Which scope for this i need ?

[Free_WiFi]

Hello there.
Since long time ago,i always asked my self which kind of equipment is required to develop your own motherboard for cpu's like Amd or Intel.
For example ,if you are developing an simple motherboard for an socket like the AM4 + 1 port pci, so which of scope do you really need to done works like this on my example?

[Zbig]

If you have to ask questions like this, you're not going to design PC motherboards anytime soon. Seriously, are you trolling? Have you designed any kind of PCB yourself yet?

There's nothing "simple" about most basic of PC motherboards and a scope is the least of your concerns if you don't even know where would you begin with a project of this scale.

[Free_WiFi]

If you have to ask questions like this, you're not going to design PC motherboards anytime soon. Seriously, are you trolling? Have you designed any kind of PCB yourself yet?

There's nothing "simple" about most basic of PC motherboards and a scope is the least of your concerns if you don't even know where would you begin with a project of this scale.

I just asking ,don't wanna to start any project like this.
However im just curious about the kind of equipment is used to craft stuff like that .
lol

[rx8pilot]

A motherboard is not simple. Not even in the same universe as simple.

Equipment is the least of your worries. You need a considerable education just to learn what the problem is. After that, you have to learn how to conceptually solve the problems. Next, you have to learn the fundamentals that allow you to understand the myriad of challenges ahead. After that, you will likely need to start learning about the test equipment necessary. To successfully test and validate a motherboard - the equipment would be $many hundreds of thousands ($USD). Then, you actually have to learn how to operate and interpret the instruments to go back and revise your design.

It is an enormous task generally tackled only by teams of people employed by fairly deep-pocketed companies.

[Free_WiFi]

A motherboard is not simple. Not even in the same universe as simple.

Equipment is the least of your worries. You need a considerable education just to learn what the problem is. After that, you have to learn how to conceptually solve the problems. Next, you have to learn the fundamentals that allow you to understand the myriad of challenges ahead. After that, you will likely need to start learning about the test equipment necessary. To successfully test and validate a motherboard - the equipment would be $many hundreds of thousands ($USD). Then, you actually have to learn how to operate and interpret the instruments to go back and revise your design.

It is an enormous task generally tackled only by teams of people employed by fairly deep-pocketed companies.

So it's for this reason,that we do not have any open source motherboard for x86 cpu's
Im right ?

[rx8pilot]

I just asking ,don't wanna to start any project like this.
However im just curious about the kind of equipment is used to craft stuff like that .
lol

You would need a very fast scope for real-time signal analysis, proper logic analyzer, TDR and/or VNA for signal integrity testing, PDN tools to design the power subsystems. You would also need software design tools that can deal with a 12-16 layer board with controlled impedance traces moving all about. That would probably include some simulation to give yourself a fighting chance of success.

So it's for this reason,that we do not have any open source motherboard for x86 cpu's
Im right ?

Open source like make your own PCB kinda thing? most people could not even deal with the soldering requirements, much less the design engineering.

[Free_WiFi]

"12-16 layer board"
really ? but the ordinary pc motherboards are not just double layer?

[rdl]

PC motherboards are very complex. Personally, I'm surprised they actually work as well as they do.

[Zbig]

"12-16 layer board"
really ? but the ordinary pc motherboards are not just double layer?

No, they're not. And once again, there are no "ordinary" PC motherboards in the sense you so insistently seem to want to see them. They're all enormously complex and way off-limits to folks like you and me. You are currently at the stage of knowing so little that you don't even seem to realize how complex these things are. Forget about PC motherboards and take baby steps, hopefully having fun in the process.

[rx8pilot]

"12-16 layer board"
really ? but the ordinary pc motherboards are not just double layer?

Most of my power supply designs are 4-6 layers. My 'simple' high-speed designs are 6-8 layers and take a TON of time for design, testing, etc....and I am just a beginner in the world of high-speed PCB design. Around the level of 'assistant' to a proper engineer that would be designing a PC motherboard. I have around $100k of equipment on my bench and it is barely enough to get through the process.

