The most accurate Raspberry Pi atomic clock IN THE WORLD!

[@deefdragon]

The technology connections quip.... take your darn like. seriously tho, I cant wait for a video showing off two computers with the EXACT same time using that hardware standard

[@rhekman]

To demonstrate PTP across multiple computers, through the magic of buying two of them.

[@vincentguttmann2231]

@@rhekman And through the magic of buying three of them, I have an already taken apart one right here!

[@ethans4783]

stole my comment! I'm super excited to see what this card can do for us in the mass market

[@arjdroid]

A Technology Connections Connection?

[@مقاطعمترجمة-ش8ث]

What's about connections latency,

[@rougenaxela]

1:13 Okay, I just have to call this out. The "> 200,000x more accurate!" card here implying that a higher frequency makes it more accurate (or that these two frequencies even related to each other in a meaningful way) is **highly** misleading, straight up wrong even. Higher frequency does not mean better, not in the slightest. You can have a 1THz oscillator that is absolute garbage at keeping time. What matters for an oscillator is it's tolerance, stability, and phase noise. In this case since the oscillator is GPS-steered, tolerance matters slightly less, but stability and phase noise are still critical, and this is what something like a rubidium oscillator excels at, but it has nothing to do with how high a frequency it is, it has to do with how stable that frequency is, how little it shifts due to both environmental conditions and random fluctuations. Now, there are technical reasons related to phase-locked loops (which is used to get a different frequency out) why an underlying frequency standard should be reasonably high frequency, but that's not what sets quartz and rubidium oscillators apart from eachother, you can get quartz crystals tuned for the GHz region. What sets it apart is that a rubidium oscillator depends on atomic scale properties of electron energy levels in individual rubidium atoms which are relatively fundamental and unchanging, rather than depending on the relatively 'macroscopic' properties of a chunk of quartz crystal as a whole which depend on physical dimensions and environmental conditions.

[@JeffGeerling]

This is true. Indeed in running the numbers after I had recorded my script, I realized the accuracy of the rubidium clock is actually more like 300,000x better than oven-controlled quartz-which shows the ~200,000x difference in frequency isn't the reason for the better accuracy. But in these videos, I can't splice in anything or even add annotations. So I'll just hit the heart on this comment, leave this reply, and hope that people see it :P

[@vincentguttmann2231]

Since you seem to know a thing or two about oscillators, you might be able to answer a question I've been having ever since I heard of this project (and the price for the rubidium clock): What is the main cause of drift in quartz clocks? Is it temperature? Or manufacturing tolerance?

[@tuppyglossop222]

@@JeffGeerling Don’t worry Jeff, I saw it...

[@771racing]

@@vincentguttmann2231 Typically temp is the primary offender. An old school method of improving an NTP server's perf was to purposely insulate it a bit to force the temp up, as holding an elevated temp stably is slightly easier than a consistent cool temp. Or you could move up to an ovenized quartz oscillator, which as the name implies does the same thing for just the oscillator itself.

[@vincentguttmann2231]

@@771racing Hmm. Now I'm thinking of how to make a PCB sandwich for a quartz oscillator and temperature sensor. It won't be as stable as a properly ovenized quartz at like 70 °C, but with no big fluctuations it should be better than a room temp quartz.

[@keithmiller9665]

You compile the Linux kernel so often it would be interesting for a video of how to do it. Just a suggestion.

[@abyagowi7]

I second it. Jeff, I think it would be great to have a video about it.

[@inventalone4339]

Yes please,sir do a video on this.

[@Anthony-rp9rg]

@Keith Miller He must have already made the video: kzbin.info/www/bejne/nXS1q6GdbLGmh6s

[@SebastianScholle]

i'm onboard for that video!

[@208Concepts]

Compiling the straight kernel isn't too bad... Cross compiling for different arches is where it gets fun 🤣

[@famitory]

love when open source projects like this come out, not because of the projects themselves, but because it makes the appliances they replace drop in price on the used market. finally, an evertz msc5700 for my very own home

[@jrlx86]

Quite. Pretty sure the timecard that Jeff has is something developed by Facebook engineers. It's open source but to buy it built is around $1k used from memory. Of course, I don't have my own atomic clock so I don't remember when I read this info, if ever. - EDIT: Just watched the video, good to know I'm part of this universe and not a parallel one.

[@abyagowi7]

@@jrlx86 I am working with several vendors to make the card (just the card with the atomic clock, SOM and GNSS) for about $200. Then you can put the parts based on need. Stay tuned

[@perwestermark8920]

I manage quite fine with just a uBlox. Given that it listens on 4 different satellite networks, it would take a GNSS jammer or a broken antenna for me to lose track of satellites. And the ubx protocol will happily show the confidence error of the emitted time, while the 1-second pulse avoids the latency from sending out the time data over a serial interface. This is the pulse that would be used for IEEE 1588.

