Its something I have been thinking about, at the level of reading
about CUDA ;-)
I did not go even as far as 'hello world' or buying a card yet!
Awfully old fashioned.
From what I understand, GP-GPU is useful for specific problems, under
the general idea of 'stream processing'
That means that if you have a lot of objects that should be treated
exactly the same way, then you would get
significant speed-up. Also, the 'treatment' has to be within what GPUs
are good at, and they should work
with small chunks of memory. btw, I wonder still what is the speed
difference calculating a few thousand
of exponentials in a GP-GPU and an SSE2-enabled CPU ? Any idea ?
In X-ray crystallography, model building seems to be indeed something
you can approach with GP-GPU.
Refinement and data processing seem to be a no-nein-niet-oxi to me,
while phasing seems to be an idea,
but most likely not worth it.
In the context of ARP/wARP its not worth it, since model building is
2-10% of the execution time.
Algorithms like ACMI's model building are worth it and they are
already multi-CPU parallel.
Then the questions becomes if its faster to wait for 16-core
processors (two years?) or spend two years
programing in CUDA a software that will be not really portable...
Having said all these, I might be trying it! ;-)
A.
On Sep 18, 2008, at 10:02, Andrzej Lyskowski wrote:
Hi,
I was just testing CUDA enabled version of VMD and saw an increase
of calculation speed just like promised - more or less 70 times!
Makes me wonder is anybody writing crystallographic software plans
to implement GPU support for the calculations? It would be nice to
use all the resources of our computers.
Regards, Andrzej
--
Andrzej LYSKOWSKI, Ph.D. ([EMAIL PROTECTED])
Institute of Biotechnology - Structural Biology & Biophysics
P. O. Box 65 - Viikinkaari 1 - Floor/Room: 4/4316
Mail: FIN-00014 University of HELSINKI, FINLAND
Delivery: FIN-00710 University of HELSINKI, FINLAND
TEL.: 358 9 191 58921
FAX : 358 9 191 59940