[nctnico]

"12-16 layer board"
really ? but the ordinary pc motherboards are not just double layer?

I've owned a double sided PC motherboard with a 386SX processor on it so it has been done. I think a typical modern PC motherboard will have 6 layers for economic reasons. Designing one will need skills in high power DC-DC converters with excellent load regulation and load step handling and routing high speed signals. A simulation package to check the signal integrity will also be mandatory. But I think designing a modern motherboard will be less of a challenge with a modern day processor compared to the Pentium 1 systems with and external memory controller, cache and a PCI and/or VESA local bus. Modern day processors have an internal cache, memory controller and PCI express as I/O. The DDR memory controller will be able to optimise (at least) the timing delays of each lane and PCI express consists of serial links with 2 high speed pairs. How to route these kind of signals is well documented.

[rstofer]

You could probably design a single board computer using an 8086, 80286, 80386 or even an 80486.  You might even get the ISA bus to work.  But those chips and that bus just aren't interesting in today's world of GHz computers.  I remember when a 100 MHz 80486 was king of the hill.  It worked very well for MSDOS and Windows 3.0 (maybe 3.1).  But that's back in 1990 and 1992.  Nobody is interested in those chips today.  I don't know if they are even available.

For the modern stuff, it's best to leave it to the chip and board manufacturers.

[timb]

You could probably design a single board computer using an 8086, 80286, 80386 or even an 80486.  You might even get the ISA bus to work.  But those chips and that bus just aren't interesting in today's world of GHz computers.  I remember when a 100 MHz 80486 was king of the hill.  It worked very well for MSDOS and Windows 3.0 (maybe 3.1).  But that's back in 1990 and 1992.  Nobody is interested in those chips today.  I don't know if they are even available.

For the modern stuff, it's best to leave it to the chip and board manufacturers.

There were DIY PC AT (80286) motherboard designs in several magazines in the mid ‘80s that worked pretty well. It was basically a handful of off the shelf chips. Same with the 8086 before it. (You could even wirewrap them on perfboard.)

By the time the 386 got here things were starting to get simpler for the large computer companies, as they started coming out with custom ASICs that replaced all the support chips, easing routing requirements. However, bus speeds started increasing to the point that a real PCB was required for signal integrity reasons. (A two sided board was still OK.)

When the 486 arrived you started to see companies selling chipsets that rolled all the support chips up into a single chip solution, so now that option was available to everyone and not just the big manufacturers like Compaq, Toshiba and so forth. At this point speed and power requirements really required a 4 layer board. That put it out of the range of hobbyists of the time. (Besides, there were so many companies selling motherboards at the time there really was no point in making your own.)

Then you get into Pentiums and PCI, which have strict signal integrity requirements.

[forrestc]

There were DIY PC AT (80286) motherboard designs in several magazines in the mid ‘80s that worked pretty well. It was basically a handful of off the shelf chips. Same with the 8086 before it. (You could even wirewrap them on perfboard.)

By the time the 386 got here things were starting to get simpler for the large computer companies, as they started coming out with custom ASICs that replaced all the support chips, easing routing requirements. However, bus speeds started increasing to the point that a real PCB was required for signal integrity reasons. (A two sided board was still OK.)

When the 486 arrived you started to see companies selling chipsets that rolled all the support chips up into a single chip solution, so now that option was available to everyone and not just the big manufacturers like Compaq, Toshiba and so forth. At this point speed and power requirements really required a 4 layer board. That put it out of the range of hobbyists of the time. (Besides, there were so many companies selling motherboards at the time there really was no point in making your own.)

This is about the same time that the electronic hobby took a (temporary) nose-dive for many of the same reasons....

It used to be that pretty much anything interesting could be replicated with a 2 layer board, maybe even just wire-wrapped.   When you got to the timeframe that the 486 came out, not only was the computing industry moving away from the ability for mere mortals to comprehend and build, but also much of the rest of the electronics industry.   With the death of the ISA bus, a lot of the experimenting with the PC dissapeared as well.