[@mikeiver]

Why not just bur a card from a decommissioned cellular site. They are already aged and easy to interface to. They have a really good GPS receiver built in for clock sync and usually a TCXO or a OCXO. The one I have also has a pair of outputs, a 2Hz and a 10MHz. Interface is via RS232. Frequency stability without a GPS lock nudging the XO is better than one cellular frame in 24 hours. Best part is that they are dirt cheap. Far less than $100.00 when I bought mine.

[@MarcoGPUtuber]

An atomic clock? It's everything I've ever wanted! It will finally ensure I am on time for work. I always end up 1 ns late. Got written up for it too!

[@BenCos2018]

lmao

[@Wordsnwood]

You're making me nostalgic, Jeff. Setting up NTP / XNTP on our network was my first duty wayyyy back when I got my first job as a sysadmin. Stratum 3 or maybe 2 was about as close as I got, though. Everybody needs accurate time, but they don't realize it. Also +1 for the Technology Connections comment. Very meta.

[@mattparker9726]

this card is doing the rounds for sure, just watched LTTs vod in it.

[@leonn3k]

Jeff and Raspberry Pi were the first things I thought of when I watched that video.

[@ccflan]

I guess that jeff speaks more technically in details more than ltt, i love when he include kernels compile sin hus talks

[@jockinafrock]

I work as a tech support field engineer for a major telecoms provider. Our entire network is synchronised to 3 Caesium clocks, cross referenced to each other and with GNSS receivers. We distribute this sync to our customers, most of whom have no idea of the ramifications. However we have one customer who are halfway through migrating their entire platform from TDM (i.e. E1 2Mbps circuits), to Ethernet. Because of the mixed technologies, their services are critically sensitive to clock jitter, and we are spending a fortune in analysing, adapting and managing their circuits to reduce these issues. I have been using Symmetricom Rubidium / GNSS test equipment as reference clocks to monitor their circuits, and the tolerance levels are so miniscule. Great video.

[@careytschritter1108]

Please don’t let Red shirt Jeff near the atomic elements. We would risk too much to the North American continent.

[@JeffGeerling]

I got a call from the EPA yesterday... they said something about a superfund?

[@zombie_pigdragon]

@@JeffGeerling Hey, that sounds like a good thing! You can get Red Shirt some toys of his own, with that kind of funds!

[@JeffGeerling]

@@zombie_pigdragon Oh, so that's how it works! Let me go tell Red Shirt Jeff he's good to go!

[@steveswartz1875]

Sorry to say Rb has two common isotopes neither of which is particularly dangerous. A typical Rubidium cell in a miniature clock probably les than a milligram of the stuff anyway. The pure metal is quite reactive, more than sodium, or potassium .

[@vincei4252]

The Rubidium unit on the card generates a 10MHz TTL clock. I have working Symmetricom Rubidium clocks that were bought a fraction of the $700 for the Microsemi unit so I think I'll go head and build one of these and connect it up to my existing 10Mhz TTL reference. Thanks for sharing, Jeff. Oh, and this is the best coverage of this device I've seen to date.

[@deefdragon]

it managed to strike the good middle ground complexity wise. Ian was for to indepth, while linus was basically a quick summary and the obligatory pointless "play games with it"

[@JeffGeerling]

Linus definitely has the monopoly on 'sexy video shots of the card' though. I'd kill to have more than 12' sq ft for my entire media production and test bench setup :D

[@dave_dennis]

Vince, I didn’t follow exactly. Does the card have a 10 MHz TTL output on one of the SMA connectors?

[@vincei4252]

@@dave_dennis Hi, that I'm not sure - I need to look at the full circuit diagram in Altiium, however, the rubidium clock source that is soldered to the board has a 10Mhz TTL clock output. What I meant is that rather than fork over $700 I can jury-rig a connector and cable from my larger rubidium sources (also made by symmetricom, now Microsemi?) and solder the cable to the clock-in on the pcie card. The source they use also has rs232 which mine does too but it'd lay bets that whatever is exchanged over that serial interface is the same protocol. I remember being able to open a putty rs232 session to my rubidium and query the device for status, etc.

[@JeffGeerling]

@@dave_dennis According to the specs, it has "1PPS / 10MHz SMA output" & "1PPS / 10MHz SMA input" - see github.com/opencomputeproject/Time-Appliance-Project/tree/master/Time-Card#interfaces I haven't dug into it yet, but it looks like those interfaces go through the FPGA, and are not *directly* attached to the Rb source.

