Originally posted by steve-dave
Pixar bought a Blade server system, which Apple does not offer. It had nothing to do with CPU power. Blades are a hell of a lot cheaper than an X-Serve and take up far less rack space. If Pixar had bought X-Serves it would have cost significantly more and taken up far more space all for the same amount of processing power.
If Apple does create a high-end server it almost certainly will be in a blade configuration.
Erm, Xserve is a blade. You cannot take up 'less rack space' than an Xserve cos it fits in a 1U rack mount space, which is the smallest you can use, just like a blade. If you mean they needed a blade
cluster then that was correct when Xserve was first released but they now have third party clustering software and I'm sure Panther Server will have a full on clustering solution. (Clustering is using a whole bunch of machines, blades or otherwise and making them look like one big machine to anyone who wants to use them. Distributed computing made easy to use - you don't have to roll your own load balancing and fault tolerance software).
As for 'blades are a hell of a lot cheaper', this only applies for the hardware for Intel blades. Sun and other *nix blades are expensive. When you factor in the software licences for handling lots of users connected (e.g. for an ecommerce website) the Wintel solution becomes very expensive, Apple keeps its software costs per server rather than per user making it a lot cheaper for that example, with the possible exception of Lintel blades.
Most blade cluster buyers want to be able to add extra performance on demand and need all that performance to handle lots of users, leading to big software licencing costs. Pixar doesn't have lots of users connecting to their blade cluster, only people in their own company, not the general public using an ecommerce site. Therefore per-user licencing would be less of an issue for them, making Wintel or Lintel blades relatively attractive in terms of price / performance. Especially compared to the original Xserve which lacked performance relative to an Intel based blade. However the Xserve G5 will be another story due the big bump in performance, with hopefully not a big bump in price.
When Pixar needed a cluster of blades, a render farm for them, Apple weren't far enough down the road with Xserve and more importantly with XRaid for storage. Can you think of a company that generates more data than Pixar and therefore needs a top class storage solution?
This new machine (if it really exists) will be rack mountable, but it won't be a blade (1U or 2U max), it will be one big box with lots of CPUs. A 4U or 5U (ie it takes up 4 or 5 slots in a rack). It will be usable in a cluster as well but it will have the advantages of lots of CPU's without the headache of managing a cluster, which is quite complex. Think about how you would handle one of your machines in your cluster going down. This is why distributing computing, the magic bullet of the 90's took a while to take off. If you've given work to one machine, how do you know whether it has a) gone down or b) is just slow. Do you give the work to someone else, how do handle the results of the calc coming back? In a transaction processing environment, this is a nightmare. Hence you'll always need big boxes for some things and you'll need blade clusters for others.
Sanj