It wasn't until the raspberry pi and the arduino brought this level of experimentation back did we see a resurgance.   

[nctnico]

Then you get into Pentiums and PCI, which have strict signal integrity requirements.

True but signal integrity is a lot easier to manage and predict than routing a massive parallel bus. What is so brilliant about DDR memory for example is that if you get the impedance of the traces right and setup the termination the DDR controller will cancel the timing problems with the help of a seperate clock which comes along with each data lane (8 bits). The same goes for PCIexpress, SATA and USB. Get the impedances and crosstalk right and you are OK. Designing a board has become RF engineering instead of fighting delays.

[vealmike]

Yes, but now we have to manage impedance mismatches from via transitions. Unused via stubs become reflecting transmission lines. Bends in traces introduce radiating common mode signals. Component pads in signal traces need voids on adjacent layers. Care must be taken to ensure a constant reference plane. Cross-talk must be modelled.

Serial traces should be modelled to ensure they meet the S-parameter limits in the specs.

And when we get to around 40GHz the width of the copper trace will become a transmission line too, we'll no longer be able to use a simple 2D transverse electrical wave model of the interconnect. The copper traces become waveguide.

All of this must be done months before you power the system on. Then you have to deal with SerDes tuning (or hints for autotune).

The OPs question shows naivety. But that's OK, we all have to start somewhere. It's not fair to mock someone for lack of education, they may be smarter than you.


To the OP:
There's a wealth of things you need to master to design a mobo that are far more important that the 'scope you use. But to answer your question directly, the general wisdom is that you need a 'scope (or comms analyser) with a analogue bandwidth at least 3* greater than the frequency of your fastest serial trace (see below). The 'scope must be capable of plotting an eye diagram and preferably should extrapolate bit error rate. It should have differential probes that do not appreciably alter the circuit probed.

So for a modern mobo, the fastest signal is 12Gb/s SAS. This actually switches at 6GHz. 3*6=18GHz.
You should use a digital storage 'scope with a bandwidth in excess of 18GHz.

Why do we need 3*signal frequency as 'scope bandwidth...
Well this stems from fourier analysis. Fourier tells us that any periodic waveform can be constructed by adding sinewaves. If you want to create a square wave, you add sinewaves at the fundamental frequency,  three times the fundamental, five times etc.
It is generally accepted that to get a good approximation you need the third harmonic.

However, the PCIe / SAS etc signals aren't square waves. And the 'scope bandwith doesn't usually stop abruptly at the upper limit, it rolls off. So it is possible to get a fairly good look at a 12Gb/s SAS waveform using a 'scope with a bandwidth less than 18GHz. But:

* You need experience to know that what you see on the screen will differ from what is on the board. And critically how it is likely to differ.

* You mustn't tell my boss this as I won't get bought any shiny new 'scopes to play with.

All in all, you'll need to spend £100K plus on a 'scope.

[Mr. Scram]

PC motherboards are very complex. Personally, I'm surprised they actually work as well as they do.

Not just that, but the plethora of different models that are out there, considering the effort every single one takes to design and test with all the various hardware configurations. You'd think that they'd forego diversity, yet we have endless variants and flavours from a herd of manufacturers.

[Monkeh]

PC motherboards are very complex. Personally, I'm surprised they actually work as well as they do.

Not just that, but the plethora of different models that are out there, considering the effort every single one takes to design and test with all the various hardware configurations. You'd think that they'd forego diversity, yet we have endless variants and flavours from a herd of manufacturers.

A large part of that is thanks to limited testing.

Remember, with consumer equipment you are the testing and QA departments.

[nctnico]

However, the PCIe / SAS etc signals aren't square waves. And the 'scope bandwith doesn't usually stop abruptly at the upper limit, it rolls off. So it is possible to get a fairly good look at a 12Gb/s SAS waveform using a 'scope with a bandwidth less than 18GHz. But:

All in all, you'll need to spend £100K plus on a 'scope.