So this is the third video I watch about this time card, yet you've added a ton of new info which I was not aware of. Like how Linux identifies the card, or that PTP requires hardware support on the network interface to work. I got a bit more excited about the possibilities this card offers. I was stuck in my head thinking on the database level only, that you can organize transactions in such way that you can essentially use optimistic locking all the time and non of them will collide. As a developer I thought "yeah, that's cool", but it's the more practical use cases which opened up my eyes. I'm really looking forward for the next video in this topic :) Also, I was thinking the other day about making a GPS based NTP time server on my network from a Pi, because "I can"(tm). GPS radios are not that expensive now. That could also be a good intro video to network time services. Shame on me, I never had the urge to look up how even NTP synchronization work, and PTP was completly unknown to me before this card. But it's an interesting topic.

Thanks for the heart Jeff by the way. But one of the typos was getting on my nerves and editing a "hearted" comment makes YT remove the heart. Well.. I'm now heartless :D

[@TechTechPotato]

11:08 Loved it!

[@MarcusPHagen]

Once again, Jeff, you provide a thought-provoking video which has me spending even more time reading through the debates & information in the comments. I'm grateful for the level of engagement that you have with your viewers, as well as as the way in which you accept corrections when additional information is presented by others. Yours is one of my favorite channels, providing both information & humor. I wish you a long & successful run.

[@JeffGeerling]

Thank you!

[@DanDrown]

Thank you for the video, I'm excited to see what you can do with this hardware. I have a few responses to your video: - The rubidium clock's accuracy is not based on it running at a higher frequency than quartz, but rather a more stable frequency. Temperature changes cause quartz crystals to change frequency more dramatically than rubidium - The 32khz crystal runs the real time clock (RTC), which can keep time between reboots. The RTC is typically ignored on the pi because it does not have a battery backup for the RTC, so the program fake-hwclock is used instead. The main system time comes from a counter in the CPU on the pi (driver is named "arch_sys_counter") which is driven at 54MHz - NTP can use hardware timestamping as well, and can reach microsecond to hundreds of nanosecond levels of sync in proper conditions

[@JeffGeerling]

I shall consider myself educated! To be honest, prior to this project I've only worked with NTP and only really cared about millisecond-level resolution. Though that darned Chicago server has made me more interested in better holdover accuracy.

[@DanDrown]

@@JeffGeerling due to denial of service attacks, it's been popular by service providers to block or rate limit UDP in general or NTP in specific. This could be what happened to your Chicago server. I know Linode is blocking IPv4 NTP outbound, but IPv6 is allowed.

[@JeffGeerling]

@@DanDrown That could very well be the problem.

[@charlesgantz5865]

Also, as I recall, while the rubidium clock frequency is very stable, its phase noise isn't very good. Crystal oscillators, on the other hand, have frequencies that aren't very stable but have very good phase noise characteristics. That is why the rubidium clock is used, through a phase lock loop or some other mechanism, to maintain the crystal frequency. Then the crystal output will be highly stable in frequency as well as have excellent phase noise.

[@Itsdirtnaptime]

I love the Technology Connections reference. It is great when people reference other channels, especially when they aren't in the same category of films.

[@JeffGeerling]

We are all in the category of "trying to satisfy our insatiable curiosities" :)

[@Itsdirtnaptime]

@@JeffGeerling that is true but uh. You are on raspberry pi's and he was on a 6 part lantern series. 🤣 So I am still sticking to different categories. But as for the passion brought into your specific areas, I agree that is the same category. You both do create unique and very intriguing videos.

[@JeffGeerling]

@@Itsdirtnaptime Well now, he hasn't yet done a 6-part series on powering Raspberry Pis off the heat generated by portable lanterns, has he? Sounds like it's time for a collab! :D

[@Itsdirtnaptime]

@@JeffGeerling that would be great. He can comes up with the fuel and wick types and the concept of how to control the fuel being supplied. You do the technical side to control numerous lanterns across different states that turn on exactly at the same time and control fuel delivery. Good way to put that $700 clock chip to work. 😂🤣😂🤣 Internet lanterns. 😳😳 old meets new. Would that be considered a Technological Connection?

[@ur1friend437]

Your humor and great work is, in my opinion, the best “family friendly tech channel”. I am eternally grateful for all your great work. I will keep sending you, and love ones, endless great wishes and blessings.🙏🏻

[@IamYuto]

It's always nice to see you giving a shoutout to small content creators like Linus.

[@JeffGeerling]

Gotta help promote those smaller channels! :D

[@ImprovisedExpletiveDevice]

That's pretty cool. I just picked up an HP5065A, an old school Rubidium frequency standard. Lot of cool things you can do with a super accurate clock.

[@DETERNET]

My job is working for one of the manufacturers that makes SMPTE 2110 gear, PTP is very important for doing uncompressed video over IP. It’s a neat field to be in.

[@abyagowi7]

Let us discuss this in more detail.

