Replace tower and mini with MacBlocks
A variety of CPUs available in Mac mini style blocks, PCI slots in another block, drives in yet another. Click them together lego style one on top of another or left to right. OS X automatically scales power with more CPU blocks, so you can still get some use out of old CPU.
Pros-
1. Simplified Product matrix, less inventory concerns for apple.
2. Consumer feels little penalty for starting small.
3. Greater upgrade benefit- no need to wait until cpu's double in performance to justify buying a new cpu.
Cons
1. Each component would have to accommodate /pass through all bus standards (firewire, USB, PCI) etc...
2. Some increased pressure on apple to support legacy equipment.
I'm playing devil's advocate on the Cons as I don't really see them as prohibitive. All in all, I don't see the downside.
Pros-
1. Simplified Product matrix, less inventory concerns for apple.
2. Consumer feels little penalty for starting small.
3. Greater upgrade benefit- no need to wait until cpu's double in performance to justify buying a new cpu.
Cons
1. Each component would have to accommodate /pass through all bus standards (firewire, USB, PCI) etc...
2. Some increased pressure on apple to support legacy equipment.
I'm playing devil's advocate on the Cons as I don't really see them as prohibitive. All in all, I don't see the downside.
Comments
A variety of CPUs available in Mac mini style blocks, PCI slots in another block, drives in yet another. Click them together lego style one on top of another or left to right. OS X automatically scales power with more CPU blocks, so you can still get some use out of old CPU.
Pros-
1. Simplified Product matrix, less inventory concerns for apple.
2. Consumer feels little penalty for starting small.
3. Greater upgrade benefit- no need to wait until cpu's double in performance to justify buying a new cpu.
Cons
1. Each component would have to accommodate /pass through all bus standards (firewire, USB, PCI) etc...
2. Some increased pressure on apple to support legacy equipment.
I'm playing devil's advocate on the Cons as I don't really see them as prohibitive. All in all, I don't see the downside.
You can't add cpus like that and you may end having to pass 32+ pci-e lanes from block to block as well.
You can't add cpus like that and you may end having to pass 32+ pci-e lanes from block to block as well.
Yeah, it just won't work. Computers can't be made like that, some of the components just need too much low-latency bandwidth. It's one thing to have external drives, quite another to modularize the CPU/GPU.
A variety of CPUs available in Mac mini style blocks, PCI slots in another block, drives in yet another. Click them together lego style one on top of another or left to right. OS X automatically scales power with more CPU blocks, so you can still get some use out of old CPU.
What you are describing is sort of like a bus based system like VME or Compact PCI. Sorta. Somewhat allied would be clustered computers (Beowulf Clusters) where old computers are tied together via fast serial network connections. Clusters though are about tying together whole computers usually for computational intensive apps.
Bus based systems tie an assortment of functionality together via I/O cards, processors and storage cards.
None of these approaches really describe what you want. To be honest the problem would be the cost to deliver what you want, the reason we have the peripheral arrangement we currently have is that it partitions funtionality into economical modules. What you desire would require an economical approach to tying those modules together. I don't think that is possible right now, the best that Apple might be able to do is to build a bus based system that can be upgraded via processor modules but the utility of those processors would limited by current software.
What Apple could offer up is a more limited snap together system with specific functions. Here things like storage modules could use something like Firewire to bus them together. But still the economics are in question.
[/quote]
Pros-
1. Simplified Product matrix, less inventory concerns for apple.
[/quote]
Actually I think you will find that things become more complex.
2. Consumer feels little penalty for starting small.
Actually there would be huge penalties as technology advances very quickly and new hardware would be hamstrung by the old. Not to mention the costs of such systems. I just don't see beating the costs of the current approach.
3. Greater upgrade benefit- no need to wait until cpu's double in performance to justify buying a new cpu.
Currently the most cost effective approach to getting more performance out of a generation of hardware is clustering. That is if you have the software to take advantage of it.
