On Thursday 18 September 2008, Andrzej Lyskowski wrote:
> 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.
During the many years I have been using computers for crystallography,
I have seen repeated cycles of a leap-frog race between specialized
hardware and faster general-purpose CPUs. As you point out, we are
currently in a cycle where customizing your code to offload certain
computations to a GPU can increase the overall throughput. At least
for crystallographic software, previous iterations of this leap-frogging
have shown that such code customization is quite expensive in development
time while yielding only a short-lived advantage over general-purpose
code run on general-purpose hardware. For instance, there was a period
in the late 1980s when you could improve your refinement time by buying
machines with separate vector array processors (Convex, FPS, probably
others I have forgotten). But it was only a few years before cheaper
general-purpose machines leap-frogged back into the lead. I am dubious
that averaged over a 5-year period the investment in specialized code
development and expensive specialized hardware was a net win.
And indeed, here is a recent harbinger that the era of GPU-based
computation may be very short-lived:
http://arstechnica.com/articles/paedia/gpu-sweeney-interview.ars
A similar alternation has been seen historically in the tradeoff between
self-contained platforms and external computational resources accessed
by relatively dumb terminals. For some years now computer use has
favored self-contained laptops or graphics workstations. But now the
pendulum is swinging back again. Yeah, it's a bit different this time.
This time around the external resource is distributed ("cloud computing")
rather than centralized, but in essence it's "déjà vu all over again".
Whether the cloud computing fad will extend to crystallography remains
to be seen. Note that distributed ("cloud") data storage has been
seriously proposed as a possible solution to the problem of archiving raw
diffraction images.
--
Ethan A Merritt
Biomolecular Structure Center
University of Washington, Seattle 98195-7742
|