[@muddyexport5639]

Good job. As usual, rewarding new vid notification. Always interesting, topical, and "in the end -- funny". Thanks! Keep on keeping on.

[@ghostdog0424]

3:30 Many time standards (at least in my grandfather's collection) are 2-3U rack mount chassis, with about a quarter to a third of that space being the cesium beam tube in its high-vacuum housing and oven. The size difference is truly dramatic.

[@hikaru-live]

There is a Dallas DS3231 chip that packs a DTCXO. It is a lot less precise than atomic clock, but much better than a regular crystal, and much cheaper too. I have a Pi NTP server project that uses GPS for time base, a DS3231 for local timekeeping, and a STM32F030 as the governor of DS3231.

[@BienestarMutuo]

Yes, your idea is the right solution. millisecond precision at low cost is a very good solution for many time problems. You can please share your project, or sell your device, i and many will buy it.

[@hikaru-live]

@@BienestarMutuo It was available on my Tindie page but I currently have no manufacturing facility available to make them.

[@thecorley]

When I got to my first submarine, we used 2 cesium clocks to provide reference signals to most of the equipment. The Cesium clocks weighed about 75 to 100 pounds and we had to take them to get calibrated every 6 months. They had a battery backup that lasted about 20 minutes, so we had to get them to the shop quick. On my last submarine we used a rubidium clock. Smaller and cheaper is better. We could sync it up with GPS ever few days to make sure it stayed accurate.

[@BrianBoniMakes]

Another bleeding edge RPi episode. I'm interested in following where this goes.

[@johngangemi1361]

Accurate time is also important in electrical power networks and telecommunications. Great video.

[@stefandriesner5026]

Already gave you a like, but the Pirates! reference earned it all by itself.

[@vincentguttmann2231]

0:05 I love the sound of cicadas. Here in Germany we don't have them, and they always remind me of our holidays in Croatia. Although the Croatian cicadas sound a lot different to yours, interestingly.

[@thomasruwart1722]

Mee too😁 I have a memory of laying in my crib when I was about 2 or 3 years old, looking at a nice blue sky and the big tree outside the open window right next to my crib. I remember feeling a gentle breeze blowing in thru the window, and hearing the cicadas "wind" up and down. Hearing cicadas always relaxes me and triggers that calm memory from more than 60 years ago.

[@thompsant]

From a personal perspective, having a potential open source PTP clock as a grandmaster for 2110 is very exciting. On the other hand, I work for a company that used to provide a Raspberry Pi as a hardware solution for a broadcast product and well... it was not received very well by sales staff or customers, which I think is kind of a bummer. I'd love to run into these when commissioning broadcast facilities.

[@jakenkid]

It was so damn pleasing to get a TC blinker reference!!! You don't even know! 😁😁😁

[@richardkicklighter8280]

Jeff, you never fail to amaze me. I'll be looking for more on this. Great subject.

[@mitchellmnr]

The puns are timeless :/

[@royk7712]

Stop, get some help

[@mitchellmnr]

@@royk7712 You cannot stop time ...

[@CristianConsonni]

@Mitchell R BA DUM TISS, sir

[@comicsansgreenkirby]

* sighs in exasperation *

[@eesachariwala2145]

Thank u sm I rlly need this video, I’m working on an amateur global radio interferometer and I needed a way to integrate the atomic clock mechanism for data recording… ur a lifesaver

[@briceperdue7587]

I've watched a handful of your videos. This is the one that really got me. Our fiber build is running ethernet frames and keeping a very good clock is a new thing i've had to even think about. Very good stuff very boring topic like NTP and DNS no one really cares until it doesn't.

[@Codeaholic1]

Jeff, Im curious what the issues are that prevent you from using NTP on your Chicago server?

[@niklasxl]

When i saw Linus and Ians viodes i was think if it was possible to run it on the raspi and i am happy to see that you did :D looking forward to more videos about time :D

[@dangeist2899]

Nice video on the Rb + PTP + CM4 setup. Looking forward to support for the HW timestamp in Broadcom chipset. Although it's not as a sexy as the "atomic clock" rubidium setup, it might be nice to do a quick build on the CM4 with a normal GPS HAT +quartz RTC setup. It could still give you microsecond-level accuracy and with chrony's partial hardware timestamping support would be a quite capable NTP/PTP source for slightly lower accuracy requirements (like homelans :) )

[@chris-tal]

It's also used in the test & measurement industry when one has to distribute the same 1PPS ,10Mhz sync and PTP time domain over a dozen hundred meters with as minimal drift as possible. Mostly for making time sensitive measurements in sync across distant data aquisition points of a distributed system. Symmetricom is a well established and known name on the market. Ofcourse this card will bring a more affordable solution to smaller companies and even individuals. Hurrah Time Appliance Project!

