Jake didn't really explain why this thing exists, so I'll try from memory: Cards like this (or similar add-in cards with small atomic clocks) are part of the PTP initiative, the Precision Time Protocol. It's a concept Meta came up with and open sourced if I remember correctly. The idea is that in a datacenter, when you send packets between servers, they have to have timestamps and constantly negotiate time, because the realtime clocks in the individual machines are not very precise. PTP syncs all the clocks in the datacenter to within nanoseconds of each other, and syncs constantly, not just on boot or once a day or so. That means you can drop timing-related bits from the network packets, which leaves more space for actually useful data. So this tech reduces network overhead and increases usable bandwidth in datacenters. That's the main point.
PTP predates Facebook/Meta by a few years (first version published in 2002; Meta was founded in 2004). For the majority of use cases (even in datacenters) you don't need the precision it provides and can rely on NTP (NTP can also be run frequently if configured as such; the reason it's not done often is because time drift is not as wild as Jake said in the video). And for dropping timing data from packets - no, that's not a thing. Most protocols use sequence numbers that are relative/based on the number of packets sent since connection establishment. But for things that do care about timing, timestamp data is explicitly included (for example, in massive transactional systems such as Spanner, all transactions have explicit timestamps that are used to decide transaction ordering).
Hey I think you are missing out something here, PTP or rather gPTP AS2020 spec makes it a lot easier for a suite of layer 2 protocols to work that collectively fall under the TSN group. I actively conduct research on TSN and the implications are pretty interesting. NTP is good for ensuring sync with some drift but fails to keep up in scenarios where you need expected latency requiremnets to kick in and work (even when for some time the time sync packet channel is being used for high speed traffic). Look into stuff like Qbv and Qav or Qcb. All theese QoS services have seen a lot of improvement with gPTP and seen wider range of applications (where NTP is just not feasible) like Automotive Ethernet and Mission Critical Software.
@@weirdguybr Open Timecard was a Meta idea, PTP is older. Open Timecard is part of the Time Appliance Project under the OpenCompute umbrella. Really should have worded that better.
@@pavuk357yes, it's a tiny amount that could add up, but very few applications need timestamps to begin with and there are other places where you can save - Take HTTPs for example - you waste more bits with all the headers than with one single timestamp.
It's a very accurate clock, there is nothing to mess around. Unless you have a large amount of servers that need very precise time to enable high speed low latency data transfer this is useless for you
In an old job, we had a system that needed accurate time and got it via GPS. Only issue we had was that we installed it in a data centre which was very secure and as a result it couldn’t see any satellites. Luckily the system had two power supply’s, so we connected one to a unplugged but charged UPS, wheeled it outside to get a signal, and then ran it back inside into our rack power so it didn’t turn off and kept time. The entire time the UPS was beeping and it must have looked like a computer heist to an outsider due to us running through corridors trying to get it to mains power before the UPS ran out.
@@thefatmoop Yeah agreed, the drift wasn’t great but we only needed it for a couple of weeks until the centre provider could source us a cable run out the building for the receiver.
@@thefatmoop a lot of time sync devices like that have also a high precision quartz clock too, which is much more precise than the default clock source in a PC
So, I'm a watchmaker as my profession, and we sometimes have people complaining that their MECHANICAL watch is not keeping accurate enough time. I now have something to point to and say "If you want perfectly accurate timekeeping get one of these" 😆
Or tell them to get a digital watch that sets time off of the RF broadcast from an atomic clock in Boulder, CO (at least in the US). I'm not aware of any mechanical watch that self-sets like this, although I'm sure it's certainly possible to do... Honestly modern smartwatches are probably more precise than even the finest mechanical watch, simply because they can sync time with the phone they talk to, or with GPS, and additionally off of cellular (LTE and 5G include timing data in the broadcast signal and you *may* be able to pull that in without even having a SIM)
So I’m genuinely curious and not trying to belittle your profession/job title, but in the timepiece industry, at what point does someone qualify to be called a watchmaker vs just an employee at a watch factory? Do watchmakers machine their own parts vs just assemble parts? Like I said, I’m genuinely curious as I don’t know anything about that industry.
My mechanical watch keeps time perfectly fine, only +/- a few seconds every month, which can't even add up over time anyways because I reset the watch every couple months to update the month counter properly (it assumes 31 days in each month). If the accuracy is bad enough to where it's noticeable enough for people to complain it might not be their fault; have you considered that your watches are not performing their sole task as well as they should?
@@catgodswim A chronometer is the highest standard for mechanical watch accuracy; it requires a watch to lose no more than four seconds per, or gain no more than six seconds per day, averaged over a ten-day testing period. If your watch is within a few seconds after a month, it's performing far better than most high-precision mechanical watches.
@@jackiecs8190 Maybe my watch is just cracked and I never noticed. Right now it's only around 12ish seconds ahead - after likely around 55 days of not having been reset.
Welcome to the rabbit hole of precision time keeping. I used to work in signal analysis and kind of found it fun to collect various clock sources from oven-controlled oscillators to cesium controlled ones. More or less just a hobby of seeing what odd vintage precision gear you could find and revive.
From someone who has written embedded PTP Master/Slave from scratch (besides other network protocols) in my job, losing internet is actually not at all a headache for a large part of customers using this high precision time sync stuff. This gets used to sync conveyor belts and other precision machines in a dedicated network, sometimes using a dedicated NIC/port, switching off every other traffic except for the PTP packets (to stabilize network stack processing jitter with mid-tier ARM devices) or using an entire second network just for the PTP connections. Also, while it is good to know that you are "n" nanoseconds precise on a hardware dedicated for this purpose, often you get worse result on the receiving (PTP Slave) side if it does not have a super stable clock itself (as mentioned in the video). However PTP in this case helps by correcting this in smaller steps way more often.
We tried a PTP only network at work beside the access network until the bean counters at work decided it was too expensive and put all the on-site devices back to irig.
@@messywill The problem with PTP is that it sounds sooo easy to implement as you basically only need one pricey Master Clock and then it is only LAN. Then you find out that if you do not connect your Slave straight to the Master you also need PTP aware switches which cost a multiple of a regular dumb switch. This is the point where most people initially considering PTP stop the project as they do not understand (or try) why stable network latency (or correction of timestamps) is so damn important. They think "I have PTP one both sides, so do your thing". If you only need to sync machines one to each other you can even get away with a software/hardware Master (depending on your timing requirements) even without internet at all. You still need those good switches though which again will break the budget of those responsible for project planning.
@Xcelsior86 pricey is an understatement. Most of the switches we use that are PTP capable are around $20k for a 24x100mbps port switch. There also environmental benefits to using these devices though, and mtbf is pretty good.
@@messywill if you look outside the IE market it's much better. You can get a Aruba 6300M doing 24-48 ports of 1G for like 2-3K. A bit more for a Cisco Cat 9300.
Makes sense, GPS is less precise at high latitudes, near the poles. But most of the Russian territory is at high latitudes so they needed and built a system that could cover that regions well and without too much precision loss. That's also why they came up with Molniya orbits for communication satellites.
That's cheap for high precision stuff. Want a Cesium clock? Used examples of 5071A run north of $10,000, often over $20,000. A new version of the successor, 5071B, is officially call us for the price (aka if you have to ask, you can't afford it), but is rumored to be about $50,000. And that's to say nothing of the sort of reference clocks NIST uses. Those are custom one off sorts of things that literally cannot be bought because nobody sells them (you can imagine the price tag). Now, these are GPS clocks, so can be alot cheaper, but even commercial competitors to this are quite expensive.
Higher end NICs also have extremely accurate built in clock oscillators as well so that packets can be synchronized to the nanosecond. It isnt usually its own time server you share out to the rest of the network (though some support this), but uses similar (but even higher end) clocks than this card uses to make sure packets are timestamped as accurately as possible. The accuracy of the clocks also matters in network streamed audio. Any time a packet is dropped because it isnt received in time, or put together in the wrong order due to a clock timestamp being wrong from jitter, or just bad network performance results in clicks, pops, or white noise type stuff in your audio.
Not really, unless for some reason you need instantaneous transmission the audio will be buffered, so network jitter is not an issue at all for audio, and those who want you to believe it are usually snake oil salesmen who want to sell 800$ "audiophile routers"
I'm a volunteer at a local observatory that has a couple of domes and telescopes and we've been having some time sync issues due to NTP not being super reliable for some reason, so this could really be useful, neat little device!
I have my own NTP server based on a PI3B running Chrony and a GPS Hat as the source and it's NTP is really accurate. Recently I've been playing with a secondary NTP server with a cheap £20 GPS receiver (with PPS) & it's time is just as accurate as the more expensive hat on the original
The presented module might fit your needs, but you could also give some other observatories a call (well or email) and ask them about how they solved it. While some use fancy expensive hardware some might know established cheap alternatives.
I'm from a small town in England called Rugby. When I was growing up we could always tell we were nearly home by the massive radio masts in the vicinity. They used to send time signals to automated clocks amongst other things. Its interesting to learn about time boards talking to gps.
Being able to measure time that precisely has some interesting implications related to general relativity - time itself actually speeds up by about 9.4ns/day per 1000m of differential elevation on earth. If you’re not at the same elevation as your reference clock, you’ll necessarily need to gain or shed nanoseconds here and there to stay in sync. This effect is especially strong for the GNSS satellites themselves - they end up doing some funny math to compensate. Now consider, for the US at least (although they do provide reference for UTC as well), the reference cesium clock is at about 1600m ASL in Boulder, CO. Higher than a good majority of the population (of course Colorado has towns above 3000m even)
That “…But this is just cool, and I like doing cool stuff like this and, I like sharing it with you.” Was really wholesome and suddenly I want more Jake short circuit
I have a stratum-1 time server at home based on a time standard from a 4G cell site running PTP rather than NTP, it also provides a very very stable 10MHz clock fed to SDRs and test equipment. The fun bit is the site survey to get an extra precise location pt use RTK 🤣🤣🤣
I have one of these on my desk. Using it synchronize multiple sensors for large-scale 3D printing (WAAM). It's pretty difficult to configure. And the NMEA hack is wild. Without an OXCO the Ubox ZED-F9T / Sparkfun controlled via serial from a Pi5 is working well at home.
Dealing with time is fun. I worked on a project where we used GPS to synchronize geographically distant channel hopping radios in sync on their hoping sequences. It's incredible stuff.
I build stuff like this for a living. It's not bad as an off-the-shelf product, but if you are willing to put in a bit of effort you could build something as good as this for a fraction of the cost.
honestly a raspberry pi with a GPS module would do the trick well enough that you don't need the extra bits. Just run an NTP server on it with a web server and then set all your devices NTP to the IP of the RPi.
One of the biggest inaccuracies is actually the GNSS signal. Every meter you are out in position is 3 ns of time error. That's why you combine it with the add on module crystal oscillator chip to help stabilise it as it drifts around.
That's also why you'd correct the position solution for high accuracy time solutions. With a fixed GPS station antenna (a good one, not a cheap one - something like a Topcon PN-A5, hard-mounted with good line-of-sight to all horizons), known-good cable that is time delay corrected in the GPS receiver (measured with a VNA or TDR) - a long term raw survey dumped into a file and sent to a correction service will give you a
This reminds me of when my physicist friend got a hold of two VERY precise clocks. One he kept at home (sea level), and the other one he brought with over the summer to 2000m altitude. They had drifted apart, just as much as Einstein said they would.
@@squishtomar1676 of course they did. You have misunderstood. Special relativity (or as we know it: "the easy theory of relativity") dictates that relative velocity affects time. General relativity (aka "the hard theory of relativity") adds gravitational potential to the time dilation equation. What we measured was probably the latter, but it was reduced a little bit by the former. I wasn't the one making the calculations, since I am a lowly musician.
The SiT5811 "Chip" that you're referring to here is actually a Crystal Oscillator (in this case an OCXO or an Oven Controlled Oscillator). These quartz "crystal" oscillate because of piezoelectric effect and the frequency of this movement is used to provide a reference for timing digital circuits. Very interesting technology Oscillators are literally everywhere and have a plethora of uses.
@@thetimelord Nope it's an OCXO. Check out their DATA sheet. Besides for an extremely high accuracy use-case like this, crystal oscillators are far more reliable and cheaper. GPS disciplined oscillators are quite literally unmatched when you want a cheap and reliable time source.
How do you know you have the right time. Run 3 of these.. 1 running - is it right 2 running - which one is right 3 running - the middle one is probably right. Lessons from a remote sys admin.
A U-Blox! I've messed around with those. It's surprisingly easy to mess around with GPS receivers, since they only need power and communicate by a serial text link. You only need to parse simple text strings to get your position and speed.
Its important in hyperconverged proxmox setup with ceph. Especially with high iops and situations when data needs to be changed frequently like in DBs. When you are building larger clusters not in same room this makes a lot of sense. Ofc this dont have any purpose in PC or 99.999% peoples watching this video :)
The amount of these things I've installed over my broadcasting years is a thing and I'm impressed that it's just a Raspberry Pi now. The ones I had were rack mounted with the GPS antenna out. Most of them were setup so satellite and our studios were in sync so commercials and triggers all went off at the same time. I also think our specialized units cost like $2000, so that thing is a steal. LOL
Once again asking why the bottom left cube in the logo light in the back is oriented the wrong way? Why and who changed it after the Dell Latitude 9440 video? Even more noticeable during the time lapse at 6:29.
In this application, it only matters that the rising and falling edge of the PPS signal is spaced exactly 1s apart or as close as possible to it. Good receivers measure this accuracy in single digit nanoseconds. The drift of the actual major time as reported by the GPS signal doesn't matter as much if you get that instead from more accurate stratum 1 sources like Cloudflare, as done in this video, and only use the PPS signal to "discipline" the clock pulse.
You should go visit the NIST-F1 clock, it's cesium fountain clock, a type of atomic clock, in the National Institute of Standards and Technology (NIST) in Boulder, Colorado, and serves as the United States' primary time and frequency standard.
A lot of new cars need a time server, especially if they have safety sensors, i.e., lots of cameras or radars. All the sensors are within a few uS, so they can sync the objects they see. In track testing, everything needs to sync within a few uS, so when tuning the sensors, they need to know where the car is and the objects it sees within a few mm at all times.
When I worked on a SDH/SONET network, clocking was super critical, bad things could happen if you lost your main clock source (usually GPS) With muxes drifting on their internal clocks and having justification errors, you'd then spend ages working your way around the whole synchronisation chain on a ring to find the culprit.
I understand Jake's wanting to do stuff fancy, like having every network component as its own device. Router provided NTP server not enough? Make a dedicated NTP server for the network.
I’ve had to build similar using only software. Managed to keep a latency within 2-5ms pretty easily between 10 computers. It was to sync video between multiple screens with panning shots across them. This was back in 2015 in Vancouver. Something like this would have made it easier for sure but we only needed accurate time in a relativistic sense. It really was crazy how much clocks drifted per day. When we were translating time to pixel offsets or arc seconds everything would be out of sync within a couple hours without our sync code and each computer would desync in a totally different way.
My previous job had to have one of these, a bit older model and company though. We ran a phone switch and it was required to keep exact time for logs. We had a GPS antenna on the roof of the ATT datacenter where we hosted the phone switch and our other servers.
Phones, especially back in the day used the RTC for sleep interrupts. Thats how you could set an alarm, turn off your phone and still have the alarm going off.
Another snag with GPS based time solutions (which I suspect the timecard mini does not worry about so much) is USB-serial jitter. Many devices GPS NMEA data will be fed via USB-serial which can have quite a significant jitter (10s of ms) if you're hitting the device with other I/O traffic at the same time. Just something to be aware of if you use USB-GPS devices which feed NMEA data.
I love these looks at tech that doesn't apply directly to me. It helps appreciate the sheer amount of invisible work that goes on to make the world we take for granted.
It is cool to learn about these niche IT products like these. I'd like to hear more about use cases and the professionals who benefit from these products.
These prices seem pretty steep to me, especially after seeing how much they rely on hobby-tier off the shelf hardware. If you only want to get time from GNSS, you can get that with pfSense/opnSense and e.g. a Garmin 18x LVC GPS for less than $100. Usually this will already achieve sub millisecond accuracy. 🤷♂
One important application where keeping GPS level time accuracy is stock markets, journaling transactions in a way consistent between actors. With high frequency trading microseconds count.
i swear to God sometimes it feels like we are living in a matrix. LITERALLY yesterday I noticed that my Windows clock was 2-3 minutes ahead of my phone and work laptop. And there is option in Windows 11's Adjust Time section, to manually "Sync Now". And i was worried that something is defecting in my Motherboard, because I am aware of the internal clock concept, but i was under the impression that it would be accurate perpetually. Never even came to me that it could be drifting. AND just now seeing this video makes me question reality.
I once had a boss that basically filled out the paperwork and paid the fees to become a cellular phone provider company so that he could get access to the US atomic clocks and sync two separate computer systems thousands of miles apart with < 1ms drift. It was only marginally more expensive than this card, but way more baroque.
even using the Standard ntp impelmentations I got about 30 ns sync for a local group as long as they share a switch. 10 is pretty impressive though. also don't forget that your clocks need about 1-2 days to actually be properly synced.
I feel like this could be fun for those homelabbers with services connected to external resources like remote VMs/docker containers hosted at rentable datacenters like Hetzner
As a kid I build a Beowulf cluster from 386 machines, all booted with a floppy that I swapped between machines to start them up! You don't want to know how many times I ran into clock skew and had absolutely no clue what to do about it :D (I did this 25ish years ago)
You don't even need a local time source. Set multiple NTP pool servers and you'll be within a few ms after everything has enough time to stabilize. I run a Nutanix cluster like this, and it's fine.
Would like to see you guys set up an IEEE 1588 PTP network for time synchronization and give a detailed explanation of how it works. It’s pretty sweet.
For a long time, I’ve been using a Mikrotik router model RB3011, and currently the CCR2116, with a USB-connected GPS antenna. Thanks to this, the router functions as an NTP server. Besides, wouldn’t it be cheaper to buy something like an RPi3 and connect a USB GPS antenna to it? The price for the presented device is a bit over the top.
Jake didn't really explain why this thing exists, so I'll try from memory: Cards like this (or similar add-in cards with small atomic clocks) are part of the PTP initiative, the Precision Time Protocol. It's a concept Meta came up with and open sourced if I remember correctly. The idea is that in a datacenter, when you send packets between servers, they have to have timestamps and constantly negotiate time, because the realtime clocks in the individual machines are not very precise. PTP syncs all the clocks in the datacenter to within nanoseconds of each other, and syncs constantly, not just on boot or once a day or so. That means you can drop timing-related bits from the network packets, which leaves more space for actually useful data. So this tech reduces network overhead and increases usable bandwidth in datacenters. That's the main point.
PTP predates Facebook/Meta by a few years (first version published in 2002; Meta was founded in 2004). For the majority of use cases (even in datacenters) you don't need the precision it provides and can rely on NTP (NTP can also be run frequently if configured as such; the reason it's not done often is because time drift is not as wild as Jake said in the video).
And for dropping timing data from packets - no, that's not a thing. Most protocols use sequence numbers that are relative/based on the number of packets sent since connection establishment. But for things that do care about timing, timestamp data is explicitly included (for example, in massive transactional systems such as Spanner, all transactions have explicit timestamps that are used to decide transaction ordering).
Hey I think you are missing out something here, PTP or rather gPTP AS2020 spec makes it a lot easier for a suite of layer 2 protocols to work that collectively fall under the TSN group. I actively conduct research on TSN and the implications are pretty interesting. NTP is good for ensuring sync with some drift but fails to keep up in scenarios where you need expected latency requiremnets to kick in and work (even when for some time the time sync packet channel is being used for high speed traffic). Look into stuff like Qbv and Qav or Qcb. All theese QoS services have seen a lot of improvement with gPTP and seen wider range of applications (where NTP is just not feasible) like Automotive Ethernet and Mission Critical Software.
@@weirdguybr Open Timecard was a Meta idea, PTP is older. Open Timecard is part of the Time Appliance Project under the OpenCompute umbrella. Really should have worded that better.
Do I understand correctly that 64 bits for timestamp isn't that much of data but it adds up really really fast?
@@pavuk357yes, it's a tiny amount that could add up, but very few applications need timestamps to begin with and there are other places where you can save -
Take HTTPs for example - you waste more bits with all the headers than with one single timestamp.
Oh that seems like a neat little thing to mess around wi….. $761?!?
Ok. Never mind.
It's a very accurate clock, there is nothing to mess around. Unless you have a large amount of servers that need very precise time to enable high speed low latency data transfer this is useless for you
@@marcogenovesi8570 nah this be cool toy to play with
Time is expensive
You only need a GPS with a PPS output to a serial port. I managed to make something of the same accuracy as this for about $30 in total.
@@luis449bp not if you make it yourself!
im still gonna be late idc
Now you can be late in miliseconds rather than minutes.😂
Time was invented to screw the working man over
but your computer won't be (which is what matters)!
666 likes
Now your employer can know how late you are to the milliseconds rather than minutes.
hearing jake say "gock" gave me whiplash every single time
same
How should it be pronounced?
the proper term is "femis"
gock on ltt 😮😮
LMG quickly learning the meaning of "gock" lol
Aw yeah I love gock
An individual of taste. Fantastic.
Us trans women are quite familiar
@@Banzai8th as am I
In an old job, we had a system that needed accurate time and got it via GPS. Only issue we had was that we installed it in a data centre which was very secure and as a result it couldn’t see any satellites. Luckily the system had two power supply’s, so we connected one to a unplugged but charged UPS, wheeled it outside to get a signal, and then ran it back inside into our rack power so it didn’t turn off and kept time. The entire time the UPS was beeping and it must have looked like a computer heist to an outsider due to us running through corridors trying to get it to mains power before the UPS ran out.
You may have got time out of it but if it can't see satellites then its dead reckoning and no more accurate than a pc clock
just taking the servers out for a walk
@@thefatmoop Yeah agreed, the drift wasn’t great but we only needed it for a couple of weeks until the centre provider could source us a cable run out the building for the receiver.
@@thefatmoop a lot of time sync devices like that have also a high precision quartz clock too, which is much more precise than the default clock source in a PC
Feed the PPS in via fibre, all good!
WAN show viewers be like:
bruh!
Wan is happening early
They'd need like 20 of these.
WAN was early this week. Clearly this thing works too well
Glad this company gave us gock representation 😊
Underrated comment
came here to post this
:)
FOR REALLLLLLLLLLLLLLL
🥴
So, I'm a watchmaker as my profession, and we sometimes have people complaining that their MECHANICAL watch is not keeping accurate enough time. I now have something to point to and say "If you want perfectly accurate timekeeping get one of these" 😆
Or tell them to get a digital watch that sets time off of the RF broadcast from an atomic clock in Boulder, CO (at least in the US). I'm not aware of any mechanical watch that self-sets like this, although I'm sure it's certainly possible to do...
Honestly modern smartwatches are probably more precise than even the finest mechanical watch, simply because they can sync time with the phone they talk to, or with GPS, and additionally off of cellular (LTE and 5G include timing data in the broadcast signal and you *may* be able to pull that in without even having a SIM)
So I’m genuinely curious and not trying to belittle your profession/job title, but in the timepiece industry, at what point does someone qualify to be called a watchmaker vs just an employee at a watch factory? Do watchmakers machine their own parts vs just assemble parts? Like I said, I’m genuinely curious as I don’t know anything about that industry.
My mechanical watch keeps time perfectly fine, only +/- a few seconds every month, which can't even add up over time anyways because I reset the watch every couple months to update the month counter properly (it assumes 31 days in each month). If the accuracy is bad enough to where it's noticeable enough for people to complain it might not be their fault; have you considered that your watches are not performing their sole task as well as they should?
@@catgodswim A chronometer is the highest standard for mechanical watch accuracy; it requires a watch to lose no more than four seconds per, or gain no more than six seconds per day, averaged over a ten-day testing period. If your watch is within a few seconds after a month, it's performing far better than most high-precision mechanical watches.
@@jackiecs8190 Maybe my watch is just cracked and I never noticed. Right now it's only around 12ish seconds ahead - after likely around 55 days of not having been reset.
HE CALLED IT THE WHAT
G.O.C. --The Global Occult Coalition
...totally ignoring the joke
gock😋😋
Welcome to the rabbit hole of precision time keeping. I used to work in signal analysis and kind of found it fun to collect various clock sources from oven-controlled oscillators to cesium controlled ones. More or less just a hobby of seeing what odd vintage precision gear you could find and revive.
From someone who has written embedded PTP Master/Slave from scratch (besides other network protocols) in my job, losing internet is actually not at all a headache for a large part of customers using this high precision time sync stuff.
This gets used to sync conveyor belts and other precision machines in a dedicated network, sometimes using a dedicated NIC/port, switching off every other traffic except for the PTP packets (to stabilize network stack processing jitter with mid-tier ARM devices) or using an entire second network just for the PTP connections.
Also, while it is good to know that you are "n" nanoseconds precise on a hardware dedicated for this purpose, often you get worse result on the receiving (PTP Slave) side if it does not have a super stable clock itself (as mentioned in the video). However PTP in this case helps by correcting this in smaller steps way more often.
We tried a PTP only network at work beside the access network until the bean counters at work decided it was too expensive and put all the on-site devices back to irig.
@@messywill The problem with PTP is that it sounds sooo easy to implement as you basically only need one pricey Master Clock and then it is only LAN. Then you find out that if you do not connect your Slave straight to the Master you also need PTP aware switches which cost a multiple of a regular dumb switch. This is the point where most people initially considering PTP stop the project as they do not understand (or try) why stable network latency (or correction of timestamps) is so damn important. They think "I have PTP one both sides, so do your thing".
If you only need to sync machines one to each other you can even get away with a software/hardware Master (depending on your timing requirements) even without internet at all. You still need those good switches though which again will break the budget of those responsible for project planning.
@Xcelsior86 pricey is an understatement. Most of the switches we use that are PTP capable are around $20k for a 24x100mbps port switch. There also environmental benefits to using these devices though, and mtbf is pretty good.
@@messywill if you look outside the IE market it's much better. You can get a Aruba 6300M doing 24-48 ports of 1G for like 2-3K. A bit more for a Cisco Cat 9300.
@@krikkits do they do fanless -40 to +85 degree celcius and have conformal coating?
Nice gock bro
Diabolical Jake out here playing with time
Jacob be like, I’ve never had my hands on a gock smh
Nice gock
0:30 thats what I call mine at least
Same lol
hell yeah sis
uwu
Same lmao
LMAO
Oh my God someone actually mentioned a GNSS network other than GPS, it's a miracle.
galileo best
My very old Sony Ericsson mobile phone had support for both GPS and glonass, curiously glonass worked better in northern Europe.
Makes sense, GPS is less precise at high latitudes, near the poles.
But most of the Russian territory is at high latitudes so they needed and built a system that could cover that regions well and without too much precision loss.
That's also why they came up with Molniya orbits for communication satellites.
Yeah, but he mentioned glonass and precision in the same sentence.
That $700 price tag gives whole new meaning to the phrase "time is money"
That's cheap for high precision stuff. Want a Cesium clock? Used examples of 5071A run north of $10,000, often over $20,000. A new version of the successor, 5071B, is officially call us for the price (aka if you have to ask, you can't afford it), but is rumored to be about $50,000. And that's to say nothing of the sort of reference clocks NIST uses. Those are custom one off sorts of things that literally cannot be bought because nobody sells them (you can imagine the price tag). Now, these are GPS clocks, so can be alot cheaper, but even commercial competitors to this are quite expensive.
@@ccoder4953 you ruined the joke bud
Time-servers. The most Jake thing ever.
Jake is a Time Wizard.
It's TIME TO
Higher end NICs also have extremely accurate built in clock oscillators as well so that packets can be synchronized to the nanosecond. It isnt usually its own time server you share out to the rest of the network (though some support this), but uses similar (but even higher end) clocks than this card uses to make sure packets are timestamped as accurately as possible.
The accuracy of the clocks also matters in network streamed audio. Any time a packet is dropped because it isnt received in time, or put together in the wrong order due to a clock timestamp being wrong from jitter, or just bad network performance results in clicks, pops, or white noise type stuff in your audio.
Not really, unless for some reason you need instantaneous transmission the audio will be buffered, so network jitter is not an issue at all for audio, and those who want you to believe it are usually snake oil salesmen who want to sell 800$ "audiophile routers"
GOCK 🤯
I'm a volunteer at a local observatory that has a couple of domes and telescopes and we've been having some time sync issues due to NTP not being super reliable for some reason, so this could really be useful, neat little device!
I have my own NTP server based on a PI3B running Chrony and a GPS Hat as the source and it's NTP is really accurate. Recently I've been playing with a secondary NTP server with a cheap £20 GPS receiver (with PPS) & it's time is just as accurate as the more expensive hat on the original
Timebeat guys do work with scientific research bodies - you should talk to them
The presented module might fit your needs, but you could also give some other observatories a call (well or email) and ask them about how they solved it. While some use fancy expensive hardware some might know established cheap alternatives.
as a great man once said, "Time is an illusion. Lunchtime doubly so"
Yes, but what about second Lunchtime?
@@ReclaimerTyphoonI don't think he knows about Second Lunchtime, Pip…
I'm from a small town in England called Rugby. When I was growing up we could always tell we were nearly home by the massive radio masts in the vicinity. They used to send time signals to automated clocks amongst other things. Its interesting to learn about time boards talking to gps.
Being able to measure time that precisely has some interesting implications related to general relativity - time itself actually speeds up by about 9.4ns/day per 1000m of differential elevation on earth. If you’re not at the same elevation as your reference clock, you’ll necessarily need to gain or shed nanoseconds here and there to stay in sync.
This effect is especially strong for the GNSS satellites themselves - they end up doing some funny math to compensate.
Now consider, for the US at least (although they do provide reference for UTC as well), the reference cesium clock is at about 1600m ASL in Boulder, CO. Higher than a good majority of the population (of course Colorado has towns above 3000m even)
They shouldve linked to LTTs "Why is this PCIe Card RADIOACTIVE?" video since it goes more in depth on why synchronised time is important.
That “…But this is just cool, and I like doing cool stuff like this and, I like sharing it with you.” Was really wholesome and suddenly I want more Jake short circuit
Completely useless for the vast majority of us but pretty dang cool. Show us more niche stuff!
That bug is the epitomy of modern software and hardware. The thing is supposed to do a thing and it does not because a bug. A classic.
I have a stratum-1 time server at home based on a time standard from a 4G cell site running PTP rather than NTP, it also provides a very very stable 10MHz clock fed to SDRs and test equipment. The fun bit is the site survey to get an extra precise location pt use RTK 🤣🤣🤣
What's rtk
I have one of these on my desk. Using it synchronize multiple sensors for large-scale 3D printing (WAAM). It's pretty difficult to configure. And the NMEA hack is wild.
Without an OXCO the Ubox ZED-F9T / Sparkfun controlled via serial from a Pi5 is working well at home.
Dealing with time is fun. I worked on a project where we used GPS to synchronize geographically distant channel hopping radios in sync on their hoping sequences. It's incredible stuff.
Gok Wan was a famous personality in the UK a decade ago....which makes it even funnier
Time sync is really one of those hidden heros to the modern computing era.
I build stuff like this for a living. It's not bad as an off-the-shelf product, but if you are willing to put in a bit of effort you could build something as good as this for a fraction of the cost.
Im interested doing a diy project like that, any sources you can recommend?
Yeah but I gotta pay the smoothbrain tax
Give us a simple tutorial or hardware list.
damn bro say please
honestly a raspberry pi with a GPS module would do the trick well enough that you don't need the extra bits. Just run an NTP server on it with a web server and then set all your devices NTP to the IP of the RPi.
One of the biggest inaccuracies is actually the GNSS signal. Every meter you are out in position is 3 ns of time error. That's why you combine it with the add on module crystal oscillator chip to help stabilise it as it drifts around.
That's also why you'd correct the position solution for high accuracy time solutions. With a fixed GPS station antenna (a good one, not a cheap one - something like a Topcon PN-A5, hard-mounted with good line-of-sight to all horizons), known-good cable that is time delay corrected in the GPS receiver (measured with a VNA or TDR) - a long term raw survey dumped into a file and sent to a correction service will give you a
This reminds me of when my physicist friend got a hold of two VERY precise clocks. One he kept at home (sea level), and the other one he brought with over the summer to 2000m altitude.
They had drifted apart, just as much as Einstein said they would.
Did they really? Only acceleration affects time
@@squishtomar1676 of course they did. You have misunderstood.
Special relativity (or as we know it: "the easy theory of relativity") dictates that relative velocity affects time. General relativity (aka "the hard theory of relativity") adds gravitational potential to the time dilation equation.
What we measured was probably the latter, but it was reduced a little bit by the former. I wasn't the one making the calculations, since I am a lowly musician.
Gock
The SiT5811 "Chip" that you're referring to here is actually a Crystal Oscillator (in this case an OCXO or an Oven Controlled Oscillator). These quartz "crystal" oscillate because of piezoelectric effect and the frequency of this movement is used to provide a reference for timing digital circuits. Very interesting technology
Oscillators are literally everywhere and have a plethora of uses.
Back in the day, you had to swap out the oscillator when upgrading your cpu! 😀
Actually, it's a MEMS oscillator.... no crystal... 🙂
@@thetimelord Nope it's an OCXO. Check out their DATA sheet. Besides for an extremely high accuracy use-case like this, crystal oscillators are far more reliable and cheaper.
GPS disciplined oscillators are quite literally unmatched when you want a cheap and reliable time source.
I checked, it seems the SiT811 module is available both in MEMS or OCXO config. The one in the video is definitely an OCXO package though.
thank you good doctor!
"Time is an illusion. Lunchtime doubly so"
Gock gock gock 💀
an actual neat product showcase! you guys need more of those
How do you know you have the right time. Run 3 of these..
1 running - is it right
2 running - which one is right
3 running - the middle one is probably right.
Lessons from a remote sys admin.
Perfect for an employee punch card system-now you can pay them by the nanosecond!
gock 💀
hawk tuah
That's a small gock ngl
@@vulpesfennex8827 Above average imo, they're not usually very big at all.
I was so not mentally prepared for this jesus
@@spoon1555speak for yourself buddy
A U-Blox! I've messed around with those. It's surprisingly easy to mess around with GPS receivers, since they only need power and communicate by a serial text link. You only need to parse simple text strings to get your position and speed.
MORE OFTHIS... This was so entertaining. Jake was truly like... a kid in a candy store...
I like these types of niche product reviews rather than everyday tech! Great work!
Its important in hyperconverged proxmox setup with ceph. Especially with high iops and situations when data needs to be changed frequently like in DBs. When you are building larger clusters not in same room this makes a lot of sense. Ofc this dont have any purpose in PC or 99.999% peoples watching this video :)
i just love jake going absolute nerd mode
The amount of these things I've installed over my broadcasting years is a thing and I'm impressed that it's just a Raspberry Pi now. The ones I had were rack mounted with the GPS antenna out. Most of them were setup so satellite and our studios were in sync so commercials and triggers all went off at the same time. I also think our specialized units cost like $2000, so that thing is a steal. LOL
Man I love these different gadget vids you do!
1PPS is common on most space applications that are under the GPS constellation altitude. Glad you guys can use to keep your steam server linked up.
Gock has a very different meaning
Definitely getting the Gokk
Once again asking why the bottom left cube in the logo light in the back is oriented the wrong way? Why and who changed it after the Dell Latitude 9440 video? Even more noticeable during the time lapse at 6:29.
Jake, next time with Jake: today, we are going to build an atomic clock!
Linus please give Jake / Emily / Dan a Linux / infrastructure / cool stuff channel PLEASE
Just watch level one techs
@@Matt-oq4jq I just like Jake's zoomer energy. I love Wendall but Jake is more relatable.
Emily does not work at LMG anymore.
@@VFPn96kQT oh really? I didn't know that
@@VFPn96kQT NOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO WHY DIDNT I HEAR ABOUT THIS :((((((((
Super cool video, please keep more of these coming!
You’re saying not everyone has a stratum 1 time server?
Just a note but GPS is only rated to be accurate to 1ms. In practice I found it was actually closer to 0.1ms, but that's what it's rated for.
Which bands are we talking about?
In this application, it only matters that the rising and falling edge of the PPS signal is spaced exactly 1s apart or as close as possible to it. Good receivers measure this accuracy in single digit nanoseconds. The drift of the actual major time as reported by the GPS signal doesn't matter as much if you get that instead from more accurate stratum 1 sources like Cloudflare, as done in this video, and only use the PPS signal to "discipline" the clock pulse.
I love that Jake always finds a solution to go overkill
You should go visit the NIST-F1 clock, it's cesium fountain clock, a type of atomic clock, in the National Institute of Standards and Technology (NIST) in Boulder, Colorado, and serves as the United States' primary time and frequency standard.
Great to see the OCP timecard making it's way to full commercial builds!
A lot of new cars need a time server, especially if they have safety sensors, i.e., lots of cameras or radars. All the sensors are within a few uS, so they can sync the objects they see.
In track testing, everything needs to sync within a few uS, so when tuning the sensors, they need to know where the car is and the objects it sees within a few mm at all times.
When I worked on a SDH/SONET network, clocking was super critical, bad things could happen if you lost your main clock source (usually GPS) With muxes drifting on their internal clocks and having justification errors, you'd then spend ages working your way around the whole synchronisation chain on a ring to find the culprit.
I understand Jake's wanting to do stuff fancy, like having every network component as its own device. Router provided NTP server not enough? Make a dedicated NTP server for the network.
I’ve had to build similar using only software. Managed to keep a latency within 2-5ms pretty easily between 10 computers. It was to sync video between multiple screens with panning shots across them. This was back in 2015 in Vancouver. Something like this would have made it easier for sure but we only needed accurate time in a relativistic sense.
It really was crazy how much clocks drifted per day. When we were translating time to pixel offsets or arc seconds everything would be out of sync within a couple hours without our sync code and each computer would desync in a totally different way.
My previous job had to have one of these, a bit older model and company though. We ran a phone switch and it was required to keep exact time for logs. We had a GPS antenna on the roof of the ATT datacenter where we hosted the phone switch and our other servers.
Phones, especially back in the day used the RTC for sleep interrupts. Thats how you could set an alarm, turn off your phone and still have the alarm going off.
Another snag with GPS based time solutions (which I suspect the timecard mini does not worry about so much) is USB-serial jitter. Many devices GPS NMEA data will be fed via USB-serial which can have quite a significant jitter (10s of ms) if you're hitting the device with other I/O traffic at the same time. Just something to be aware of if you use USB-GPS devices which feed NMEA data.
I really hope it's pronounced like Gock. Please, dear gods, let there be some femme memeing hardcore up in there.
I love these looks at tech that doesn't apply directly to me. It helps appreciate the sheer amount of invisible work that goes on to make the world we take for granted.
It is cool to learn about these niche IT products like these. I'd like to hear more about use cases and the professionals who benefit from these products.
Raspberry Pi ... accurate timekeeping ... those are kind of mutually exclusive.
You should be using a Caesium time source if you want accuracy.
These prices seem pretty steep to me, especially after seeing how much they rely on hobby-tier off the shelf hardware. If you only want to get time from GNSS, you can get that with pfSense/opnSense and e.g. a Garmin 18x LVC GPS for less than $100. Usually this will already achieve sub millisecond accuracy. 🤷♂
One important application where keeping GPS level time accuracy is stock markets, journaling transactions in a way consistent between actors. With high frequency trading microseconds count.
i swear to God sometimes it feels like we are living in a matrix. LITERALLY yesterday I noticed that my Windows clock was 2-3 minutes ahead of my phone and work laptop. And there is option in Windows 11's Adjust Time section, to manually "Sync Now". And i was worried that something is defecting in my Motherboard, because I am aware of the internal clock concept, but i was under the impression that it would be accurate perpetually. Never even came to me that it could be drifting.
AND just now seeing this video makes me question reality.
I once had a boss that basically filled out the paperwork and paid the fees to become a cellular phone provider company so that he could get access to the US atomic clocks and sync two separate computer systems thousands of miles apart with < 1ms drift.
It was only marginally more expensive than this card, but way more baroque.
even using the Standard ntp impelmentations I got about 30 ns sync for a local group as long as they share a switch. 10 is pretty impressive though.
also don't forget that your clocks need about 1-2 days to actually be properly synced.
I feel like this could be fun for those homelabbers with services connected to external resources like remote VMs/docker containers hosted at rentable datacenters like Hetzner
I used to work on phone equipment and some would have a stratum 3 time card. The point of those were to maintain the T1 sync.
I see you guys got that redbull sponser. Nice!
I came here to say exactly this 😀
Cheer~~a card used to record an employee's starting and quitting times, usually stamped by a time clock.😊
For reference, the most accurate wristwatch is +/-1 seconds per year.
As a kid I build a Beowulf cluster from 386 machines, all booted with a floppy that I swapped between machines to start them up!
You don't want to know how many times I ran into clock skew and had absolutely no clue what to do about it :D
(I did this 25ish years ago)
It's about time we saw some gock love given the amount of programmers in this community.
You don't even need a local time source. Set multiple NTP pool servers and you'll be within a few ms after everything has enough time to stabilize. I run a Nutanix cluster like this, and it's fine.
I just love this stuff i cant wait till it is just built into our motherboards
I think this is the coolest 😎❤ SETI please 🥺 I had a killer Xeno during the NapYears
Would like to see you guys set up an IEEE 1588 PTP network for time synchronization and give a detailed explanation of how it works. It’s pretty sweet.
Jeff Geerling went over this twice a few years ago. glad to see things are spreading.
Something interesting, GPS time is 18 sec ahead of UTC time. Because of GPS leap seconds.
Jake: "I really want to buy with this unnecessary thing to solve a problem I already solved"
Taran: "Fine just make a video on it I guess"
For a long time, I’ve been using a Mikrotik router model RB3011, and currently the CCR2116, with a USB-connected GPS antenna. Thanks to this, the router functions as an NTP server. Besides, wouldn’t it be cheaper to buy something like an RPi3 and connect a USB GPS antenna to it? The price for the presented device is a bit over the top.
0:39 "The G0KK One and you can probably guess what this one is called..."
Me: The G0KK Two!
"The time card mini."
Me: .... oh.
When you think of the grief people went through to get accurate time to permit the measurement of Longitude... 10ns is insane!
Disappointed this didn't go into any depth regarding Precision Time Protocol
Surely this won’t cause any issues in the future. We all know how well LLT does with installing over complicated components to their infrastructure