Cons
1. Each component would have to accommodate /pass through all bus standards (firewire, USB, PCI) etc...
Not really. How I/O is accessed depends upon the organization of the hardware.
2. Some increased pressure on apple to support legacy equipment.
Actually the cost to Apple would be huge. Part of that would be from the departure from the currently mass produced hardware to odd stuff.
I'm playing devil's advocate on the Cons as I don't really see them as prohibitive. All in all, I don't see the downside.
The downsides are pretty huge as you have special hardware and software to keep working. Then you have the reality that within a year or two your carefully crafted system will actually be holding up new hardware.
Bus based systems are very popular with military, industrial and communications due to the need to maintain those systems for years. That and a well defined bus allows for many I/O cards that can be interchanged on systems. The high costs of these systems means really nothing comapred to the long term support required. On these sorts of systems it is nothing to spend 3 to $5000 on a CPU card, that doesn't include your power supply, backplane and any other required cards.
Dave
___________________
Great thing to know about busby seo test.
I suspected that there would be latency issues with tying together the CPU modules to act as multiprocessor machines, but I assumed that connections to drives and PCI ports would not be an issue.
With respect to CPU modules if you cannot do multiprocessing then at worst you are talking about deriving a benefit by having the modules and the OS automatically cluster multiple CPUs. OS X already supports grid computing I believe. That would still be a big advantage for many applications provide the pipe was fast enough.
As far as the cost of legacy support I just don't see the problem... Say some new architecture comes along that makes it prohibitive to try and integrate old architecture at the level of CPU sharing/clustering/whatever. Well then you just write some software for the old CPU that relegates it to being a hub responsible for passing/translating io to the new CPU and maybe support it as a virtualized OS. Every subsequent architectural change requires only writing some departing code for the last architecture, not all previous modules. Supporting the most recent architecture is something Apple usually does anyway, it usually only abandons tech 3 or more gens old.
Done right it would actually encourage consumers to buy tech at every architecture change just to maintain continuity in their systems. And since modules could be comparitively cheap compared to whole systems, it would seem no more prohibitive than buying the new OS upgrade every year.
Finally, this seems right up apple's alley since they are the only ones that could do this as they are the only software and hardware computer company.
This type of product will stay in the Telecom/Military/Datacenter environment.
I remember there was a thread on these boards a few years back with this idea (too lazy to do a thread search
I personally believe that the consumer market is going to revert back to the old philosophy of mainframe/thin client....Have your "home server" in the closet and the family all has thin clients that tap into that compute power. This makes the 500 pound gorilla (Intel) happy....sell a few multi-core somethings into a household and a bunch of atom processors...
I happen to work in the CompactPCI/ATCA/MicroTCA market and am very familiar with this type of technology and I can tell you it will never happen in the consumer market. The biggest cost of electronic devices are quality PCBs and chipsets to support connectivity (at reasonable speeds for modular architectures). Infrastructure costs alone will keep modular computing out of the consumer market...Oh don't forget, each device has to have it's own power regulation circuits...this inflates the cost of each processor unit you add.
I believe you have a lot more expertise in this area, so I'm not arguing with you. Rather, I'm wondering what could be done toward this model.
I don't know the intricacies of power regulation circuitry, but it would seem that the existence of fairly cheap external peripherals would mean this isn't a big obstacle. Combined with something like GreenPlug tech and I think it would be pretty slick.
I think many here are misreading my expectations. I'm not suggesting that every CPU is modular. I would expect that it would be necessary to have dual processors in the same module just like the Mac Pros do now. In adding additional processors I don't expect that they would communicate as fast as true dual processors, but I think it would be possible for Apple to achieve something better than gigabit clustering through a high speed, short range connection.
Even if there was no attempt to share processors through a clustering model, it would still be an advantage to access your old CPU and Os through a window in the new OS, if only to avoid having to buy a KVM switch, but it could clearly be much more.