[@autohmae]

I wonder what the Low-Frequency Array (LOFAR) project uses.

[@chris-tal]

That's a good question. It's way larger than to sync inside a factory building's several business units which manufactures and tests high-freq radio equipment. On that large scale (across bordering countries) I guess it's the same GPS+Rb per receiving station. Combined with singlemode fiber on shorter distances just like the CERN white rabbit network. Plus some magic code that handles exceptions and maybe even mapreduces large chunks of data. You can easily find some articles on the net about both projects if you're very interested. It's a deep hole. Just "follow the white rabbit"! :)

[@artemartemov1242]

​@@chris-talPlease explain to me how to get 1pps with a frequency of 1hz taking into account the time update. Raspberry doesn't allow this. Using a python script I get a stable pulse, but it doesn't have synchronization like pps

[@LimitedWard]

If every self-driving car had one of these atomic clock modules in them, they could communicate with each other to warn about the position and speed of potential obstacles with a high degree of accuracy. They could also use it to coordinate traffic and reduce congestion.

[@phillee2814]

You only need precision to a few milliseconds for that though, and if the cars cross synchronised in an NTP peer group, you could have µS precision easily - and with GPS practically universal in cars anyway (but most lack PPS, which would be a simple fix), each car could be stratum 1 PPS peered to all the others around it also at stratum 1, so the worst that could happen on the loss of GPS, Galileo, BeiDou, IRNSS, QZSS and GLONASS signals simultaneously is that the whole group could drift by a few µS. One car with a faulty GPS would just drop to stratum 2 automatically, as that is built into NTP already. That just means it gets its time from the other cars around it, with very little loss of precision (offset due to distance [a nS is about 30cm] is built-in). Galileo gives horizontal

[@ConnorisseurYT]

You definitely don't need an atomic clock to do that

[@theoriginalrecycler]

Key use for this is transactional analysis, stock trading etc… minutes give 60 transaction windows per hour, seconds give 3600 transaction windows per hour. The rubidium source allows more time slots per hour, hence more transactions per hour executed. I won’t list assumptions etc.. you get the idea.

[@seangraves3509]

Wow there I thought having just got my RPi4 being an NTP server was great a time piece ... I do hope that card does not glow in the dark ;-) Nice video Jeff 👍

[@ЕгорК-ж6о]

To actually sync the clock in a data centre over the network you need a support from network devices. You mentioned that the card does not have the support yet but it is only software. Well many networking devices also do not have hardware timestaming. This is actually a heavy topic if you want to dive into it. For 5G application for example the accurate time is used so multiple base stations can send data over to a handset at the same time. There are multiple options to distribute time over the network the first one is via Synchronous Ethernet links and the second one is EEE 1588v2 PTP Boundary Clock. Many if the telecom providers use the second option as the hardware is cheaper. Same as on your cards the vendors like Broadcom created ASICs with support for hardware timestamping.

[@unperrier5998]

Many cristals and oscillators are running at a high frequency (dozen of MHz) and are actually precise. And they can be calibrated, even in software. The problem is not the low frequency but environmental changes in temperature (like you said) but also voltage that make them drift at runtime and unpredictively.

[@AustinSteingrube]

On the industrial manufacturing side, there are exceptionally few things that need more than 1/10 sec precision. The things that DO need to be coordinated are not done over a network, but rather point-to-point discrete signals. It is far faster to look for a rising edge of a 24V signal over 20m than it is to wait for network traffic to propagate to the devices. The PLC’s that run and connect things are themselves very slow compared to a netbook, so any PTP precision would go straight out of the window.

[@TrollingAround]

Jeff, you could make a video about drying a towel and I'd watch it, probably to the end, and then click 'like' - please don't! Please keep the fascinating content coming. :D

[@ezzeldin101]

Dude your videos are REALLY... REALLYYYY awesome

[@maximusboscus]

I think synchronized motion in the industrial automation does not normally uses PTP as is. If Profinet is used in IRT mode (isochronous real time), delays are within the nanosecond. PTP is included in a wrapper protocol called PTCP (Precision Transparent Clock Protocol) and besides allowing you to synch the clocks, it can calculate the delays in the switches and cabling. For this to work, special hardware is required at both ends as well as for the switching equipment. The next best thing in Profinet RT (real time) with which you can get down to the microseconds. You can use normal hardware for that, even through Wifi. Oh and fun fact, both of these modes skip completely layers 3, 4 and 5 of the OSI layer, so basically from the Ethernet layer you go directly to the layers 6-7. No need for IPs.

[@ghostdog0424]

My grandpa helped pioneer CSACs! I have one of his 10mhz frequency standards and it's rock solid. He still fools with cesium beam tubes using a raspberry pi for logging the data

[@ghostdog0424]

i'm going to send this video to him, will post a response with his reaction

[@CaioVinicius-nc3cy]

Very nice Jeff, as always. Thanks for the video.

