160.

Lost Samsung Gear. The watch fell off while she was carrying 吉吉. According to her, the watch has a tendency to fall off. She was sure I would chide her, but I did not say a single word. This has not been a good year for our devices.

Delivered in the nick of time

I usually deliver parcels to my workplace, but this time, Amazon estimated it to be delivered during my company shutdown period, so I delivered it to my home.

Well, Amazon must have been trying heroically to deliver all parcels before Christmas, because it split my order into two parcels and adjusted the dates to three days before Christmas.

Should I re-route them to my workplace? There is still a risk they will be delayed.

And they picked Amarex. Based on online feedback, they have the lowest reliability by far among the delivery companies.

Amarex delivered the first parcel to my home in the afternoon, but no one was home. The courier did not call me and left me a note. They don't do same-day re-delivery either. (I think.)

I called the number and no one picked up even after one hour of being on hold. I tried again at night, and found their Customer Service was not 24/7, only from 9am to 6pm.

The second day, I called the number again. Network busy. I tried for tens of minutes before I got through, then I was put on hold. After 45 minutes, I finally reached an operator.

The parcel was in their Changi warehouse. I could either collect it myself, or I could schedule another delivery, even to another location, on the following day. The second parcel was being delivered that day, but it was already out, so it could not be re-routed.

I knew the same thing was going to happen, so I left a note on my door asking the courier to call me. He did, and I asked him to leave the parcel at the doorstep.

Interestingly, I have a third parcel — from Tao Bao — that was delivered on that same day. The courier saw the same note and called me. (I don't know if he would call me otherwise, but I suspect yes.) In this case, he was not willing to leave the parcel behind, even though he already saw one there, but he would re-deliver at night. (Amarex does day-time delivery only, I believe.)

So, what's the takeaway?

Don't miss the initial delivery, or it will be a hassle. Track the parcel and give instructions prior to local delivery. Leave a note on the door. I took it for granted the courier will call if no one was home. Also, not all couriers will do same-day re-delivery or deliver at night.

Amarex is low-cost, I don't expect too much. :lol:

I would say my experience is not too bad, given the peak Christmas period. At least they delivered on time. The Tao Bao parcel was late by 5 days!

Board games 2015

Only three new games made the cut this year. The criteria has become more stringent due to my existing game base. There were actually a few more, but they were unavailable in one way or another (for example, free Amazon shipping).

7 Wonders (2010) [US$31.99]

7 Wonders is a hot-selling award winning game. On BGG, it is ranked first in the family game ranking and #20 overall. In comparison, the classic gateway game Ticket-to-Ride ranks #10 and #89 respectively.

I gave it a pass last year, but it made it this year — it is on my second-tier list. :-P

The key mechanics are simultaneous card drafting, variable card powers and set collection.

This is the next gateway step after TtR.

Fury of the Dracula 3rd ed (2015) [US$43.95]

Fury of the Dracula (3rd ed) is one of many "10th anniversary" remakes. It is a remake of the 2nd edition (2005), although it first came out in 1987.

Mechanics: deduction, co-operation, variable power, hidden movement.

Unlike my other "Euro" games, this is an Ameri-trash game. The gameplay can be somewhat unbalanced, but it is oozing with theme — if you are a vampire person.

This time, they corrected the most glaring problems with the 2nd edition, so it should play even better — especially for Dracula!

Due to the long play length (potentially 2+ hours), I doubt I ever have a chance to see this in play, just like my 2nd edition copy. I need to introduce new win/loss criteria to fix the play time at one hour.

I can foresee myself buying the 4th edition in 2025! :lol:

Mission: Red Planet 2nd ed (2015) [US$36.58]

Mission: Red Planet (2nd ed) is sort-of a 10th anniversary remake. I'm surprised I have never heard of it. Just as well. The 2nd edition has better art and bits, and slightly better rules.

The key mechanics are: simultaneous action selection, hand management and area control.

My first impression is, the role selection feels like Citadels, and no surprise, it has the same designer. It is done simultaneously, so there is no down-time. Much of the time in Citadels is spent waiting for others to select their roles — for example, 80% of the time in a 5-player game.

I think of this as Citadels done right. I also buy into the "conquer-the-Mars" theme, so that is another plus point. There is something about fascinating about seeing rockets take off, even cardboard ones. :-D

Snatoms


The Magnetic Molecular Modeling Kit

$418.8k from 5,585 backers, with 3 days to go.

I don't really need this, but I think it is a good model and is worth supporting.

Kickstarter is a good idea, although I think backers generally get a worse deal compared to regular buyers — they pay full retail price.

Have a good idea but no seed money? Kickstart it!

Update: 6,581 backers pledged US$519,851.

Tomica Pixar Cars: core characters assembled!

For the longest time, the Tomica Pixar Cars were sold exclusively in Japan. A local hobbyist shop imported them, but at a premium price of S$15 apiece. After a long hesitation, I finally got the core characters. But the shop ran out of Doc Hudson, so I did not have him.

A couple of months ago, I found that the Cars characters were available at regular retailers — and cheaper! They are sold at S$10.95 at Metro, no discount available. I don't know if other retailers offer any discounts or not.

So I got my C-06 finally.

Flawed game even after ten years

Manila (2005) is a gateway game with beautiful art. It went out of print a few years ago, but it was reprinted end of last year — just in time for its 10th anniversary — with slightly cheaper components. It is no classic, though. I want to like it, but the game is too flawed. Its problems?

Die rolls are too chancy, the Harbour Master (HM) is too strong, and never play with 5 players.

Die rolls are random, but there is no averaging-out effect because there are only 3 turns (9 rolls) per round.

The Harbour Master gets to buy shares, chooses goods to load on the boats, chooses their starting positions and starts first. Woe to the player who sits to the left of him (last to go). The HM position is bidded in an auction, but if one player gets an early lead and gains more money relative to the others, he can out-bid the others.

The game is sort-of a worker-placement game, so you are at a disadvantage if you get to place your workers last. And there are just enough spots for 4 players.

Last of all, for such a light game, it takes too long to explain and play! It can be as short as 5 rounds, or as long as 17 rounds! Each round has 3 turns, and each turn has 3 phases (the first turn has 4 more).

There is a missed opportunity to improve the rules for the 10th anniversary reprint. I won't play it again without changing some of the rules. For example,

  • the HM can buy up to two shares, the others one
  • the HM can load goods on two boats only
  • the second-highest bidder pays his bid and load goods on the last boat
  • players can bid for the turn to go clockwise (change of wind direction)
  • the Pilot (move boats) is a free turn
  • the share price starts at $5 (at least two rounds shorter)

As mentioned, the HM is way over-powerful. He is the only one who can buy shares. If a player is not able to become the HM, he is at a distinct disadvantage.

Another HM power is choosing the goods to load. We now let him load two goods only.

One side-effect of bidding for the HM is that the second-highest bidder can cause the winner to overpay, while he gets to keep his cash. No more. He has to pay. As compensation, he gets to load the last boat.

Players can bid for the turn to go clockwise to counter the HM-effect. Otherwise the player on the left of the HM always goes last.

The Pilot is often a poor choice because you cannot place your worker on an income-generating spot and thus effectively losing one turn.

The game is often too long and repetitive. By starting the share price at $5, we cut short a few rounds and shave off 10-20 minutes.

The game that failed to put up a good show

Colosseum was released by Days of Wonder in 2007. This was their next A-list game — the art and components were gorgeous — but it failed to sell. DoW even had to give it away a few years later to clear their stock. It went OOP around 2011.

What's wrong with the game?

Well, it is repetitive, has limited paths to victory, little strategy and it takes too long.

The player who puts out the show that attracts the most spectators win. You do this over five shows. For each show, you improve your arena, upgrade your event, recruit more performers (via auctioning), perhaps even star performers, and try to lure the Emperor and the nobles into your arena (by die roll and Emperor Medals).

As you can see, you are doing the same thing five times.

Your starting events and event asset tokens (performers and props) often determine your event upgrade path.

There is also a certain sequence to get the bonus powers: get Emperor Loge (use two die), get as many Emperor Medals as possible (+3 VP or move a noble 1-3 spaces or 6 coins), expand arena, buy season tickets (+5 VP).

