I still don't think at this point a mac mini cluster would be a suitable replacement the mac pro -- especially since it lacks drive access and expandability.
You're missing the point with this sort of configuration... Storage would be on a SAN system, and the Mini cluster attached via Fibre Channel or 10G Ethernet (InfiniBand over TB wouldn't really be a good solution due to the additional latency introduced by TB; InfiniBand was designed explicitly for low latency).
An iMac with a bunch of TB devices plugged into it, 10 minis? Why not just buy a self contained machine like a Mac Pro or an HP Z series and be done with it? You're trying to compensate for Apple's minimalist approach.
If you're talking about a desktop, the Mini's have limitations. But as a cluster, those limitations actually reduce. They don't disappear entirely (FC or 10G Ethernet over TB is still throttled a bit for example, and InfiniBand isn't an option). But the cost/performance ratio is quite good for such systems (not just the Mini), and will make such solutions attractive where viable.
The point is, soon when we get dual channel 100mbit TB, daisy chaining minis will give the same CPU power as Mac Pro, so as long as we can add GPU's, it'll actually kill any need of a Mac Pro.
TB is currently 10Gb/s, with it claimed to be capable of 100Gb/s (optical).
The Mac Pro is at the top of the Apple ecosystem. Make it extinct, and you may discover that its value to the ecosystem was far more than its per-quarter profits.
I'm not convinced of this, as creative content can be developed on Windows and Linux platforms (OS X doesn't have a monopoly here).
And for software development, they could chose to use an iMac, especially if it's the top tier desktop. Or they could also opt for a Linux or Windows system (they own the OS X IP, so hacking their own OS to work on different hardware is actually viable).
I'm not sure that Apple gets it, though. MBA Tim and the bean counters might not get it.
They may, or may not get it. But regardless, they're business people, and will use things like ROI to guide their decisions.
So if the ROI on the MP is horrible to their way of thinking, it's going to be EOL'ed. It's really as simple as that.
I love your style of thinking. I seriously hope this the way Apple goes, towards modularity. In the future, this modularity could even apply to the MBA, MB Pro & iPad line. You want more CPU, GPU, & storage, just daisy chain through Thunderbolt.
With optical cabling, Thunderbolt's potential is 100Gpbs. Once that happens there's no need for internal expansion.
Workstation-class equipment has needs that can't be faked by daisy-chaining laptop-class equipment. Either you need the power or you don't. If you need it then you have to buy this class of equipment that is almost at the same level as server-class equipment.
Keep in mind, that computing is turning back the clock in a sense. More and more users are going to be turned to the cloud, so their systems/devices are more like terminals of the past (no need to be very powerful, just access the main system over a network).
Now I realize your point, and it's correct. But the workstation market is changing, as is Apple. And unfortunately, Apple has already abandoned the server market, and the MP isn't generating the growth rates that would keep it a highly profitable product. So the likelihood of it being EOL'ed is high.
Now given Apple's focus has been directed towards devices and other consumer products, it seems to me they're well aware of this, and have been planning accordingly (probably began in earnest once they realized what they had with the original iPhone). Which means they've known for some time that the MP, at least in it's current form, wasn't going to remain around for much longer.
The list of add-ons you can't attach to a Mac mini is very long. Even if there was a third-party market for SAS controllers and SDI add-on card that would work through Thunderbolt, how would you ever attach something that approaches two NVIDIA Quadro cards connected through SLI?
- Think of a cluster attached to a SAN (Mini's aren't meant to replace workstations, particularly with all but one model are dual core systems, but would make decent, cheap clusters).
- In terms of GPGPU processing, the data between the system and the GPU's isn't that large compared to the data generated during the processing between the GPU's (resides in the GPU's VRAM). Then the completed results are sent back, but it's not that large either (i.e. single value returned per variable). So TB for GPGPU processing is actually possible.
Wouldn't it be great, if you could just add as solid dual-power-unit server to your server rack that runs OS X to manage mobile devices and desktops in your company or institution? That should gain them more users and really solidify their position from an ecosystem standpoint, shouldn't it?
Apple blew it in the enterprise market, as there wasn't sufficient follow-through (not enough hardware options, or software integration). So their offerings suffered from low sales, and they had to EOL the XServe as a result.
I would suggest instead that people lobby Apple to allow OSX to run on a small set of top-tier x64 servers from HP/IBM/Dell/SuperMicro, including running virtualized on ESX or Hyper-V.
I'd love to see this, but I don't see this happening unless they dump ever computer model they sell, and concentrate solely on devices, iOS, and content delivery.
Just my perception, anyway.
