[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: Stata (gllamm) benchmarks for different platforms?
> Philip Ryan:
> No doubt the stuff about chips and flops is important but if the objective is
> making -gllamm- more useable shouldn't the first step be to encourage
> to incorporate -gllamm- or -gllamm-like code for mixed/hierarchical models
> *into the executable* as soon as possible?
I agree that anything that can be done to increase the execution speed of
-gllamm- is appreciated. If I have got it right, the last changes in the
executable improved -gllamm- performance with 100%. Which was a significant,
and well appreciated, improvement.
However, if the (supposed) performance of Stata on the G5 is 27% of the
Opteron but the G5 really is (according to benchmarks) 50% faster in double
precision floating point operations. The theoretical room for improvement
(in floating point calculations) is in the neighborhood of 600% for anyone
sporting an Apple G5. That potential should be appreciated as well.
Now, I know this is not up to StataCorp alone. Apple and IBM has to supply
the right tools and environment as well. And, the 600% improvement is
probably just wishful thinking. However, if StataCorp is serious about
making advanced statistical analyses (-gllamm-) available on high
performance computing platforms, the G5 should be on their top priority
* For searches and help try: