Why some cables cost so much - The Adam Savage video.

[Blue4130]

6 minutes ago, MarcWolfe said:

Don't worry too much about it; it's an over complicated method anyway.  

The most modern HDMI standards meet or exceed 40 Gbit. HDMI has a reference clock; don't need to re-time that.

If you exceed the specified length you do.

[MarcWolfe]

5 minutes ago, Blue4130 said:

If you exceed the specified length you do.

What do you think the reference clock is for? 

[Blue4130]

7 minutes ago, MarcWolfe said:

What do you think the reference clock is for? 

Standard length cables.

[MarcWolfe]

7 minutes ago, Blue4130 said:

Standard length cables.

Right, you can't just boost a reference clock signal /s

[Blue4130]

1 minute ago, MarcWolfe said:

Right, you can't just boost a reference clock signal /s

Ok, so how do you boost the signal?

[MarcWolfe]

14 minutes ago, Blue4130 said:

Ok, so how do you boost the signal?

Basic ass FETs.

[starsmine]

14 minutes ago, MarcWolfe said:

Basic ass FETs.

In what arrangement would these Fets be? 

[MarcWolfe]

3 minutes ago, starsmine said:

In what arrangement would these Fets be? 

One that results in a higher voltage, that get's lowered at the other end. You want me to try to design a good PCB? better throw a wad of money at me.

[starsmine]

3 minutes ago, MarcWolfe said:

One that results in a higher voltage, that get's lowered at the other end. You want me to try to design a good PCB? better throw a wad of money at me.

you said fet.... mosfets are current sources, not voltage.

I showed you the chips needed here (the titan ridge part from intel) and the ones used for this use case (the cortex m0 to be used as a microcontroller)

[MarcWolfe]

3 minutes ago, starsmine said:

you said fet.... mosfets are constant current sources, not constant voltage.

... I think I've forgotten more than you know right now. 

Logic level MOSFETs come to mind. Easy to find those that are triggered by 5 volts and switch 12 volts on and off.

[MarcWolfe]

And then there's the simple approach to optical. Don't mux signals and just use a fiber per lane/twisted pair. About as complicated as having multiples of these for one end

And then the photosensors for the other side. Likely couldn't use the cheapest parts, just stuff with a high enough slew rate. And you can just extend the power lines alongside the optical. 

 

[Distinctly Average]

12 hours ago, MarcWolfe said:

And then there's the simple approach to optical. Don't mux signals and just use a fiber per lane/twisted pair. About as complicated as having multiples of these for one end

And then the photosensors for the other side. Likely couldn't use the cheapest parts, just stuff with a high enough slew rate. And you can just extend the power lines alongside the optical. 

 

I cannot think of any hybrid cables in mainstream use yet. Mixing power and optical probably has limited situations where it would be useful.

 

As for FETs, I have fond memories of buying one from Tandy with my pocket money in 1980. It was for a project in EE magazine but I cannot remember what. Probably an amplifier with a voice changer facility as I was all into that back then.

[starsmine]

13 hours ago, MarcWolfe said:

And then there's the simple approach to optical. Don't mux signals and just use a fiber per lane/twisted pair. About as complicated as having multiples of these for one end

And then the photosensors for the other side. Likely couldn't use the cheapest parts, just stuff with a high enough slew rate. And you can just extend the power lines alongside the optical. 

 

 Thunderbolt 1 wanted to be optical. Its not like this wasnt tried. 
it was called Light Peak. 

Also that has issues with the cables needing to be high quality glass. 
SPDIF is SLOW its 6Mbits per second

for optical to work at 40Gb you need clean fiber and you need to protect both ends, its not good for a solution wher eyou are constantly connecting or disconnecting the device, it also provides ZERO power. 
 

[Distinctly Average]

15 minutes ago, starsmine said:

 Thunderbolt 1 wanted to be optical. Its not like this wasnt tried. 
it was called Light Peak. 

