Posted by
mahogny on
URL: http://confocal-microscopy-list.275.s1.nabble.com/Advice-for-offline-image-analysis-computer-tp1109084p1109153.html
Christophe Leterrier wrote:
> Dear listers,
>
> I have to buy a new computer for our team that will be used as for
> offline image analysis. The software required (ImageJ, Metamorph,
> Matlab) require to build a Windows/Vista machine. The cost would be
> around $5,000 / 5000€. The job would be processing (a lot) and
> rendering. My questions are :
this sounds like a lot to me if it just includes the hardware
> - Should I go for a "traditionnal" computer, I mean a 32-bits, dual
> core, XP computer with maxed RAM (I guess it is 3GB or so) ? Is it
> usefull to go to more fancy stuff like 64 bits, 4 or 8 core, 8 to 16GB
> RAM machine ? Would ImageJ/Metamorph/Matlab really benefit from it (I
> would love to have an answer from the Metamorph/Matlab people) ?
with 32-bit, there is little/no point in more than 4gb (of which 3gb is
available for your program).
imagej can use the full 64-bit as it is java. I think matlab exist in
64bit edition. I can not say anything about metamorph. endrov makes full
use of 64-bit. expect all maintained software to at some point support
full 64-bit, legacy software will not improve (recompilation needed).
but remember that more ram only speeds up the computer if your software
actually will use all of it. this depends on your application. I would
buy 4gb for rather heavy normal use, more if you have a special purpose.
cores: matlab has essentially no use of them (but you can buy a toolbox
for it), imagej has limited use, endrov full use, no idea about
metamorph. they will become more important in the future but require
special programming. 2 cores are always useful, the OS can use one. get
at least 2-4 cores. I doubt 8 cores will be fully useful in a while,
speed-up is not linear.
> - What about graphic cards and GPU ? What is the best choice ? I've
> heard about new strategies to speed up processing by making the GPU
> churn data as well as graphics, but I don't think it is really
> commercially available now or implemented in commercial software yet.
you need the card for 3d rendering. recent cards support GPU calc as
well with no extra cost.
some rather obscure software can do GPU calc but GPU programming is
still taking a lot of effort. expect this to be a hit in 2-4 years,
earlier if you are doing something really intensive (deconvolution would
benefit). the nv8-series is a rather safe bet, can do CUDA, and nvidia
has rather good linux support if you need to upgrade your software later.
--
------------------------------------------------
Johan Henriksson
MSc Engineering
PhD student, Karolinska Institutet
http://mahogny.areta.org http://www.endrov.net