Agreed, it's bunk. The reason you have modular designs is precisely because some parts aren't good enough so you need to be able to swap in better ones. When all the parts are good enough, and are going to stay good enough, modularity offers insufficient advantages.

The thing with software is that you can have your cake and eat it. App platforms are intrinsically modular - every app is a little module of functionality. Looked at that way, the Apple ecosystem is staggeringly modular, with hundreds of thousands of modules available from many thousands of developers.

It's possible what Christensen means by modularity is really commoditisation. The idea that eventually base operating systems will be so mature and stable that it won't matter which one you're running, so there will be no reason to pay a premium for a proprietary system. I don't think that will ever be true. Human society, fashion, behaviour and needs are too dynamic and advancing too fast. OS and platform development has been going at breakneck speed for the entire history of computing and shows no signs of slowing down. Apple apparently 'lost' the PC wars because modular commoditised modular windows ate it's lunch. Yet now we live in a world where innovative mobile OSes, have totally redefined the computing experience and in the form of the iPad is chewing a great big hole back into the established computing market. The truth is that OSes were never commoditised and just good enough. It just looked that way because the stagnation of the dominant platform was misread as being stability.

>> Back when PCs weren't very good and you had to make real choices about what processor, hard drive and so on to get, DELLs customization and modularity business flourished.

The cpu/memory/chipset remained integrated , due to performance reasons. But external bus performance was good enough to separate some functionality into external cards.

CPU, memory and chipset have always been very much modular parts. Chipset is determined by which motherboard you buy, memory comes as sticks you put into the motherboard, and even the CPU sockets in to keep it modular.

I mean just imagine the trouble to keep the CPU seperate from the motherboard, as is still done today! It is very very difficult to build a socket (and matching CPU) for the 1000+ pins of a modern CPU where each pin can possibly carry signals at multiple GHz frequencies.

Yes you're right ,they are modular.

The change christensen refers to is the shift from companies that built whole computers, like IBM - to the IBM pc , which was an assembly of parts from different companies.

My guess is that at that time , the pc-XT came with memory chips that weren't soldered , but in sockets. At that context that didn't mean much loss of performance. That habit of pluggable memory stuck. I'm not sure you lose much performance due to it.

And if we're talking about integrated memory - intel does have caches. Probably those are the best ways to deeply integrate memory.

And regarding CPU and board: it would be quite hard to integrate chips and board. There's was one attempt i know but it failed as far as i know. It's complex and not economical, but it does offer great performance.

I think what really happens is that the complexity of each modular component increases over time. It's kind of like Sutherland's Wheel of Reincarnation. When microprocessors were first created they didn't have any internal RAM, then SRAM became part of the microprocessor die. A machine with four cores used to require four sockets. Now you can do it with one, and that one will have a GPU on it.

But we still have external RAM, we still have discrete GPUs, and multi-socket motherboards, etc. The modularity doesn't go away, it's just that some applications require only one of the increasingly powerful modular components.

But when the increase in power doesn't come with an increase in price, Jevon's paradox kicks in and we start coming up with new applications for the increased power. So now Seagate has announced a hard drive which uses Ethernet as the primary interface -- they've effectively integrated a (simple) database server into the disk drive and made that into the new modular component.

So the real question with phones is whether they're yet in a position to become the new modular component. For that you have to answer the question of whether they can integrate with the other components to do what the user wants. How do I add storage to it? How do I make it faster? How do I make it take better pictures, etc.? What you want is to be able to do these things somehow without throwing away your entire existing investment, and until the common user can do it easily (e.g. by syncing the phone to a NAS device that provides bulk storage) there will be a market for modular phones that let people walk into the shop and have the tech insert another 8GB of memory or storage or upgrade the camera etc.

Interestingly the trend toward integrated devices has been held together in part by The Cloud allowing network storage and processing to take place off of the device and therefore reduce the demand for on-device capacity. If the trend away from trusting third parties with your data continues to gain steam then we could see changes in the market demand for modular devices.

From that perspective, you are right. However I think Christensen's perspective still holds if you go back further.

When the first semi-mainstream GUI-based PC (the 128k Mac) came out, the limitations of hardware at the time meant the thing was a mass of clever (perhaps genius) hacks to get the whole OS to run in with such limited RAM and CPU constraints.

Apple simply needed to be making the hardware and software to tweak them enough in tandem to work together.

As CPUs and memory improved, you had enough leeway so everything didn't have to be custom-engineered down to every 1 and 0 and bit of silicon to work well enough, and so the PC market entered the "modularized" phase.