Also that has issues with the cables needing to be high quality glass. 
SPDIF is SLOW its 6Mbits per second

for optical to work at 40Gb you need clean fiber and you need to protect both ends, its not good for a solution wher eyou are constantly connecting or disconnecting the device, it also provides ZERO power. 
 

Why? The optical can be converted in the plug body. This gives you an electrical connection at source and destination, but a sealed optical system that would not be prone to dirt. Not a single core I grant you, but a 19 fibre link ran at 1.7 petabits over 41miles recently. SPDIF was running on a slow protocol using an old laser (in the TOSLINK implementation, SPDIF was the protocol which could also be used over a coaxial connection) both of which limited it, not the quality of the glass. Fibre design has come along lots since 1983 when Toshiba launched TOSLINK

[starsmine]

4 minutes ago, Distinctly Average said:

Why? The optical can be converted in the plug body. This gives you an electrical connection at source and destination, but a sealed optical system that would not be prone to dirt. Not a single core I grant you, but a 19 fibre link ran at 1.7 petabits over 41miles recently. SPDIF was running on a slow protocol using an old laser (in the TOSLINK implementation, SPDIF was the protocol which could also be used over a coaxial connection) both of which limited it, not the quality of the glass. Fibre design has come along lots since 1983 when Toshiba launched TOSLINK

Im not saying fibre hasnt come a long way. 
but its exactly that long way that would be problematic for this use case. a way to seal the connection ends is not a simple or cheap task. 


That 1.7 petabit link is with a permanent installation so the problems im refering to do not apply. 
 

[Distinctly Average]

26 minutes ago, starsmine said:

Im not saying fibre hasnt come a long way. 
but its exactly that long way that would be problematic for this use case. a way to seal the connection ends is not a simple or cheap task. 


That 1.7 petabit link is with a permanent installation so the problems im refering to do not apply. 
 

I’m not saying it is, just that it is possible. It has actually been done, not in USB, but in the mainframe world. Think of a fibre cable with a built in SFP, wrapped in copper cables for power. 
 

In a data centre, we often pull the SFP rather than disconnecting the fibre from it. It can be easier and the risk of contamination is lower. Also because those annoying plastic clips sometimes end up in the wrong place. Anyone who has worked in a data centre will know exactly what I am referring to.

[MarcWolfe]

14 minutes ago, Distinctly Average said:

It has actually been done

I'm not aware of an option that hasn't used muxxing though.

[starsmine]

32 minutes ago, MarcWolfe said:

I'm not aware of an option that hasn't used muxxing though.

 

48 minutes ago, Distinctly Average said:

I’m not saying it is, just that it is possible. It has actually been done, not in USB, but in the mainframe world.

You misunderstand the issues I'm bringing up. 
yes, these are used in data centers.
Yes, it's configurable and swappable. 

but people are not just tossing Infiniband cables in their backpacks and tossing that around. and expecting it to work. you have to be VERY careful with the endpoints of those cables, relative to the abuse a thunderbolt or USB connector receives. 

Consumer and Datacenter are two different markets with two different needs.

But InfiniBand is why Intel was designing light points the way they were back then. 

Im not saying it cant be done, but arguing that it would be cheaper than thunderbolt 4 over USB C for a consumer is a different argument altogether. 
and again, POWER. Like I have said, you are misunderstanding the issues. Not to say you cant slap a copper cable next to the fiber. Hell, InfiniBand can use copper for its slower connections. but it can't hit >40Gbit without using a 12-lane endpoint and a maximum length of 2.5Meters, which is a relatively massive connector. 

Not that that cant be engineered around to make something smaller... which people have... its called Thunderbolt 4. 

[MarcWolfe]

9 minutes ago, starsmine said:

arguing that it would be cheaper than

Well, one point I'll bring up is how cheap ARM can be. 

And that's after the consumer markup without bulk pricing. To act like Apple is selling it for little more than they need to is a slap in the face to people who know better.

[starsmine]

7 minutes ago, MarcWolfe said:

Well, one point I'll bring up is how cheap ARM can be. 

And that's after the consumer markup without bulk pricing. To act like Apple is selling it for little more than they need to is a slap in the face to people who know better.

