[vasqw]

I bought one of those recently and I was surprised by how inaccurate it was. It drifted like 15 seconds every month. From what I've been reading, that is normal, but I was surprised - with all the technological advances that we have it is odd to think that cheap timekeeping is not a solved problem yet!

Also: how is the software of these watches without a CPU designed? Is it something like Verilog or whatever?

[Tor3]

At around 1976-1977 I bought one of those fancy new LED wristwatches, where you pressed a button and the time lit up in red. Straight out of James Bond, as I recall.

As I was studying electronics at the time I brought the watch to the lab and opened it and connected a frequency counter to the oscillator, you could clearly see the crystal and a little trimming pot (I don't remember by now if that was a pot or something else - capacitor maybe), and adjusted it to exactly 32768Hz.

After that the watch drifted less than one second per month, and it kept the stability for the rest of the year (until a bicycle accident which resulted in a smashed watch).

I've never since owned a watch which was even close to that. They're drifting so much that I can't even rely on my watch to catch the bus (there's a stop outside my home and the bus is there exactly on time).

[galangalalgol]

My f-91w gained less than a second over several years... While it was sitting on my desk in a giant climate controlled building within a degree of 22C year-round. When I wore it, it deviated quite a bit more, but still far better than the 30s per month quoted. I was very excited with my experiment at first comparing it to a friend's HAQ that he did wear. Then I realized the flaw in my method.

[dekhn]

People who are curious about this, it's worth reading about quartz oscillators in watches and in particular, methods of calibrating them: https://en.wikipedia.org/wiki/Quartz_clock#Inhibition_compen...

[akdor1154]

> and the bus is there exactly on time

So that would locate you pretty unambiguously in Switzerland then?

[xattt]

> At around 1976-1977 I bought one of those fancy new LED wristwatches, where you pressed a button and the time lit up in red.

This was the vibe I got when the first pre-always-on Apple Watches were first introduced with Raise to Wake - where you had to actively do an action to see the time.

[smackeyacky]

Get a Casio G-Shock. The 800 models can sync with your phone and the time is auto corrected. I bought a GBA-800 from a pawn broker - out of the dozen or so watches in my box it's the one I reach for if I have to know what time it is accurately.

[2143]

> the bus is there exactly on time

What's winter like in Japan? Is it freezing?

[_ph_]

Almost all quartz watches have a similiar movement, running of a 32kHz crystal. Those give you about 15s/month accuracy, which is an oder of magnitude better than most mechanical watches and that for a very few bucks. More precise quartz watches exists, but they require quite a bit more effort, which of course translates into costs.

Actually one of the earliest quartz watches on the market from Omega used a 2MHz crystal and was very accurate. Personally I own a Seiko with a 200kHz system, which is good for about 20s/year. Then there are the thermo stabilized systems, which are even more accuarate. The problem is, that the frequency of the crystal depends on its temperature which is the main source of time inaccuracy (there are some watch enthusiasts which do get egg breeding cupboards which have constant temperatures up to tenths of degrees, they make for very accurate watches).

Currently, the best movements on the market are accurate to about 5s/year, which is pretty amazing considering the watches are worn on your wrist in varying conditions. With my precise Seiko watch I could even notice a slight change in speed when I moved quite a distance to a different town with slightly different weather. That shows how big a challenge a really accuarate wrist watch is.

Then there are the market concerns. The most basic and cheap quartz are already accurate enough for most non-enthusiasts. Then, after almost being killed by quartz watches in the 80ies, the luxury watch industry managed to establish a mechanical movement as the desirable item. So there are few expensive quartz watches left on the market, which would feature more sophisticated movements with higher accuracy. And finally, there is a range of higher value watches which receives time signals, be it official time signals in several regions or just GPS signals.

Cornered like this in the market, unfortunately not much money went into high-precisions movements. There are still a few on the market from Seiko, Breitling, Omega and Citizen (there might be more, but those come to my mind). And of course there is the Apple Watch, which is rather affordable and just uses NTP to get absolute precise timing.

[klodolph]

To add to this—I’d say that watches are quasi-temperature-controlled. You wear the watch in varying conditions, but most people spend most of their time in controlled environments, and the watch is directly next to your body, which itself is temperature-regulated. This just means that temperature drift may be less of a factor for watches than it might be for, say, a clock on the wall or in a car.

Assuming you wear the watch.

[erosenbe0]

Indeed, another challenge is crystal aging. For example, PCF2129 has an aging offset register. These devices are rated to +-3ppm over the huge range of -30-80C but the register allows for -7 to 8ppm offsets, if necessary.

[teraflop]

It's easy to make a digital circuit that counts clock cycles with perfect accuracy. But for any clock, the actual length of each cycle is dependent on analog processes, and it's impossible to completely eliminate errors.

15 seconds per month is about 6 parts per million, which is already better than the manufacturing tolerances of a typical cheap quartz crystal. There are very few objects of any kind that you can obtain cheaply with that kind of accuracy. That suggests that, as this article says, Casio is trimming the frequency for each watch to compensate for component variations.

[somat]

A fair assessment, but I find it funny that 15 seconds per month is considered inaccurate, analog watches would kill for that sort of accuracy, a railroad quality pocket watch for example must loose no more than 30 seconds per week. I would guess that as smartphones are replacement for the pocketwatch, The way their network connected, never drift clock works becomes the new normal.

