Getting back to the whole 'Datacenter on a chip' concept, this chip can seriously help developers and datacenter managers.

Developers can use this chip to virtualize a large amount of OS instances either via LDOMs or Solaris Zones and model their whole datacenter within 1 physical machine.

What do I mean by this?

A developer using an UltraSparc T1 or T2 chip running Solaris can divvy up the system into application, database, webserver, identity/LDAP zones and deploy her code on this box to mimic how it would really look in a datacenter. This allows developers to get a true feel for how their application architecture will behave in the actual datacenter where each layer may be a separate physical box. Now, you may be thinking, "Wait, I can do that on any CPU running Solaris!" and this is true, but the difference here is that developers can actually beat on their architecture without the machine falling over. With Solaris Zones set up in this manner, a developer can easily pinpoint which layers (db, app, web, etc.) are being hit the hardest and plan accordingly.

So, how does the UltraSparc T2 help datacenter managers? Massive reduction in costs. Everyone is talking about racking up hundreds of 'low cost' x86 servers nowadays, but what is the real cost? Racking up servers is not cheap, it requires man-power, electricity per machine, at least one network switch port per machine, and management of each physical component, not to mention the actual physical space that each machine takes up in a datacenter. Network switch ports are not cheap, electric bills and cooling are not cheap, and datacenter space is nowhere near cheap. Would you rather rack up a handful of 'cheap' x86 servers, or one UltraSparc T2 based server?

With all this being said, when the original Niagara/T1 chip was released, many of Sun's competitors claimed the chip was for "niche computing". Let's see now, which niches do the Niagara chips fill?

The fact is, the niches that the UltraSparc Niagara chips serve are quite obvious, they're the computing niche, the datacenter niche, and the multithreading niche.