R devel - Feb 2009 - cpu bound cases

I'm considering some things AMD could do to accelerate R using GPU
processors. In an internal discussion I was asked 

"Are there interesting R computations which are currently cpu bound?"
I'm sure there are lots but I'd like to be able to name some real world
cases.

 

Does any one know of some good examples?

 

Thanks

Norm Rubin

 


	[[alternative HTML version deleted]]

On Thu, 19 Feb 2009, Rubin, Norman wrote:
> I'm considering some things AMD could do to accelerate R using GPU
> processors. In an internal discussion I was asked
>
> "Are there interesting R computations which are currently cpu
bound?"
> I'm sure there are lots but I'd like to be able to name some real
world
> cases.
>
It depends a bit on what you mean by cpu-bound.  Some of the arithmetic and
mathematical functions are fairly clearly cpu bound, since Luke Tierney's
multithreaded math library speeds them up.

The matrix operations in regression can easily push the CPU usage to 100%, but
the success of ATLAS suggests that they may really be limited more by cpu memory
bandwidth. I don't know if this counts.

Other people may have different suggestions.

        -thomas

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle