Re: cpu-gpu vs cpu

From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Tue Nov 29 2011 - 07:41:51 CST

On Tue, 2011-11-29 at 07:44 +0100, Francesco Pietra wrote:

[...]

> The answer should not be that any NAMD user has to develop the
> application he is interested in, or pay for that. Such a (recurring)
> answer forgets how much time is required to stay aware, for example,
> as a chemist dealing with biology. On the other hand, the inventor of
> GNU or free-to-use-for-noncommercial-purposes software is paid back in
> visibility (citations/year) much more than those using his software
> are.

please explain on which authority you can make these statements?

i contend that the "payback" in citations is rather minimal in
relation to the effort that is required to implement a feature.
if at all, only the authors of the "reference publication" get
credited, rarely the ones that provide additional features, let
alone the ones that provide the user support that is essential
to many projects, particularly for people that have little
technical expertise of their own. claiming that you can pay
pack for that by simply including a citation in your paper
is neglecting most of the effort that goes into maintaining
and supporting an free to use open source software package.
with an "essential" package like NAMD, it is almost inevitable
to no cite it, but with VMD, for example, the situation is already
different. i have seen countless paper where VMD was used for
visualization (or analysis) purposes, but no credit was given
at all. ...and then how about the many, many contributed modules
in VMD. people may cite VMD, i've very rarely seen a mention
of any contributed feature. how does that match with your
assessment that a citation of the software pays for it?

i have been working on several open source or free for use software
projects for more than 10 years now, but it would have been completely
impossible to have done this without being "cross financed" through
other activities, and i know that the majority of people working
on such software are. there is _very_ little funding to be had
for scientific software development.

nobody says that you do have to be an expert in implementing a
specific feature that you need, but have you ever thought about
including an "software feature development fee" or a like in any
of your grant applications that will allow you to (part-time) fund
somebody to have a feature that *you* need to be implemented?
if more people would do that, there would be a "market" so that
folks can get paid to focus on developing scientific software
and that people with the proper skills would want to go that
route, since there is a chance that you can make a living off it.
right now we are stuck in a situation, where the number of users
has been continuously growing, but the number of developers
(and the amount of funding) at best remained constant, but
in many cases declined. people entering "the business"
have in general less IT/programming skills than they used to
have and that makes it much less likely that want to do
software development of any kind. in fact, even the number
of people volunteering to support "no-cost" scientific software
is declining, since many consider this time as "lost", and
would rather spend it on doing more research work. in the
extremely competitive "business of research" that is an
understandable reaction, but that doesn't help the
software project.

in my observation, it is particularly people with an attitude
of "i am a researcher, i don't need to care about software
development. that is what 'the developers' are for, they
already get paid to do what i need" that are essentially
pulling the rug underneath themselves. if the current trends
continue, there will only be a small number of overworked
enthusiasts left that have to face a huge number of inexperienced
users that need so much "hand holding" that there will be
no more time left to develop new features, improve the software
or fix bugs.

> Thanks for predicting a possible date for that transition. It is known

what transition?

> that software for QM calculations is being developed (for example
> GAUSSIAN-NVIDIA). I know that these are commercial enterprises, but
> all us help (the little anyone can) NAMD-VMD getting public funds.

what has that to do with the price of rice in china?

i fail to see how the gaussian folks jumping on the
GPU bandwagon as one of the last(!) in the field is
helping NAMD or VMD? this is just a "must do" PR move.
there are already packages out there like TeraChem
that, being developed from ground up to run on GPUs,
will outrun anything that the gaussian folks will ever
come up with (but of course nobody will be permitted
to provide any numbers proving it).

cheers,
    axel.

> francesco pietra
>

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:21:02 CST