[@ed.puckett]

Another typically informative video! Thank you!

[@SebastianScholle]

the field of metrology uses highly stable and calibrated measuring equipment. this device promises a more widespread adoption of such science, which i think is a very exclusive club at the moment. Having this card run on a PI really makes this field in science more accessible to a larger demographic therefore helping further the research and opening up more opportunities.

[@trelauney]

I was waiting for this. And KZbin waited a day more to show it to me.

[@jwillisbarrie]

Thanks for adding actual captions for the Deaf and clear explanation.

[@vmsysprog]

Timing is everything.

[@Openspeedtest]

I was waiting for Atomic Pi

[@JeffGeerling]

A _truly_ Atomic Pi ;)

[@rhekman]

@@JeffGeerling and here I thought you couldn't trust atoms. Since they make up everything. ... I'll see myself out.

[@phillee2814]

For the vast majority of users, a version of this at a fraction of the cost with ONLY the GPS system - and with backup to other NTP servers on the internet. Selling such a simplified version would be a real money-spinner, and provide easily sufficient precision for the vast majority of PC or LANs, and potentially increase the number of available stratum 1 NTP servers available in the pools to the point where the load on each would be insignificant. Such a basic GPS/PPS PCIe card (with drivers for all current operating systems) is a major missing link. Once you get distribution around a variety of devices on a network, NTP is quite sufficient. PTP is major overkill for almost everything we do at less than data centre or physics laboratory level.

[@perwestermark8920]

You don't even need PCIe for GNSS-only time reference to host a NTP server. But you do need the 1 pps signal if you want IEEE 1588 where the network data is timestamped in hardware.

[@mitnoxin]

I used to use a similar system for calibrating the A2D in seismographs. It would output a square wave at a series of fixed frequencies. It was a lot bigger than your card there.

[@porm1930]

Good stuff you create, both professional and entertaining. I am planning to build a small server at home with some 10-20 Pi. How do you do and which Pi to chose? Is Kubernetes something to use? Which board is the best? Do you think you can do a video about this?

[@blacktrooper100]

This is so AWESOME I want to build my own stratum-1 server now. 🤩🤩🤩

[@betterlifeexe4378]

10:04 an open source, simplified time synchronization platform could be made more secure and cross-platform. If adopted as part of a standard for communicating info between smart cars, we might see automatic safety features get a lot better. particularly in traffic streams where several smart cars can respond to each other predictively. Autonomous and maybe semi-autonomous caravanning also becomes more practical.

[@KirilMiloshev]

Nice video! There is a good option with GPSDO Gps disciplined oscilator. You can get one for about 100 bucks and hook it to the raspberry. It has several nanoseconds drift.

[@johnkunze5362]

We use an array of AC's all around the world interlinked for NASA, correlated to our telescopes to determine the wobble of the earth. The timekeeper is Dr. Phillip Arctender(really...birthname). A polar shift potential is high. Loss of Antarctic ice is prime factor. Nice to know we have it on a pi.,jpk

[@MrMftech]

Thanks for your clever explanation...clocking is everything in Digital :-)!

[@briangoldberg4439]

wow man. this is a very pun-heavy video. good job

[@JeffGeerling]

I had a few more to add, but I ran out of time :(

[@net_news]

audiophiles could be interested in this clock to be used as master clock for DACs.

[@perwestermark8920]

Better CD/DVD/BD players sends out the clock signal and better amplifiers have an input to control the DAC. It is normally not that important to run at exactly 44100 Hz absolute frequency. But it's important that the DAC runs at same frequency as the generated PCM data so the DAC doesn't skip some samples or dual-samples some.

[@RandornCanis]

Snake oil. Accurate sync between hardware clocks is important. Dante, AES67, and SMPTE already use PTP V2 with the oscillators we already have to achieve this in professional audio. Beyond sync, a precise global clock would still yield frequency stability in ppb from ppm. It takes specialized frequency counters to properly characterize stability at this order. Hearing any difference - not remotely possible. Phase noise of any local oscillator would dominate. Does sound like the bait audiophiles would burn cash on, but only meaningful to broadcasters mandated to deliver a middle A at precisely 440.000 000Hz.

[@perwestermark8920]

@@RandornCanis I was one time told to buy a $200 (more like $400 in todays money) optical cable (with gold-plated connectors) from CD to amp. As if that would make it sound better. Best is the gold-plated connectors for the *optical* cable that doesn't even need any electrical circuit to transfer the sound...

[@RandornCanis]

@@perwestermark8920 Sounds about right. Probably for the directional wide-stage copper. Buzzwords!