The game takes around 1.5 hours to play. Every round is 90% similar. And that is the real problem. Three rounds is the most people can take before their minds get numbed from the repetition.

But boy, is the game gorgeous. :love:

Games that just missed the mark

Another year, another round of board games. To my surprise, after a big catch-up last year, quite a number made my list this year. And I have not even heard of many of them last year! (This is why sometimes I read or watch other people's top 100 games — to be exposed to new games.)

But here, I will talk about some that did not make my list, at least this year.

Dice Town (2009)

In Dice Town, you roll five die and that determines which of the seven locations you will go. Each location offers its own reward.

I like this game. My main beef? It takes too long. This is a light game, but it takes more than an hour to play! Why does it take so long?

The problem is that the die are resolved one at a time. First, you roll all five, keep one, then you roll four, keep one, and so on. (You can pay the Bank to keep more if you like your die roll.) This adds strategic element to the game. You can see what other players are choosing and react accordingly.

Then, each of the seven locations is resolved sequentially: the player with the most die for that location gets the reward. For the second last location (the Town Hall), you have to compete for the best poker hand. That takes more time.

If I play this game, I would use a variant. First, every player would get ten colored die (of their own color) and roll them. They may re-roll by paying, but they must keep at least three die every time. The die are then placed at their locations, with five at Town Hall. This speeds up the die rolling phase and allows the locations to be resolved quickly.

Dice Town has one expansion. It adds a few new mechanics. The biggest change is that the second highest player at a location gets the secondary reward. It sounds great, but it slows down the resolving phase and it reduces the chance of players visiting Doc Badluck, which is a very powerful location!

Kingsburg (2007)

I have poor affinity for worker-placement games, so you will never find Caylus (2005), Pillars of the Earth (2006) or Agricola (2007) in my collection. Which is why I'm surprised I rather like Kingsburg, which is a worker-placement game.

I suspect a lot of it has to do with its die mechanic: roll three die and choose any seven combinations from it to influence one to three of the eighteen advisors in the court (numbered 1 to 18).

The problem is that the board can get pretty crowded with 4 or 5 players. There should be more mechanisms to put two die per location.

This game is still available, but is starting to get hard to find! But no worry, there is some news about a reprint early next year.

Small World (2009)

Small World is Euro-style Risk. There are three mechanics I like: race+power combo, pay-to-skip and go-into-decline.

There are 14 races and 20 power badges, so that means 280 combo.

The races/power are put in a row. You have to pick the first one. If not, you put one VP each on the races you skip as you go down the line. This has the effect of making them more attractive.

After you have "maxed" out your existing race, you pick a new race and let the existing one "go into decline". They still retain their existing territories — and you still earn points from them — but they are sitting ducks for new invaders. (Hey, it's a small world.)

So, what's wrong?

I would say the main problem is that all the games feel "samey". How can that be, with 280 combos? Actually, there are only 9 effective combos, because in the name of balance, the cards are either poor, okay or good, so 3 x 3.

And the gameplay is very tactical and repetitive: go for the weakest spots and spots you have the best defense.

Small World has a ton of expansions. That is a sign of its popularity. Plus, Days of Wonder released a pimped-out Small World Designer Edition (2015) at US$320 (Kickstarter price; leftover were sold at US$450, and snapped up in minutes). Only a rare handful of games get this treatment.

Are these dinnerware worth it?

Set of 3Reg $5pts +10pts +
27cm dinner plate$49.50$24.95$16.95
21.6cm luncheon plate$45.00$23.95$15.95
550ml cereal bowl$46.50$24.95$15.95
900ml noodle bowl$58.50$31.95$19.95

There are 4 months to accumulate the points. Each point requires $30, so 5 points means a minimum of $150.

Having said that, these can be worth it if you can accumulate 10 points anyway. To do that, you have to strategize your spending so that they are in multiples of $30.

Math combo fail!

A shop has an offer of 20% off for $20 and above. I have $15.70 worth of stuff. I can buy $4.30 more of stuff and pay only 30 cents more for them!

The question is, can I?

The items are $2.00, $2.30, $2.90 and $3.90.