And I don't see why it would not be possible to divide up the PCI slots into one module and the drives into another.
As far as the thin client thing- I'm still wondering why we can't simply do this with existing macs. My kids have desks right next to mine. Why can't I just add Displaylink monitors, keyboards, and audio lines from my mac to their desks, allowing them to use the idle CPU while I'm typing this response. I suspect the economics are against this model for Apple.
I believe you have a lot more expertise in this area, so I'm not arguing with you. Rather, I'm wondering what could be done toward this model.
First I have to say your model is confused as you seem to be mixing different system techniques randomly.
Frankly I can't see Apple offering a bus based general purpose computing system. They simply target applications that I don't believe Apple is interested in!
For small scale clustering the best approach soul involve high speed serial interconnects. The problem is I don't really think you want or understand clusters. The way the majority of clusters are implemented offer very little for the average Joe.
I don't know the intricacies of power regulation circuitry, but it would seem that the existence of fairly cheap external peripherals would mean this isn't a big obstacle. Combined with something like GreenPlug tech and I think it would be pretty slick.
Personally I don't see power regulation as a big obstical. At least not on the boards. The problem with buss based systems is proper sizing of the power supply. Once you have a system where anything can be slotted in you have to carefully do your engineering to make sure that the power supply can manage. A card rack might look plug and play but it is far from that.
I think many here are misreading my expectations. I'm not suggesting that every CPU is modular. I would expect that it would be necessary to have dual processors in the same module just like the Mac Pros do now. In adding additional processors I don't expect that they would communicate as fast as true dual processors, but I think it would be possible for Apple to achieve something better than gigabit clustering through a high speed, short range connection.
Well yeah anything is possible but the question is how would you use those extra processors. How would your OS organize the system and access those processors. If you don't have an SMP arraingement then you need software written for the architecture you create. Then you would need vendors to possibly supply multiple versions of their apps, one for normal SMP and another for xyz parallel architecture.
Even if there was no attempt to share processors through a clustering model, it would still be an advantage to access your old CPU and Os through a window in the new OS, if only to avoid having to buy a KVM switch, but it could clearly be much more.
You can already do this though. It is usually over a high speed network connection but it is no problem to have a window on screen that is actually connected to another machine. Trying to move this sort of utility to a bunch of machines (cards) connected over a high speed bus offers only a lot of expenses and support issues.
And I don't see why it would not be possible to divide up the PCI slots into one module and the drives into another.
It's very possible. A long time ago I worked on a multibus system that had two processor cards and bubble memory. Bubble memory should date this system well. In any event only one processor ever accessed the secondary store. Now today you might implement a smart storage controller for the different processors bit if those different processors have different system images sharing might not happen. You would likely be better off with some sort of network attached storage.
Speaking of which it would pay to look into how data centers have evolved over time. I don't think you will find many implemented in the way you describe. Instead storage is handled via some sort of network attached storage. Or alternatively a disk array attached directly to a server. Either way functionality is partitioned a bit.
As far as the thin client thing- I'm still wondering why we can't simply do this with existing macs. My kids have desks right next to mine. Why can't I just add Displaylink monitors, keyboards, and audio lines from my mac to their desks, allowing them to use the idle CPU while I'm typing this response.
Actually with Linux you can do just that. At least there was a project going at one time to make a time shared Linux distro. I believe this was in South Africa and the goal was 4 stations per "computer" chassis. I'm not sure if there was four seperate audio channels though.
Yeah I know this is an off shoot of Linux but it highlights that it isn't to difficult to do with the right OS. You are still talking a lot of hardware though as you need GPU cards and such. A thin client approach with your machine acting as a server would most likely be faster and easier to implement. I'd have to search a bit to find the Linux project doing this but you may have some luck youself. The other option with Linux and most likely OS/X is to simply allow terminal sessions via terminal emulators; a throw back to the 70's but who really needs a GUI?
