8 processor cores in one chip

The Niagara chip sounds really wonderful.
I got a T-2000 for a free trial.

To do so you needed a highly threaded application that runs on solaris, and uses
little floating point. I had such an application. I got it quickly running on solaris and
benchmarked it. It turns out that the 4 hardware threads are useful to hide the
latency of multiple threads of execution. If the threads don't have significant latency,
then two threads can easily saturate a single cpu-core (as I found out). I think the design
is great for webservers or applications like that, but not a good general solution.

My loaner T-2000 had 16 threads (4 cpus each with 4 threads). I had a sun engineer
test my application on a 32 thread system, and the performance was significantly
slower than my dual processor 2.4ghz xeon. This makes the cost of the Niagara
prohibitive.

My specific application (which I wrote) easily scales to 512 or more threads.

I have all the details of my testing at http://www.weasel.com/com-perf.html
including runtime, power usage, and some basic cost comparisons.