The problem is the combination. While $2 + $2.30 works, I don't have anything I want at the $2.30 price point. $2 + $2 is too little, and the $2 and $2.90 items overlap (iirc). $3.90 is short by itself, but is too much if combined.

I gave up after 10 minutes of searching.

x265 1.8 presets test

x265 1.8, CRF 22. It turns out we can "drop" it directly into HandBrake 0.10.2.

PresetFPSQPkbps
ultrafast12.34650525.636575.41
superfast7.67444825.576881.28
veryfast6.21370623.888501.43
faster6.51545123.718704.24
fast6.50957825.794722.86
medium4.41635425.874864.72
slow1.30407925.805220.39
slower0.39606225.895093.71
veryslow0.31465325.744881.17
placebo0.23297025.714782.56

Trading negligible quality for lower bitrate and 30-50% increase in speed in most cases.

Distributing the tests over 6 pools

PoolCPUPresets
00placebo
1slow
2ultrafast, veryfast, medium
31superfast, faster, fast
4veryslow
5slower

Each pool has 3 real cores and 3 HT cores.

x265 1.5 presets test

x265 1.5, CRF 22.

PresetFPSQPkbps
ultrafast9.12708025.866164.25
superfast7.21205925.576958.28
veryfast5.67086923.049026.12
faster5.77016823.079277.26
fast5.46333225.085371.28
medium3.92086825.155121.40
slow1.17524025.135491.04
slower0.30899825.315247.88
veryslow0.22481625.325147.98
placebo0.15337825.325124.20

Presets affect speed, but there is no clear effect on quality and size.

This is the same clip that I used for x264 testing. x265 uses around ~5 Mbps, vs ~17 Mbps for x264. Is HEVC really that efficient? Sadly, no. This clip is grainy and x265 removes it partially.

As expected, ultrafast and superfast requires more bitrate, but not as much as their counterparts in x264. However, veryfast and faster requires much higher bitrate. Is it because they try to preserve the grain?

There is a huge speed hit from medium to slow, and then to slower. They better be worth it!

Consecutive B-frames:

ultrafast   0.1%  0.0%  0.0%  0.1% 99.9%
superfast  40.6%  1.9%  2.4%  1.4% 53.8%
veryfast   40.6%  1.9%  2.4%  1.4% 53.8%
faster     40.6%  1.9%  2.4%  1.4% 53.8%
fast       40.6%  1.9%  2.4%  1.4% 53.8%
medium      8.8%  4.4% 12.6% 39.4% 34.9%
slow        8.6%  4.1% 13.2% 40.5% 33.5%
slower      9.8%  2.7%  2.8%  9.6% 11.7% 45.2% 10.7% 3.9% 3.6%
veryslow    9.9%  2.9%  2.6% 10.1% 11.4% 43.1% 10.7% 4.7% 4.6%
placebo     9.8%  2.5%  2.7%  9.8% 11.7% 43.9% 11.5% 4.1% 4.0%

x265 relies less on reference P frames and heavily on B-frames, even for faster presets.

Is x265 usable today? Yes, but with the following caveats:

  • encoding is very slow
  • there are few hardware decoders (but this is improving)
  • it smoothens the video, as if it has a built-in weak low-pass filter
  • it shines in low-bitrate and 4K resolution

When using x265, it is worth keeping in mind that the default parameters are tuned for 4K encoding, so they may not be optimal for lower resolutions.

What is mildly funny to me is that x265 has a longstanding history of smoothing the video and the people involved as well as the advocates just refuse to admit it! :lol:

Based on postings on doom9 forum, it still happens today for x265 1.8.

From a distance

I first heard the song From a distance during the first Gulf War, when the world was united against the evil dictatorship of Saddam Hussein and the forces of righteous were going to crush him and we would enter a renewed golden era of Pax Americana for centuries to come.

Of course, it did not turn out that way.

At that time, I could not figure why George Bush (Senior) did not finish the job. Ten years later, I prayed George W. Bush would not finish it, but he did. And we knew what happened after that. Without an iron-fist strongman, the region disintegrated into chaos.

From a distance
We all have enough
And no one is in need
And there are no guns, no bombs and no disease
No hungry mouths to feed

From a distance
We are instruments
Marching in a common band
Playing songs of hope
Playing songs of peace
They are the songs of every man

I cannot help but get misty-eyed when I listen to this song. If not the above, then it is this:

From a distance
You look like my friend
Even though we are at war

From a distance
I just cannot comprehend
What all this fightings for

Back then, it was because I was overwhelmed with joy that the new peaceful era was within grasp. Now, it is because the world is a much more dangerous place. Instead of making it better, we have made it worse. Where did we go wrong?

I especially like the first part of the song:

From a distance
The world looks blue and green
And the snow-capped mountains white

From a distance
The ocean meets the stream
And the eagle takes to flight

That is our beautiful planet Earth. We must take care of it.

The missing sides from 2.35:1 to 16:9

By chance, I acquired two different sources of Bill & Ted's Excellent Adventure (1989).

I did not notice the difference at first, then I compared them top-n-bottom to compare the quality. One is in 16:9, formatted for widescreen TV. The other is bona fide 2.35:1. Wow.

But truth be told, the director will always make sure the scene fits within the 16:9 area — and likely the 4:3 area too, since this show predates the widescreen era. The area outside is merely window-dressing.

I used to think that 2.35:1 gives the most complete image, since it is the widest. However, the reverse is usually true. Shows are shot wider, then matted to create 2.35:1.

Tweaked x264 presets

I found that some of x264 presets were "sub-optimal" when I tested them.

Original:

PresetFPSkbpsMB
slow5.45374217788.92667.15
slower2.88157618731.36702.49
veryslow1.71429317489.24655.91
placebo0.52977417696.63663.69

slow and slower are hampered by their low B-frames. veryslow and placebo have high reference frames and B-frames that are not used effectively (except in animation). The tweak is simple, limit reference frames to 4 and b-frames to 6:

PresetFPSkbpsMB
slow6.42411717484.95655.75
slower3.72791218314.58686.86
veryslow3.011477 17493.34656.06
placebo1.20060217708.58664.13

#ref for P frames, starting with 1:

slow     45.7% 10.1% 26.6% 15.7%  1.9%  0.0%
slower   45.8% 10.3% 25.7% 16.4%  1.8%  0.0%
veryslow 45.8% 10.1% 25.9% 16.4%  1.9%  0.0%
placebo  48.4%  9.1% 25.0% 15.4%  2.2%  0.0%

Consecutive B-frames, starting with 0:

slow      2.5%  1.8%  4.0% 10.1% 14.9% 59.7%  7.1%
slower    2.4%  1.6%  2.6%  9.4% 14.9% 61.5%  7.7%
veryslow  2.3%  1.3%  2.8%  9.6% 14.6% 61.4%  8.0%
placebo   2.7%  1.7%  3.3%  9.0% 14.9% 60.3%  8.1%

As expected, slow and slower benefit from more B-frames. As a bonus, they are also slightly faster.

For veryslow and placebo, they are much faster, and only negligibly bigger.

Big contributor

According to a post I read online, trellis is one of the main reason for the size difference.

TrellisPresets
0ultrafast, superfast, veryfast
1faster, fast, medium, slow
2slower, veryslow, placebo

It does explain why there is a big gap from veryfast to faster. It should also mean veryfast has worse quality. However, it should be very slight because I cannot see it on casual viewing.

This table also suggests the best preset to use for each trellis setting:

  • veryfast: for speed and size, at some expense of quality
  • tweaked slow: optimal trellis 1 setting
  • tweaked veryslow: optimal trellis 2 setting

Cosmos and math

I saw browsing through the Astronomy section in the library when I saw Carl Sagan's Cosmos. For some reason, I decided to skip it because I was fixated on borrowing just one book and I thought there was no hurry as Astronomy books are shelf warmers. :lol:

Imagine my surprise on my next visit when I found the book absent from the shelf. The real shock was when I checked the online catalogue and found only 6 out of 30 books are available! Wow, it is popular.

A couple of weeks later, I borrowed two Math books after watching the YouTube channel Numberphile. Now, they were readily available: Alex Through The Looking-Glass, 14/20, and The Simpsons and Their Mathematical Secrets, 36/44.