I have to admit, for my day to day use I am happy if I am less than five minutes off.

[Maursault]

> analog watches would kill for that sort of accuracy

Seiko 5 automatics have been reported to lose 30 seconds a month, even as little as 12 seconds a month.

[lmz]

You mean Mechanical watches, not Analog ones. There are lots of Analog quartz watches with 15s/month accuracy.

[faisalhackshah]

15 seconds per month is about 6.7ppm. That's well within typical crystal oscillator accuracy. If you need more than that (without external time correction such as gps or radio/internet), you need expensive oscillators with temperature compensation or even ovenized units.

[jimmyjazz14]

Citizen has a quartz movement that is rated to only drift plus or minus 5 seconds a YEAR which is quite impressive though they cost around 2000 dollars and are exclusive to the Japanese market. Several other brands out there make high accuracy quartz movements (which is done with thermal compensation) but they do generally cost more than cheap Casios.

[serf]

>yet

I mean, we have accurate electric timepieces now, and the F91-W came out in 1989; we had accurate electric timepieces then, too.

it's a cost versus value thing.

[sn_master]

> all the technological advances that we have

This is 34 years old technology. You can get GPS watches relatively cheap and they'd sync with satellites automatically and always remain accurate.

[vasqw]

Yes. I ended up buying a GWM5610 which synchronises through radio. It's cool. But still I would have hoped there would be a better "local" (i.e. self-reliant) solution.

[iron2disulfide]

The functionality of the F-91W is simple enough that I don't think a CPU would even be needed. Probably the digital parts of this chip are just state machines. That being said, the left half of the die shot looks like some kind of gigantic ROM, which could either be used by a CPU (as static program memory) or just transition logic/data for any generic state machine(s).

Verilog came out in 1984, but its use for synthesis (i.e. actually compiling text into circuits) was not popularized until much later, after correctness bugs in synthesizers and various other advancements in design tooling came around. It might have been used as a simulation/verification language for the digital portions of this chip.

[somat]

vhdl started as a project to document integrated circuits. The department of defense was getting a lot, and more complicated, integrated circuits and wanted a standard to document the functionality. At some point someone thought, "you know, if the documentation is good enough we could reverse it and synthesize a circuit from it", and thus why you use vhdl(or more likely verilog) to program your fpga.

The two languages fill the same role in the ecosystem, I have to say that I have never used ether, but my impression is that vhdl has clearer syntax(if you can stomach it's ada look and feel) and verilog has better tooling. which makes sense considering that one was a documentation project and the other was an internal tool for simulation that escaped into the wild.

[hotpotamus]

Citizen has a movement where they throw all the tech they have at it to guarantee 1 second per year accuracy. I remember things like quartz crystals have aging effects, so they used pre-aged crystals - little details like that. It even has an anti backlash mechanism on the second hand so that it ticks perfectly with no visible wobble. It probably doesn't make sense when it would be much cheaper to just use an external radio/GPS time source, but as a watch company, it makes a statement about their craft.

There are other high-accuracy watches (it's mostly a Japanese market thing), and I believe +/- 10 seconds a year is considered fairly pedestrian in that world.

[kube-system]

While 15 seconds per month is within Casio’s spec for this, it’s also worth noting that there’s reportedly a lot of counterfeits of the F-91W floating around. I got one once from a major marketplace.

[aidenn0]

~1s per day is normal for a non-temperature-compensated quartz crystal. The cheap solution is to just sync to LF time transmissions (or GPS if not running on battery power).

The F-91W might not actually have anything resembling software; just fixed function mixed-signal circuitry, but I haven't investigated it. Given the timing of its release, it's possible that an HDL was used, but also possible it was designed at the circuit level with CAD. The F-87W (which the 91W replaced) predates either VHDL or Verilog.

[ashwinne]

I was surprised by this inaccuracy too. But was impressed when I learnt that there is no analog wristwatch that can match this accuracy - they are all far far worse. (Isn't it shocking that all those super-expensive Rolexes and whatnot are all going off by minutes per month!) Get one of the Casio watches that syncs to atomic clock radio signals every night - it will be second-accurate all the time.

[Maursault]

> But was impressed when I learnt that there is no analog wristwatch that can match this accuracy

Seiko 5 automatics apparently[1] can.[2]

[1] https://www.watchuseek.com/threads/seiko-5-is-pretty-accurat...

[2] https://qr.ae/prvf6L

[chinabot]

It will be calibrated for room temperature, temperature changes do affect the oscillator frequency that is why highly accurate frequency sources have an "Oven" to keep the temperature stable https://en.wikipedia.org/wiki/Crystal_oven

[gleenn]

You might be surprised to learn a lot of servers even have largish time drift. I might be repeating info which is getting out of date but I heard 30 second drift isn't uncommon. Obviously doing NTP syncing more frequently would probably help that.

[post-it]

Yeah, the React Native debugger is broken on my Pixel because the time is different from my laptop by ~0.5 s, so short taps register as long presses or something like that.

[nightowl_games]

Anecdotal but my F-91W is currently 8s behind my phone, which I sync'd together about 6 months ago. I've always felt very impressed with how accurate it is.

[leeoniya]

my automatic Aqua Terra (8800 movement) maintains this (or better) accuracy, which i find astounding for a mechanical gear train that oscillates at 3.5Hz. i definitely would expect a well-tuned quartz movement to outperform this significantly. but i guess not for < $20 :p