Why don't they make miners with swappable chips?

[Lorenzo]

I've always wondered why miners don't seem to come with removable/swappable chips. Pretty much all of the advancements in ASICs are in the chips themselves and the rest of the miner doesn't really change that much from generation to generation. Instead of throwing out an old Antminer S1, for example, and replacing it with a brand new S3 or S5, wouldn't it make more sense to make the individual ASIC chips replaceable? Kind of like how you can swap out an older processor on your computer's motherboard with a newer model (provided they both use the same type of socket).

Seems like this would be more economical than buying a completely new miner. Lower manufacturing and shipping costs would also mean that miners would have a greater chance of being profitable. It would be an easy way to fix the problem of dead chips too.

[1715116248]

1715116248

[1715116248]

1715116248

[1715116248]

1715116248

[1715116248]

1715116248

[1715116248]

1715116248

[1715116248]

1715116248

[TheRealSteve]

quoth self:

Processor socket: because physical sockets are expensive and for most current designs make thermal management a pain the butt.

If you were thinking of package and pinouts: because not every chip design lends itself well to the same package and pinout.  While some of it could be harmonized, there's really little incentive to do so.  Most new chip designs are going to require new board designs anyway, be that for power management, decoupling, thermal management, etc.  Going with some 'agreed-upon' package/pinout would be limiting the possibilities.
( Note that there have been exceptions, e.g. the BitFury Rev1/Rev2.  Rev2 was pretty much a drop-in replacement, and could be applied to existing designs, and even replace Rev1 chips for those willing to desolder the old chips.  Not that much 'waste', though I think applying to new boards + components would be more cost-effective; can't easily resell desoldered older chips, but reselling an old functional miner isn't terrible. )

[Lorenzo]

quoth self:

Processor socket: because physical sockets are expensive and for most current designs make thermal management a pain the butt.

If you were thinking of package and pinouts: because not every chip design lends itself well to the same package and pinout.  While some of it could be harmonized, there's really little incentive to do so.  Most new chip designs are going to require new board designs anyway, be that for power management, decoupling, thermal management, etc.  Going with some 'agreed-upon' package/pinout would be limiting the possibilities.
( Note that there have been exceptions, e.g. the BitFury Rev1/Rev2.  Rev2 was pretty much a drop-in replacement, and could be applied to existing designs, and even replace Rev1 chips for those willing to desolder the old chips.  Not that much 'waste', though I think applying to new boards + components would be more cost-effective; can't easily resell desoldered older chips, but reselling an old functional miner isn't terrible. )

Those are some good points but couldn't the same also be said for CPUs?

Back in the late 2000's, Celerons, Pentium 4s, Pentium Ds, Xeons, and Core 2 Duos all have completely different thermal profiles, power consumption figures, and even underlying architectures and yet were compatible with each other since they all used the same standard socket (LGA 775):

http://en.wikipedia.org/wiki/LGA_775

I would have thought that ASICs are much simpler than general purpose CPUs, so compatibility issues would be less significant.

[TheRealSteve]

Sure, but they're not really comparable markets.  E.g. counterpoint: GPUs.  Just to touch on CPUs anyway.. that's a much more mature area, the underlying board (motherboard) itself is also a fairly significant investment, the manufacturer (for that socket: Intel) had a roadmap based the market where being able to swap out a CPU would be beneficial for their bottom line, etc.  Even with that, that socket was eventually abandoned and later sockets found far fewer re-use across major architectures (and not just relabels for market segmentation purposes).

I'm not saying it can't be done - absolutely Bitcoin ASICs would make such a design simpler if anything (very few data lines, mostly power) - just that for the time being, it doesn't really make business sense to do so for either the manufacturer or the customer.

[helipotte]

Most ASICs for cypto hashing have very specific mounting profiles.  These chips produce very high amounts of heat for there size.  It would be very difficult

(expensive) to introduce additional physical sockets/connectors/interfaces just to make them upgradable.  Most manufacturers of these miners are mostly

concerned about getting them out the door as fast as possible and with minimal cost.  It is easier and cheaper to just design a whole new platform around each

ASIC.