It so happened Cosmos was available too. This time, I grabbed it.

I set a personal record for reading the Math books in one day each. Are they really that engrossing? :lol:

Value staycation

Room$357
Dinner$8.20 + $54.14 + $28.05
Parking$10 * 3
Total$477.39

This is one of the cheapest room rate with a window view in Singapore, but it is still not cheap...

x264 stats for nerds

Live action

IPB
Preset#QPSz#QPSz#QPSz
ultrafast3018.47271,022717021.49144,451---
superfast9415.53255,998379418.02173,144331220.8096,528
veryfast14618.08186,781502321.6793,996203123.0756,674
faster14118.04190,120501721.62101,850204223.0763,250
fast14218.30194,627500721.93104,547205122.9867,262
medium14218.30194,632500721.95103,880205123.2865,439
slow6217.49206,300218521.43118,549495322.8679,804
slower6217.53203,526220421.52122,099493422.9885,266
veryslow4518.81200,009138722.23127,091576824.3581,583
placebo3918.69218,712141622.29127,552574524.3782,591

#ref for P frames, starting with 1:

ultrafast –
superfast –
veryfast  –
faster    72.6% 27.4%
fast      69.6% 30.4%
medium    53.2% 13.7% 21.0% 11.7%  0.4%
slow      39.1%  8.6% 23.4% 12.9% 11.6%  4.2%  0.3%
slower    34.8%  7.8% 19.0% 10.6%  8.9%  7.2%  6.1%  3.8%  1.6%  0.1%
veryslow  35.9%  7.6% 16.9%  9.7%  6.1%  4.5%  4.0%  2.6%  2.3%  1.9%  1.7%  1.6%  1.4%  1.3%  1.3%  1.0%
placebo   38.6%  7.1% 17.0%  9.5%  5.9%  4.4%  3.8%  2.4%  2.1%  1.8%  1.6%  1.4%  1.3%  1.2%  1.2%  0.8%

Consecutive B-frames, starting with 0:

ultrafast –
superfast 21.2% 49.8%  8.4% 20.7%
veryfast  44.4% 53.0%  2.5%  0.0%
faster    44.2% 53.2%  2.6%  0.0%
fast      43.9% 53.3%  2.8%  0.0%
medium    43.9% 53.3%  2.8%  0.0%
slow       2.7%  4.4% 36.8% 56.1%
slower     2.8%  4.6% 38.2% 54.3%
veryslow   2.2%  1.0%  2.0%  8.3% 13.2% 58.0% 12.2%  1.6%  1.5%
placebo    2.5%  1.8%  2.5%  8.2% 11.9% 56.2% 11.0%  1.1%  2.9%  0.8%  0.5%  0.3%  0.2%  0.2%  0.0%  0.0%  0.0%

Anime

IPB
Preset#QPSz#QPSz#QPSz
ultrafast3017.87253,758716120.7072,974---
superfast12416.44270,094386818.59118,169319918.9828,264
veryfast25419.55191,505444521.5253,234249221.447,460
faster24219.41187,417443221.4655,466251721.9510,539
fast23719.41208,754441921.4660,045253521.8916,031
medium23719.40208,960441921.4459,414253522.2711,101
slow9317.78232,767229021.2173,580480822.2424,691
slower9317.78228,830229221.1773,107480621.7424,193
veryslow7418.01230,037177221.8675,884534522.7923,083
placebo5917.81234,708202222.0575,766511022.7122,526

#ref for P frames, starting with 1:

ultrafast –
superfast –
veryfast  –
faster    73.0% 27.0%
fast      72.9% 27.1%
medium    61.3% 13.8% 16.6%  8.1%  0.2%
slow      48.2% 11.9% 20.2%  8.8%  8.2%  2.6%  0.0%
slower    44.1% 11.2% 17.6%  7.1%  6.3%  4.8%  4.6%  3.0%  1.2%  0.0%
veryslow  39.1%  9.5% 16.2%  6.6%  5.6%  3.9%  3.7%  2.1%  2.3%  1.9%  2.0%  1.8%  1.6%  1.3%  1.4%  1.0%
placebo   43.9% 10.0% 15.5%  6.0%  5.1%  3.6%  3.3%  1.8%  1.9%  1.5%  1.5%  1.3%  1.3%  1.2%  1.1%  0.9%