[@perwestermark8920]

@@RandornCanis All audio cables are marketed as using oxygen free copper. Just that is was relevant maybe 1950. Today all copper manages the conductivity levels required for oxygen-free copper. But audiophile sales pitches are magical. Who would dare to not use gold-plated mains fuses for their equipment - the shame...

[@vincentguttmann2231]

9:35 I think you meant to say that Red Shirt Jeff has barely scratched the surface of the card...

[@Balam-m9n]

For over a decade I have been aware of the systemic benefits of hardware clock module upgrades. These benefits go beyond synchronization issues, and improve board functions resolution and noise levels. Might there be some way to rig this clock to REPLACE the clock functions that run processing and GPU functions? There might be significant benefits to be realized!

[@IMBlakeley]

My last job we used the GPS receiver on a small outdoor radio to let it run as a mini GM clock, the proper big Grand Masters were so pricey there was quite a bit of customer interest. I'd bet they'd be even more interested in a Pi solution.

[@koitsu2013]

7:08 This is partially incorrect; /dev/pts is a directory containing pseudo tty allocations (ex. /dev/pts/0); see pts(4) man page. You may have seen these mentioned when using the w(1) command. I believe you meant to say /dev/ppsX ("pulse per second"), which is a userland interface for a high precision timer -- which is not present in your video. You probably did not include the option in your kernel config, and/or 5.12 may not offer this to userland (unsure); 5.15 may in fact do this, but I'm not sure as the GitHub page for OCP TAP is not clear, while FB's own engineering page does mention it (though under a very different tree). Finally, it's also possible you confused ptys and ttys, the latter of which are also associated with serial interfaces (which are quite common on several receivers; many GNSS and NMEA devices include one as their way of userland interface).

[@nathanieldoyle4972]

Random question, where does one acquire the new time card? Had a quick google but didn't find anything

[@rougenaxela]

2:43 I see you there corner text that is a critical factor in the story

[@JeffGeerling]

Maybe I am insane.

[@mauisam1]

I liked your outatime joke and the shaun the sheep clip.

[@abelgerli]

There's also another option. When the atomic clocks send their signal on local positions on eart you can do differential GPS with centimeter accuracy. Would may be a genius projects out of the ordinary.

[@benverdel3073]

In Europe we have DCF77. A cesium based signal from the hart of Germany with a range of about 100 miles or 1500 km. It the reference for all of Europe. Trains, postoffices etc. You can get receivers for home use as well.

[@JeffGeerling]

The US has WWVB, which does a similar thing-once you know your relative distance to it, you can get pretty accurate time from the radio signal, too!

[@RoelBaardman]

If you have the antenna well positioned, you might be able to run an NTRIP server. You can help people nearby improve their GPS accuracy.

[@DJ-Manuel]

Or you could run NTRIP client and get more accurat timing 😉 With NTRIP server it is best to have multiple receivers and have a differential setup, in order to have more accuracy in location.

Oh yes, it's about time!

[@creesch]

Interesting video :) Small bit of feedback on editing, you are putting text near the bottom of the video which disapears quickly which is normally fine as you can pauze vidoes. However due to the position of the text the youtube video controls cover it when the video is pauzed.

[@JeffGeerling]

Gah, I have to try to remember that. I wish KZbin would let you hide the controls while paused.

[@locusf2]

Also if you are going with the hardware timestamping route, use level 3 multicast first, ie. UDP. The ethernet multicast can be tricky if there are multiple switches or routers that don't support it.

[@shubinternet]

I think the hardware support for timestamping only works at the Ethernet layer. I agree that multicast is the most scalable solution, but many switches and routers don’t support it, or it’s turned off by default. This has been an issue for many years.

[@catzymbs]

What do you know, you need to put a linux box near windows to get accurate time

[@SuperDavidEF]

Funny joke, but ultimately wrong. The rubidium clock is orders of magnitude more accurate than the GPS signal that he is putting the card near the window for.

[@catzymbs]

​@@SuperDavidEF joke aside, there is a reason why you put GPS antenna plug there and to sync the clock to the rest of the world, you need a master clock.

[@FennecTECH]

There is actualy a use for all clocks having a time card like that…. Temporal accuracy is necessary for self driving cars to cooperate creating a more efficient network. Being half a second off when deciding who is going to do what when at 200 mph can and will have deadly conciquences. So for cooperative self driving to work you will need accurate clocks.

[@perwestermark8920]

The cars don't need nanoseconds. And proper ntp does not have 0.5 seconds time errors. Given that this board sets the time from a GNSS module, the current GNSS in the car is already managing quite well. Even redundant GNSS modules with redundant GNSS antennas is a small fraction of the cost of the rubidium module.

[@matuzaato]

This piece of hardware is sooo cool!

[@stefangr1]

The cesiusm used in atomic clocks is not radioactive, it is cesium-133, a stable isotope.

