[geeks] PS/3's, Linux, etc, was: Second Life is not a game?

[nate at portents.com]

> Can I get the same GPU for a PC and will the COMPILER for the GPU run on a
> PC?

Well the CELL is not a GPU, however, you have a number of options...

> Since the GPU does the work so much faster than an X86 processor,
> I asking if I can wrap a PC around it instead of (if you pardon the
> expression) a toy?

Sounds like you should look at the Mercury Computer Systems Cell
Accelerator Board (CAB):

http://www.mc.com/microsites/cell/ProductDetails.aspx?id=2590

They also have a 1U Dual Cell-Based System, 25U/42U Dual Cell-Based Blade
System, Cell Technology Evaluation System, Cell Workstation Development
System, and PowerBlock 200 System.  IBM also has it's BladeCenter QS20:

http://www-03.ibm.com/technology/splash/qs20/

Next year the Los Alamos National Laboratory is getting the IBM
Roadrunner, a supercomputer with "16,000 AMD Opteron cores (~2200 IBM
x3755 4U servers, each holding four dual core Opterons, connected by
Infiniband) to an equal number of Cell microprocessors resulting in a 1:1
ratio of Cell cores to Opteron cores":

http://en.wikipedia.org/wiki/IBM_Roadrunner

However, what you should keep in mind is that if you do develop some great
software for the CELL, it probably won't be useful on anything else.  In
particular, you should look at one of the last conclusions in the paper
"Limitations of the PlayStation 3 for High Performance Cluster Computing":

http://www.netlib.org/lapack/lawnspdf/lawn185.pdf

"Programming paradigm. The most attractive features of the CELL processor
are its simple architecture and the fact that all its functionalities can
be fully controlled by the programmer. In most other common processors,
performance can only be obtained with a good exploitation of cache
memories whose behavior can be controlled by the programmer only
indirectly. As a result, the performance of applications developed for the
CELL architecture is very predictable and straightforward to model; it is
possible to get very close to the peak performance of the processor, much
more than what can be done on cache based processors. This high
performance and predictability, however, comes at a cost. Writing
efficient and fast code for the CELL processor is, in fact, a difficult
task since it requires a deep knowledge of the processor architecture, of
the development environment, and some experience. High performance can
only be achieved if very low level code is produced that is on the border
line between high level languages and assembly. Besides requiring a lot of
effort, source code developed for the CELL processor is not portable at
all on other architectures. Even though higher level programming models
have been proposed for the CELL processor (see [8,6,5]), it is only
possible to achieve a small portion of the processor's performance relying
on them."

- Nate