The problem is that what you want isn't that easy to implement with todays hardware when compared to the thin client model. The CPU hardware performance is certainly there but the willingness to combine the extra hardware and software into an easy to use package isn't. Which is probably why a few Linux hackers are rolling their own. I'm not to sure if Apples software even allows for multiple video cards for example.
In a nut shell the simplest and quickest way to meet some of your desires is the thin client approach.
Dave
I suspect the economics are against this model for Apple.
But even if they can't, the stackable bricks/blocks is still a very good idea for a build your own system. Perhaps they'd link together by eSata or something like that. A base mini style machine with just a CPU, onboard graphics and a small HD. Snap in a stack for a DVD burner, more for extra hard drives, even one for PCI cards, etc.
You could just upgrade the CPU by replacing the base unit, and snapping all your modules into that. More than one base unit snapped together might not be a cooperative cluster, but it could still work like target disk mode or like a KVM switch allowing each separate CPU to access the snapped in accessories as if they were on a network.
It's a really good concept, that would lend itself well to Apple designing some really cool enclosures. There's a LOT you could do with this that would be truly different from what others are doing.
If anyone could figure out a way to get "home clustering" of CPUs to work, it's Apple.
Clustering no! When I hear clustering I hear about tech that really has very little in the way of home applications right now.
But even if they can't, the stackable bricks/blocks is still a very good idea for a build your own system.
A modular system certainly has a sort of appeal but again I don't think it is cost effective. This in a way is like a throw back to the days of the Vic 20 and commodore C 64, it actually makes for expensive low performing machines.
Perhaps they'd link together by eSata or something like that. A base mini style machine with just a CPU, onboard graphics and a small HD.
A small base machine is certainly an interesting idea but is it marketable. I think not. The reasoning is as follows look at the disgust that people have for the Mini and the way it is marketed. People are already displeased with the unbundled nature of the keyboard. Now you want to get rid of the CD drive and whatever else.
That will be tough. There may be a way though if Apple markets the device as a machine to connect to the Internet. Unfortunately people still like to load software into their machines and CDs are the common format. In the end I think people would feel like they have been taken advantage of or that the product is a poor deal.
Snap in a stack for a DVD burner, more for extra hard drives, even one for PCI cards, etc.
How about a nice integrated stack of major components. That is build a full up mini with todays latest accepted configurations. Allow a disk drive module to mate up with the main unit physically but electrically you need to plug in cables. Most likely these would be Firewire 3200 cables. This is not much different than what we have today except for the idea that there would be a family of devices that all mate up in a physical stack. The base machine is still an optimal allotment of hardware for a desk top and the aux units fill in the odd requirements of individual users.
You could just upgrade the CPU by replacing the base unit, and snapping all your modules into that. More than one base unit snapped together might not be a cooperative cluster, but it could still work like target disk mode or like a KVM switch allowing each separate CPU to access the snapped in accessories as if they were on a network.
This is where I lose it in this thread. What is the point in special hardware that reproduces functionality we already have on cheap hardware. You might just as well rack up a bunch of Minis like is often done on server farms. I know it doesn't sound like much but look at the price on some of the rack based server hardware from IBM or others. All that I/O hardware costs money.
It's a really good concept, that would lend itself well to Apple designing some really cool enclosures. There's a LOT you could do with this that would be truly different from what others are doing.
Unfortunately I have to disagree with the idea of breaking a PC down into a bunch of little boxes to snap together. It just adds considerable costs and the more parts you have the more external communications paths.
On a larger scale I wouldn't mind stackable boxes of well defined subsystems to extend a base computer. A subsystem might be a storage array or a I/O box of some sort. I just don't see it as cost effective way to deliver all the parts of what we call a PC today.
As a side note; PCs in a way are already made up of standardized parts that sort of snap together. The modules being RAM, disk drives,
They've already tried that
Amorya