[notlist3d]

Upgradable is hard to do.  One of the few to do one is bitmain S1 upgrade kit.    There are so many different types and sizes that you cant really just switch them in and out.  It's surprising how few upgrade's there are for asics.

When researching a new chip I think one of the last things on their mind.  They are more worried about the cost to produce.

I would love to see one case that you can slip chip's in but I don't see that ever happening.

[Amph]

well if you are a good engineer, you might do something about it, i mean you can swap the chip and remove also the unprogrammable chips and replace those too

you can keep the board i think

[notlist3d]

well if you are a good engineer, you might do something about it, i mean you can swap the chip and remove also the unprogrammable chips and replace those too

you can keep the board i think

If chips remained the same size it could be possible.  You would be pretty stuck on a design if you try to reuse the heat sinks.   If you just reuse connectors and ship new heat sinks you have a lot more options.

[alh]

I expect that the big killer here is the power input to the ASIC has kept changing with every generation. Kind like why there have been at least a half dozen new socket types, each designed with a specific range of voltage and current inputs. The number of power and ground pins must just continue to grow, since there really isn't any significant I/O to speak for a Bitcoin ASIC (i.e. no PCie or memory pins).

[notlist3d]

I expect that the big killer here is the power input to the ASIC has kept changing with every generation. Kind like why there have been at least a half dozen new socket types, each designed with a specific range of voltage and current inputs. The number of power and ground pins must just continue to grow, since there really isn't any significant I/O to speak for a Bitcoin ASIC (i.e. no PCie or memory pins).

They would be a huge advantage to a standard socket.  It would have to plug in to a socket or fit the cables depending on design. It could be done but it would restrict their designs having to follow something from a previous generation. 

It just goes to how much they can reuse.  Case and PSU would be easy (assuming it has a psu in it).  The control boards and hashing boards are where it would get much harder.  And if you reuse things such as heat sinks some are it won't be plug and play.  They would have to put parts together properly.

[Dexter770221]

Miners with sockets to play with different chips will not come to live for several reasons. But it's possible to make a miner that may accept different PCB's with different type of chips. Heatsink, DC/DC converter and controller board may be reused. That idea is in my head for some time but no time and money to bring it to live

[Lorenzo]

Thanks for the answers.

I wonder if swappable chips might make more sense on USB stick miners then. Stick miners have a lot of hardware built around a single chip so the ROI is very low. Being able to swap out the single chip for a new one while keeping the rest of the hardware might improve the ROI for these types of miners. There would be less cooling/design/power issues to worry about since you would only be dealing with a single chip and a single interface (USB). Also, their designs don't usually differ that much from each other:

[MRKLYE]

I would like to see this idea come to life in the near future..

Imagine being able to buy an upgradable minger tat instead of becoming obsolete you could just swap out the hashing banks (which would have built in on heatsinks or something to compensate for heat). You could have lets say 20 hashing banks per a miner and be able to have each hash bank have 8 chips or something of the like that are able to pin to the main board something like how modern CPUs are attached. Hell, Even have build your own miner kits and allow people to mix and match hash banks and cooling systems and you might be onto a really neat and succesful business idea for ASIC hardware in the near future.

[notlist3d]

I would like to see this idea come to life in the near future..

I would to but it will not happen.  Companies have a hard time making upgrade kits, they are not as easy as they sound.  Making a new miner is easier and requires in most cases no putting together.

I don't really expect to see swappable chips anytime.  Upgrade kits... possibly depending on miner and company.  All depends on if its worth their time.

[TheRealSteve]

I wonder if swappable chips might make more sense on USB stick miners then.

More sense, yes.  Sense, still kinda no.  I think I pointed you to GekkoScience's thread in another thread - in there you should find a post by vs3 who made modifications to his board to switch from the BE200 (unsuitable for a StickMiner*) to the BM1384 - it's not just an IC footprint change

* and that's a reason why it doesn't make too much sense: there aren't very many chips suitable for StickMiners in the first place - and that's not likely to change any time soon - so you'd have to plan in advance for swapping out to future unknown chips that may not even get made.