Why can't we have a motherboard with hot swappable socket?

[e22big]

Just curious, what prevent people from creating a motherboard where you can just swap out an old CPU socket for the one with the pin layout you want? Sound like a much better idea than throwing out mobo after every few years (and could make high end motherboards actually make some sense investment wise)

[manikyath]

the pin layout isnt the problem, it's just about everything else on the motherboard that has to be tailored to work with the specific CPU.

[Shimejii]

8 minutes ago, e22big said:

Just curious, what prevent people from creating a motherboard where you can just swap out an old CPU socket for the one with the pin layout you want? Sound like a much better idea than throwing out mobo after every few years (and could make high end motherboards actually make some sense investment wise)

Its usually more then the socket, but usually its because expecting an end user to properly replace a socket type item would be WELL beyond the scope of their ability. Plus its just an added cost that the vast majority would never use so they never really invested any time to make them as a consumer board. 

 

Also Bios chips have limited space, although thats pretty easy to fix. Generally it just comes down to Intel wanting more money and they change the socket layout now instead of just saying they arent compatible since they got so much flack for that

[e22big]

5 minutes ago, Shimejii said:

Its usually more then the socket, but usually its because expecting an end user to properly replace a socket type item would be WELL beyond the scope of their ability. Plus its just an added cost that the vast majority would never use so they never really invested any time to make them as a consumer board. 

 

Also Bios chips have limited space, although thats pretty easy to fix. Generally it just comes down to Intel wanting more money and they change the socket layout now instead of just saying they arent compatible since they got so much flack for that

I don't think it has anything to do with Intel though, if, say Asus decided to create a hot swappable socket mobo, there's nothing they can do about it. 

 

And yeah, I understood that it's not a gurantee than just swapping out socket will suddenly make the mobo compatible with the new CPU but it's still a selling point. And Asus could also just test it out themselves whether there's any configuration that allows it to works. After all, these were the people who managed to bring back AMD backward compatibility when the company themselves broke their promise (but fix it eventually)

[Eigenvektor]

Because the socket is just a small part of the system. The socket is connected to all kinds of devices on the board through wires, that you would somehow need to re-arrange (and add more of or remove) to match the different number of pins.

 

~edit: aside from number of traces, their arrangement might also change between CPU models. If pin 1-30 connects to the memory bus on one CPU and pin 500-530 ob another, you'll have a very hard time to account for that.

 

23 minutes ago, e22big said:

I don't think it has anything to do with Intel though, if, say Asus decided to create a hot swappable socket mobo, there's nothing they can do about it.

 

Oh, but there is. Intel could decides to withhold future specs and engineering samples from Asus for violation of contract. Designing and validating a board takes time, so it might be months before Asus can bring a new model to market after all of their competitors do.

 

Last but not least, creating a super flexible board that eats into future sales also isn't exactly in the interest of board manufacturers either.

[Kilrah]

25 minutes ago, e22big said:

Asus decided to create a hot swappable socket mobo, there's nothing they can do about it. 

Even if the signals were in any way compatible which they aren't ASUS can't know what the next socket will need, and can't integrate a chipset that doesn't exist yet that supports the next CPU that comes.

 

When there's a socket change it's not just for fun, it's to prevent the user from putting in a CPU that can't work on that platform...

[manikyath]

20 minutes ago, e22big said:

I don't think it has anything to do with Intel though, if, say Asus decided to create a hot swappable socket mobo, there's nothing they can do about it. 

 

And yeah, I understood that it's not a gurantee than just swapping out socket will suddenly make the mobo compatible with the new CPU but it's still a selling point. And Asus could also just test it out themselves whether there's any configuration that allows it to works. After all, these were the people who managed to bring back AMD backward compatibility when the company themselves broke their promise (but fix it eventually)

they would have to engineer a new bios to work with the new chips, they would have to validate everything, the VRM's have to happen to be powerful enough for the new chips, the PCIe layout needs to make sense with the new platform, the chipset might not play nice with the new cpu, ...

 

the thing about AMD's side is, the ones that were not compatible at first, but then eventually made compatible with new chips... are quite limited as to what they can do with the new chips (pcie generations, power limits, etc.)

 

the reason AMD was sort of fine with this, is because they stuck to a fairly standard layout for what all the cpu's I/O does. if a cpu manufacturer wants to innovate their I/O (DDR5 for example), there is no good way to prepare a motherboard for that, that will not cost so much to prepare for, that the cost makes it so you may as well buy two.

 

as for 'if asus wants to, nothing they can do about it' - i bet you your rear end that they have a very strict contract with the board partners about do's and dont's, to get access to the resources they need to design boards.

[Poinkachu]

48 minutes ago, e22big said:

I don't think it has anything to do with Intel though, if, say Asus decided to create a hot swappable socket mobo, there's nothing they can do about it. 

 

And yeah, I understood that it's not a gurantee than just swapping out socket will suddenly make the mobo compatible with the new CPU but it's still a selling point. And Asus could also just test it out themselves whether there's any configuration that allows it to works. After all, these were the people who managed to bring back AMD backward compatibility when the company themselves broke their promise (but fix it eventually)

It will be a selling point that ASUS or other brand gonna use to raise the board price atleast $50 higher.
And the worst part of it, is that it's not guaranteed to work.

[e22big]

39 minutes ago, Eigenvektor said:

Because the socket is just a small part of the system. The socket is connected to all kinds of devices on the board through wires, that you would somehow need to re-arrange (and add more of or remove) to match the different number of pins.

 

~edit: aside from number of traces, their arrangement might also change between CPU models. If pin 1-30 connects to the memory bus on one CPU and pin 500-530 ob another, you'll have a very hard time to account for that.