Consecutive B-frames, starting with 0:

ultrafast –
superfast 32.1% 23.4%  7.2% 37.3%
veryfast  43.6% 25.1% 16.5% 14.8%
faster    43.0% 25.5% 16.6% 15.0%
fast      42.6% 25.7% 16.9% 14.9%
medium    42.6% 25.7% 16.9% 14.9%
slow       4.5% 11.0% 24.5% 60.0%
slower     4.5% 11.0% 24.9% 59.6%
veryslow   3.7%  6.4% 10.8% 21.6% 14.5% 35.1%  5.5%  1.3%  1.1%
placebo    6.3%  9.6% 12.9% 17.9% 14.4% 31.0%  5.3%  0.6%  0.8%  0.6%  0.0%  0.3%  0.0%  0.0%  0.4%  0.0%  0.0%

Takeaway

It is interesting to see the distribution of I, P and B frames and their sizes. You can re-construct the "strategy" behind each presets. Unfortunately, they do not always work out, that is why the slower presets are not significantly smaller.

The #ref P-frames and consecutive B-frames distributions provide support that high values are seldom used.

Unexpected x264 presets

x264 (core 142), CRF 20.

Live action:

PresetSize (MB)Size (%)FPSFPS (x)
ultrafast1,043.9153%73.897.530
superfast1,000.7147%45.844.671
veryfast614.690.1%31.083.167
faster667.097.8%20.422.081
fast689.1101.0%12.011.224
medium682.0100%9.8131.000
slow667.197.8%5.4540.556
slower702.5103.0%2.8820.294
veryslow655.996.2%1.7140.175
placebo663.797.3%0.52980.054

Anime:

PresetSize (MB)Size (%)FPSFPS (x)
ultrafast530.2156%72.025.865
superfast581.0171%48.783.972
veryfast303.989.3%35.472.888
faster317.893.4%24.191.970
fast355.5104.5%15.081.223
medium340.3100%12.281.000
slow308.990.8%7.6120.620
slower305.289.7%4.2860.349
veryslow274.980.8%2.4210.197
placebo282.282.9%0.66170.054

I used to think that the presets trade off quality for size and speed. Not so, they only trade size for speed. Quality is still largely determined by CRF.

The video from ultrafast looks "just as good" as the rest, just twice as large. veryfast is actually a good preset. It is both smaller and faster than medium!

I still think the slower presets have slightly better quality, but it looks like they are very marginal — at least at first glance.

The next surprise is that placebo is bigger than veryslow, despite being 3x slower.

On the trail of an advanced technology

If there are aliens at KIC-8462852 and they can build a Dyson sphere, I hope they don't detect us. Really.

Another round of DBSS complaints

Pasir Ris 4-room DBSS flat 'too small' for family of 7

Date: 15 Oct 2015. Source: AsiaOne.

A WOMAN who bought a four-room Design, Build and Sell Scheme (DBSS) flat at Pasir Ris said she was appalled to find the completed unit too small to accommodate her household of seven, reported Shin Min Daily News yesterday.

"I planned to occupy the master bedroom with my mother, give one room to my brother and his wife, and the guest room to my two nephews and the maid," Tracy Tan, 36, told the Chinese evening daily at the 86 sq m unit in the Pasir Ris One development which she is about to move into.

86 sqm is only slightly smaller than the standard 90 sqm HDB 4-room flat, so I'm not surprised if Ms Tan assumed they would more or less fit, if they were "upgrading" from one. That is, until you scrutinize the floor plan:


Pasir Ris One DBSS 3BR

A quick trick is to use the doors as reference. Look at bedrooms 2 and 3. Their width is just double that of the door. That is very small.

Plus the two balconies and a/c ledge. That effectively makes the living area only around 75 sqm. DBSS flats have such large balconies because they count 10% less in GFA (Gross Floor Area). Before 2008, bay windows and planter boxes were not counted. Guess what happened?