[@funskaterfriedberg579]

Hi Jeff, in our Mainframe we use a STP (sysplex timer Protocol) that Provider an Atomic Signal also. STP is a mix of the Hardware and the Software of the operating Systems. It is doing s nandshake, so that very Mashine measures the runtime of the Signal and corrects automatically the time to be correct. Do normal Servers are Build in the Same way)? Because, if not, Evelyn Server will run with different timestamps again.

[@raulsaavedra709]

@1:10 More than 200K times more accurate than a quarz clock! Gee, quite the jump in orders of magnitude of improved accuracy from the beginning of time keeping at this scale of relatively small devices. Such darn accurate clocks becoming ubiquitous at even smaller and smaller scales in the future (hopefully and progressively also more and more affordable as well) will indeed likely change quite a few things.

[@johnmurrell3175]

As light travels 1 foot in a nanosecond the lengths of cables are all critical. To get accurate time to 1ps you need to measure cables to 1/1000 of a foot & understand how much slower the signal travels in the cables used - signals in optic fibre cables only travel at about 70% of the speed of light.

[@perwestermark8920]

The electromagnetic wave moves at similar speeds for electrical signals.

[@maxheadrom3088]

Curiosity: mechanical watches (the type with the torsion pendulum - wristwatches) was created to know the longitude! It required a sun dial to do that, though. another: never had an iPod - still have my Sandisk mp3/ogg/flac player, though.

[@FolixOrision]

Nice call out to the synchronized blinkers

[@ordepcaetano]

Congratulations on the video. How do you know how many gps satellites can it sync to? Synching via ptp is one job, creating a non blocking switching fabric via multicast is another job. Would the rpi cm4 handle it?

[@FennecTECH]

Ah yes. Technology connections. I love his stuff.

[@datachu]

Lol, love the tech connections reference. And the bad puns.

[@JeffGeerling]

I see you too are a person of culture!

[@jmr]

That's just a little more accurate then what I personally need right now but maybe I'll need something like that next week.

[@JeffGeerling]

Once you start running HFT in your homelab... 😂

[@romanticroman11]

Jeff Girling , you 've won a prize ! Pick it up at the address of the island of Seychelles .

[@MikeStavola]

I have a few atomic clock modules. I need to figure out how to use one to make my own atomic clock.

[@rudolfhans9106]

Liked the video a lot. Just some curious questions about the applications - for example, in communication or cars. I see that adding accurate timekeeping to a cell tower would help, but where would that help in adding it, for example, to a car or a phone? For sure, one could improve GPS or make it more robust against jamming, etc. But does it help with 5G if my phone has it? Is the local oscillator currently a problem? And for using that in a car, I would imagine one could use it to synchronize the sensors, but if we have PTP, which can synchronize them to nanoseconds, why would we need the better clock? Would be nice if someone could help me! This topic is really fascinating.

[@highendcustomdesigncaraudi1777]

Thank god keept all this material in side the earth .

[@borbetomagus]

Thanks for this project! Just wondering: Are you overstating the accuracy of the the SA.31m? Datasheets for the Microsemi SA.31m, SA.33m and SA.35m (along with the SA.45s) imply they all have an accuracy of '±5.0E-11 accuracy at shipment', while Microsemi vendor Sparkfun says: 'That puts the SA.35m Miniature Atomic Clock in the picosecond category of eight seconds per 31,689 years. Said differently, the MAC loses a second every 4,000 years or so.' Then again, I may not understand the long-term accuracy.

[@upcsatx]

Show us how to deploy a low cost array of radio receivers at known locations, each equipped with time cards, and each on a network so they can be remotely tasked to receive a specific radio frequency being transmitted from an unknown location (e.g. from a drone). The receivers would then send the peak signal spike times back to a command unit, which then geolocates the transmitter using TDOA(?).

[@Jerry_from_analytics]

Now that's an application I'd love to see.

[@ndupontnet]

Rubidium in a DAC, right in my alley :)

[@wyohman00]

This is an interesting topic and one that I have long term experience with. Back in the day, we used WWV as one primary time reference. I'm having a hard time understanding why I should care about the differences between these references and NTP? What is the to gain in the specific examples you presented?

[@autohmae]

NTP has less accuracy.

[@wyohman00]

@@autohmae Clearly. Internal NTP can be accurate to 1ms or less whereas Internet-based NTP is often 10 or more ms. What I'm finding challenging is the use case for time more accurate than 1ms in normal use. I'm just looking for some examples as to why I might want to consider this for my own stuff. Maybe there's something I'm missing...

[@muhammetfurkancankaya777]

Yargı reformu paketi: judicial reform package. I didn't expect to see something in Turkish.

[@biggiejohn3360]

did they demo one of those to every tech KZbinr ?