Originally Posted by RobM
Right there wiz !
We've always had the ability to add whatever we might have needed through ye olden PCI/e slots.
If you work in an environment where that is possible, great but not every Pro user is that free to swap cards around. With today's electronics I find the economics of these upgrades harder to justify anymore. It is a certainty that some industries can justify GPU upgrades with every GPU revision but that is a diminishing crowd. I see most users being better off buying the best performance they can get at the time of the computer purchase.
Gone - the future for the MacPro is yet to be written.
I like to think of it this way, Apple is free to screw up the future of the Mac Pro. I really like the new design personally but that doesn't mean that Apples pricing will align with my willingness to pay. My big fear is a machine priced astronomically high and out of line with the hardware contained. It is my hope that the new design effectively lowers costs to Apple.
Yet capability is there through TB, but as we've discussed in other threads - still an X factor unless TB become ubiquitous.
I've been hearing about Ethernet protocols over TB, if this is real it could lead to very cheap but high performance clustering over TB of Apple hardware. This becomes very interesting in the case of the Mac Pro as one would never need to throw out a Mac Pro if the software supports clustered compute nodes. So every two years or so when the GPU manufactures deliver a real GPU improvement one can simply buy a new Mac Pro and plug it into the cluster with you costs being the one TB cable.
If this is in fact a reality TBs future is not in doubt, it is in fact sealed. There are many things TB can do that no other low cost port can, it is really just a matter of Apple leveraging the hardware. The idea that TB needs to be "ubiquitous" is in my mind misplaced. All we really need for TB to get a lock on its future is a couple of key apps. We already have the use of TB for docks and high performance storage, if this IP over TB is real and Apple supports clustered installations nobody will have to worry about TBs future. We will have three compelling uses for the port, each with rather broad support.
Could end up like fw ... great tech, adopted by the PC crowd late. Then swamped by usb3. Bleeding edge here ...
Actually I've heard this crap one too many times, TB and USB 3 aren't even in the same ball park as far as application and performance goes. I'm not sure why this keeps coming up in discussions, I've yet to see a video monitor with a USB connection that also acts as a dock for example. The idea that TB was designed to replace USB is a joke too, even though Intel and Apple have tried to dismiss the idea people still seem to grab on to the idea that TB is a USB replacement. One look at cable construction should dismiss that idea completely, but yet the idea continues on in the publics mind. TB will be successful doing things other ports can't. That might not be the massive market seen with USB but who cares?
In any event let's get back to the clustering speculation, people often complain about the lack of dual processors in the new Mac Pro even though there are two high performance GPUs in the machine. OK I can understand that some apps just need a lot of I86 cores but is a massive Mac Pro box the right way to go to deliver those cores. For many apps I would say no, it makes more sense to cluster a bunch of low cost nodes to get the cores you need. TB, at 20GB per second, would allow for that to happen with good performance over a wide array of apps. A Beowulf cluster with a high performance interconnect if you will.
With such a clustering ability many people win. Apple for one would sell more computers and their computers would be seen as an easy avenue into high performance low cost computing. Users will win big time as the majority of the users out there, that don't need this technology, would enjoy lower cost hardware as they wouldn't need to support the small minority needing a big box. User also win due to malleable systems that don't completely die if one node goes belly up. Software developers win also as they will have a limited number of Mac Pro configurations to work with. More so if developers have customers complaining about performance they can suggest adding another node ( given that the software supports operation on a cluster).
For some developers a mind set change is probably in order, others will have apps that just don't fit into a clustered environment. However for many industries Apple targets, clustered computers are nothing new. What would be new here is doing it for a relatively cheap price and perhaps on a smaller scale. However small scale is relative here, a cluster of 4-8 Mac Pros would be nothing to sneeze at, especially for GPU accelerated apps. I can see many developers getting excited at the prospect of cheap low pain clustering and are probably hoping that this feature is real.