Lastly, show units are always deceptive. Caveat emptor.

Thankfully, this is the last DBSS project.

Millionaire cameos

The first release of Millionaires' Express is a subpar transfer of film to DVD. This is typical of first-gen HK DVDs.


Universe Laser R0 DVD 2000 (626 x 278)

The video is heavily cropped, although nothing important is actually lost.

When the first-gen DVDs were released, many people thought these were the best possible, because Hong Kong films were not known to be preserved properly. Luckily, the films held up better than expected, or the restoration worked magic.

The next release, six years later, is so much better.


Joy Sales R3 DVD 2006 (842 x 370)

But it is surpassed very slightly by the Dragon Dynasty DVD because this has the widest shot of all (2.32:1 AR).


Dragon Dynasty R1 DVD 2007 (854 x 368)

The blu-ray is even sharper, but not by much (this is scaled down to 360p):


K&M Blu-ray 2013 (1920 x 817)

It has the same look and color as the DVD, so both of them must be from the same digital master.

Aspect ratio can cut both ways. 10 pixels are cut off from the top and bottoms to create 2.35:1 aspect ratio from the wider 2.32:1 AR.

The film with the widest cast

Millionaires' Express [富貴列車] (1986) by Sammo Hung is well worth watching. It is mostly a light comedy, but finales with 15 minutes of an all-out fight. But it is not perfect. It has multiple storylines and goes on for too long. And there are too many sub-plots at a total tangent to the main plot.

This is a big budget release. Imagine a whole town built as a location set! There will never be anything like it in Hong Kong movies again.

Universe Laser released the show on non-anamorphic DVD (R0) in 2000 with a run-time of 96 minutes. The video is just 626 x 278. The sides are cropped to give an aspect ratio of 2.25:1. It has Cantonese and Mandarin audio.

In 2006, Joy Sales released it on anamorphic DVD (R3) with run-time of 101 minutes, but it is not a superset. It has several new scenes, but it is missing a few scenes too. The video size is 842 x 370 (aspect ratio 2.28:1). It has Cantonese audio only.

In 2013, K&M released it on blu-ray with 90 minutes run-time. The video is 1920 x 817 (aspect ratio 2.35:1), but the top and bottom are cut off very slightly. It has Cantonese and Mandarin audio.

To get the full set of audio commentary, interviews and trailers, we need to get various international releases, such as the US (R1), UK (R2) and Australia (R4).

Hopefully, one day there will be one ultimate release that contains the full cut (around 107 minutes) and all the extras. :lol:

It is said there are a lot more scenes — mostly cameos by other actors — left on the cutting room floor. Maybe they will be found in time for the 30th anniversary next year! :-D

On the run-time

I want the complete cut as a completionist, but I feel the show is too long even at 90 minutes and can cut away more scenes. Many of them are to give the huge secondary cast some screen time. It is a treat to the viewers, but it is detrimental to the show's pacing and cohesiveness. Many of them don't interact with the main cast a single time!

The definitive release today

The best release is Dragon Dynasty's R1 DVD from 2007 with run-time of 101 minutes. It has anamorphic video (854 x 368), the most zoomed out video with 2.32:1 aspect ratio, Cantonese (including the original mono) and English audio, four deleted scenes (6 minutes), audio commentary and all interviews except for Yukari Oshima.

The only problem is that it is now OOP. :-(

Defective good, blacklisted shop

This double-chain pearl bracelet caught Missus's attention. It had the best design of all the sub-$100 pearl bracelets.

Unbelievably, the catch became loose on the first day I bought it. The saleswoman had "warned" me there would be no exchange or after-sales support when I bought it. At that time, I wondered why would I need that. Jewellery don't break down.

It got worse. After self-repair and Missus wearing it for just one day, the metal ring on one end detached.

After re-attaching it and wearing it just one more time, one end of the pearl beads detached and the ring was lost.

After a very long time, I decided to salvage the bracelet by removing one set of beads and making it a single chain. It lasted just one day. One end of the pearl beads detached from the ring.

This is my first and last purchase from Citigems.

The modern life skills

Three essential life skills that will make your living much easier in the current era:

  • scripting
  • networking
  • audio/video encoding