Cool, you have 1 of two microcontrollers needed, now where is the rest of the board and circuitry? 

you are arguing points not being made. We know apple tax is being applied to what is a 70USD cable.
But you also know it will work, and they are not selling a scam cable. 

The CCG3PA2 is used for the USB3 complience and the titandridge is there for thunderbolt. 

https://www.mouser.com/c/?q=CCG3PA2 and its cheaper. 

[MarcWolfe]

8 minutes ago, starsmine said:

Cool, you have 1 of two microcontrollers needed, now where is the rest of the board and circuitry? 

you are arguing points not being made. We know apple tax is being applied to what is a 70USD cable.

Apparently Adam and his pals don't.

The majority of circuitry needed gets integrated into the ARM IC. Just like Arduinos tend to have ADCs integrated, and other somewhat complex timing circuits. Doesn't need to be a very competent ARM core either. That SAMD21 for example can do some fancy things without even engaging the ARM core's compute, or the main loop code, or interrupts; it can just sit there and fuck off after setup. Also doesn't even need to be ARM; that's just a dominant cheap option.

[Distinctly Average]

21 minutes ago, starsmine said:

 

You misunderstand the issues I'm bringing up. 
yes, these are used in data centers.
Yes, it's configurable and swappable. 

but people are not just tossing Infiniband cables in their backpacks and tossing that around. and expecting it to work. you have to be VERY careful with the endpoints of those cables, relative to the abuse a thunderbolt or USB connector receives. 

Consumer and Datacenter are two different markets with two different needs.

But InfiniBand is why Intel was designing light points the way they were back then. 

Im not saying it cant be done, but arguing that it would be cheaper than thunderbolt 4 over USB C for a consumer is a different argument altogether. 
and again, POWER. Like I have said, you are misunderstanding the issues. Not to say you cant slap a copper cable next to the fiber. Hell, InfiniBand can use copper for its slower connections. but it can't hit >40Gbit without using a 12-lane endpoint and a maximum length of 2.5Meters, which is a relatively massive connector. 

Not that that cant be engineered around to make something smaller... which people have... its called Thunderbolt 4. 

I was not arguing about price at all, just that technically it could and has been done. Having said that, a pair of 10G SFPs can be had for under £20 retail so probably cost £2.50 each to produce. Add in a little custom electronics and put the thing in a more user friendly form factor and it looks like it could be done at a reasonable cost. 
 

Still not saying it should be done though. And ai actually think Apple have done a great job on that cable despite it being subject to Apple tax. As I also said, I think Adam and the team did this video more to show off their scanner tech and the used it could be put to than to show the cables. Last time Adam worked with that company it was to scan a model mould that was otherwise too hard to reproduce.

[starsmine]

Just now, MarcWolfe said:

Apparently Adam and his pals don't.

The majority of circuitry needed gets integrated into the ARM IC. Just like Arduinos tend to have ADCs integrated, and other somewhat complex timing circuits. Doesn't need to be a very competent ARM core either. That SAMD21 for example can do some fancy things without even engaging the ARM core's compute. Also doesn't even need to be ARM; that's just a dominant cheap option.

Adam and his pals dont what?
Nothing I have said has disagreed with them in their video. I have said their framing could be more clear already, but that's the only complaint. 

I never said it needed to be arm. But I also have never built a controller board like this. 

The apple cable is very well built with very high precision and consistency, and it is not a snake oil cable. 

At this point, I honestly do not know what you are arguing other than ignorance of how Thunderbolt or any other highspeed connect works, and a refusal to dig into it.  I told you the model numbers of the chips used on the apple cable. I gave you a generic block diagram of what you need to retime a PCI 3 signal. And you know that the manufacturing is incredibly difficult on these cables with the machines and tooling being no more than a couple of years old (Remember, 5 years ago, USB C cables were made.. by hand, and those had fewer connections as back than it was just USB 3, not thunderbolt).

[MarcWolfe]

4 minutes ago, starsmine said:

Adam and his pals dont what?
Nothing I have said has dis

Go back and re-read things if you really need to.