Or get an older sampling scope for 20 to 200 times less. Serial links contain highle repetitive signals and are therefore excellent candidates for equivalent time sampling. You only need realtime if you want to do things like decoding SATA, PCIe, etc but that is not necessary for doing signal integrity checking.

[capt bullshot]

Hello there.
Since long time ago,i always asked my self which kind of equipment is required to develop your own motherboard for cpu's like Amd or Intel.
For example ,if you are developing an simple motherboard for an socket like the AM4 + 1 port pci, so which of scope do you really need to done works like this on my example?

Basically no scope at all, or a simple 100MHz two or four channel one. The only thing that you would use a scope to debug such a mainboard is the various power supplies for correct voltage, sequence, ripple, stability. All the fast data signals must be properly designed with matching length, impedance and whatever, most of this is done by simulation while you design the PCB layout and some verification on the final PCB using a TDR, impedance analyzer or similar means.
There will be some slower signals, like system management bus or fan control, your favourite entry level scope will be fine for that.
You simply won't touch the fast busses and PCIe lanes with a scope, and if you intend to, you'll need the fastest one you can get and still won't be able to do meaningful measurements since the attachment of the probe kills your signal integrity.

[Mr. Scram]

A large part of that is thanks to limited testing.

Remember, with consumer equipment you are the testing and QA departments.

Server equipment isn't very different, surprisingly enough.

[Monkeh]

A large part of that is thanks to limited testing.

Remember, with consumer equipment you are the testing and QA departments.

Server equipment isn't very different, surprisingly enough.

Some of it, anyway. Supermicro spring to mind.. consumer-grade server equipment!

[vealmike]

Or get an older sampling scope for 20 to 200 times less. Serial links contain highle repetitive signals and are therefore excellent candidates for equivalent time sampling. You only need realtime if you want to do things like decoding SATA, PCIe, etc but that is not necessary for doing signal integrity checking.

Yes, this will work. I did say a communications analyser is an alternative.

A real time oscilloscope (RTO) can be used to do a whole bunch of stuff that a sampling 'scope can't.

• The RTO can construct a "golden clock" from the received waveform post capture using the same math model required by the serial standard.
• The RTO can tell you by how many picoseconds each signal transition varies from it's ideal location using the extracted golden clock. Your jitter shown as a waveform. 
• Really good RTOs will allow you to plot an FFT of the jitter and then to average the FFT, allowing you to hunt down deterministic jitter sources.
• A good RTO will simultaneously display transition and non transition eyes in separate windows, very useful for tuning transmitters.
• Some RTOs will accept mathematical models for the receivers. This allows you to 'view' the signal at the input of the deserialiser, after the equalisation. In an embedded system this is all you really care about.

I've never successfully used an RTO to decode a highspeed serial signal. I tried it once at PCIe Gen1 on a fully loaded Lecroy 25GHz waverunner. There wasn't enough memory to show anything useful. Neither could the 'scope show the protocol at a high enough level to be useful.

The 'scope will address signal integrity issues. If the bus is stuffed for any other reason, you need a protocol analyser, or a good deal of luck, patience and expertise.

[rstofer]

For what you get, low end commercial MOBOs are cheap at around $50.  OTOH, a high end gaming MOBO might be fairly pricey.  The last one I bought was about $270 or a little over 10% of the project cost.  Speed costs money!  How fast can you afford to go?

[timb]

Then you get into Pentiums and PCI, which have strict signal integrity requirements.

True but signal integrity is a lot easier to manage and predict than routing a massive parallel bus. What is so brilliant about DDR memory for example is that if you get the impedance of the traces right and setup the termination the DDR controller will cancel the timing problems with the help of a seperate clock which comes along with each data lane (8 bits). The same goes for PCIexpress, SATA and USB. Get the impedances and crosstalk right and you are OK. Designing a board has become RF engineering instead of fighting delays.

For modern PCI Express, sure. However, Conventional PCI *was* a massive parallel bus! It was 32-bit, so you basically had to do microstrip routing for 40+ lines across 3/4 of the motherboard.