 

Oh, but there is. Intel could decides to withhold future specs and engineering samples from Asus for violation of contract. Designing and validating a board takes time, so it might be months before Asus can bring a new model to market after all of their competitors do.

 

Last but not least, creating a super flexible board that eats into future sales also isn't exactly in the interest of board manufacturers either.

I don't think it's within Intel's interest to prevent that from happening - after all, they are making and selling the CPU, not the motherboard. 

 

And yeah it maybe within Asus or other board partner interest to not make that happen, but they simply added the cost of testing, BIOS and valuation into the price of the hot swap socket they had for sales - or just put it in the cost of the mobo in the first place. I dunno, it seems to make some sense business-wise, they can get people to buy their expensive mobo which have a chance to work with more chipset than Intel advertised spec. 

 

Say, going from Sky Lake to Coffee Lake type of situation. Doesn't seems like something that will cost too much compatibility. And if you already bought into the ultra-premium mobo, you probably already have all the VRM in the world already anyway. 

[manikyath]

7 minutes ago, Poinkachu said:

It will be a selling point that ASUS or other brand gonna use to raise the board price atleast $50 higher.
And the worst part that of it, is that it's not guaranteed to work.

$50 is a very conservative estimate.

[e22big]

1 minute ago, manikyath said:

$50 is a very conservative estimate.

Here's the pitch - make it available only on the ultra premium, balls-to-the-wall model, Maximus Heroes or whatever staff that cost even more than the CPU itself. And you might not even have to charge a buck more to make it worth, if it's attracted just 5 percent more sales from the usual crowd, that's already margin you get from those sort of lower volume production

[manikyath]

12 minutes ago, e22big said:

Here's the pitch - make it available only on the ultra premium, balls-to-the-wall model, Maximus Heroes or whatever staff that cost even more than the CPU itself. And you might not even have to charge a buck more to make it worth, if it's attracted just 5 percent more sales from the usual crowd, that's already margin you get from those sort of lower volume production

here's the problem with that.

 

the ultra premium boards are a very low volume product, and the lower the volume, the higher the R&D cost 'per board' is, and by extension the more the board will cost, the lower the volume will be, so the more it will cost, ... you get where this is going.

 

adding a costly feature to a premium board doesnt attact more sales, it increases the price, and results in less sales.

 

the ultra premium boards are already balancing on this exact issue, because high end boards just dont sell much, and it costs just as many hours to make an ultrapremium board as it costs to make a basic one. and those hours need to be split among less customers.

 

also, i'd dare state that in the ultrapremium range, the potential customer would much rather buy a new board each time they upgrade to guarantee they get the most out of their new platform, rather than re-use their board and have the potential to be stuck debugging and tinkering to make it work smoothly.

 

on top of that... in this range it's not uncommon for people to flip their old cpu+mobo combo, or pass it on to a secondary rig or a family member's rig. it's not because they upgrade that the hardware goes in the dumpster.

 

case in point - i'm still running my previous rig, (parts of) the rig before that, and the majority of the rig i had before that ended up in one of my legacy machines i keep on hand for hideously dated software.

[Eigenvektor]

9 hours ago, e22big said:

I don't think it's within Intel's interest to prevent that from happening - after all, they are making and selling the CPU, not the motherboard.

However they are interested in motherboards that work well. Anything else could tarnish their brand. There are certification programs to be able to use Intel's logo to advertise your own stuff for a reason.

 

----------

 

In any case, what you're asking for is basically an impossible engineering feat. Whoever wants to design such a motherboard would need to be able to look into the future to have any chance of designing a system that can work with CPUs that have a different number of pins and pinouts.

 

Imagine it's 2016 and you're the engineer tasked to design the motherboard for Intel's upcoming 8-9th generation. You know that the CPU has 1151 pins and you also know which pin needs to connect where.

 

At this point in time, neither Intel nor you have any idea how many pins the 10th generation is going to have, nor what their pinout is going to be. Of course we already know that 10-11th gen has 1200 pins. So there are 49 additional pins that need to connect somewhere. Your past self doesn't know this yet.

 

This means you would need to design your system in a way that allows for more (or less) pins with future generations and also come up with a system that allows these pins to be routed somewhere else in case there are substantial changes to the pinout. From all you know, a pin position that connects to memory in this generation could be used to connect to the northbridge next generation.

 

Speaking of northbridge: The number of lanes connected to the northbridge might change between generations, which means your chipset needs to be designed in a way that it can work with a, as of yet unknown, number of lanes whose functionality and signaling requirements might also change. That, on it's own, is not something you could realistically design.

 

For different pinouts, you might be able to do some re-routing in your socket adapter, but that brings along its own set of problems. Crossing lanes are subject to electromagnetic interference between one another, which would increase signal error rate and could render your system inoperable if too high.

 

If that isn't enough, then you should also know that signals that are sent in parallel typically also need to arrive in parallel. Which means certain lanes need to have the same overall length. So you'd have to re-route your pins in a way that minimizes interference while also ensuring trace lengths are within spec. If you ever wondered why traces on a motherboard are zig-zagging all over the place, now you know.

 

Speaking of trace lengths: PCIe 5.0 has much tighter tolerances for signal quality and the maximum length of lanes is shortened compared to PCIe 4.0. The system you're designing in 2016 would have to be over-engineered to a point where you could comfortably say that it should be able to be within spec of all kinds of, currently unknown, upcoming standards.

 

Even if you could overcome all of these challenges, your board would most likely be multiple times more expensive than replacing it every two years.