[deadalvs]

olà !

* * *

a few questions about scientific computing and potential maxwell supercomputing

* * *

previously info and «colorful pictures»:
http://www.dalco.ch/press
http://www.fluent.com/

* * *

about using «supercomputers»

1[
software like fluent or any other scientific application: do they run ONE large OS with a software that distributes all calculations over all processors OR is there a «small» OS running on each rack/blade that is fed with tasks for just this unit.
my question is pointing to shared memory technologies (over the whole supercomputer) that are used today which seems to imply just ONE OS.

2[
are these applications especially written to run on supercomputers and thus there is no way to implement a workflow with a linux/redhat version of maxwell that would «see» the computer as ONE unit with n cpus instead of just one/two/four.
i want to underline the question if a supercomputer in such a workflow would rather render an image as ONE .mxi, as a standard computer does, OR if such a workflow would spread .mxis with seeds (1) to (n+1) across all available racks. what would be the optimum?

3[
is it possible to use a «standard» software like maxwell for linux to run on such a machine? are there special OSes/software implementations around that would spread a maxwell task over the whole cluster ?

4[
which is the max numbers of cores, maxwell can allocate under one OS?

* * *

please don't just tell me this is all bullshit i ask, i'm just interested in the technical facts - not based on economy or whatever...

i have not really been in touch with supercomputing. i'm just at the moment in contact with one of the leaders of renderrocket.com, which makes me interested of a possible use of a supercomputer in conjunction with maxwell rendering.

thanks !

[Hervé]

hehe, I'll be dead by then....

[deadalvs]

so - You're an optimist ...

Hervé - we need You here at least for the next 50 years !

* * *

no, seriously.

there are already diverse firms active and online that sit in front of a supercomputing service. not all, but some of them (i mentioned one already, and one other is used by some NL members) have developed a workflow to upload manage and render .mxs scenes and download the result.
of course this is all not ripe for a larger amount of users or even mainly animation studios, but maxwell's quality certainly has the right to ask about more cpu power from the industry. it screams fot such solutions!

there are many posts on the web that say supercomputing is good for scientific calculations (chemistry, cfd, structure analysis, ...), but many times, rendering is also mentioned! but do i see a revolutionary step ? naah.

imagine. if the game world needed a supercomputer instead of graphics cards - where would we be !!
by the way. supercomputing on a GPU cluster is even more interesting ! speed factor 40 today !! well, and we also discussed about a maxwell version that uses the GPU to render...

* * *

remember the discussion about a p2p program that would render like SETI does it's calculations? we've seen this would not work simply based on data size.

but we could form a group of interested people that kicks in the door at such a firm and say: «we need Your 2000 cpus ! what price can You make?»
the price is always the problem. pay 3$ per core per hour on opteron cpus ? this is insanely expensive ! running 5 nodes à 4 cores for 6 hours would cost 350$ !

of course, supercomputing is Xtremely cost-intensive, but there must be a way and ther must be an interest from user-side.

am i so wrong ?

[Hervé]

.. he he thanks for the 50 next years... ehe

well sure there is an interest in supercomputing... sure there is an interest in finishing Maxwell renders much faster... but like you said, it is going to be intended to Mega Studios..

Now don't forget big studios represent a small % of the license owners.. not saying we should overlook that solution.. but for some years, $$$$$$ will be a show stopper.. unless of course a large company starts to mass produce them at a reasonable cost... which I doubt..

call me pesimist... but realist is better in this case I think...

(as always I could be wrong... well our children might play world of wars on a personal supercomputer... err.. dad !!! bring me back some "cryo cans" from the store... would you..?)

hehe
Hervé

[ivox3]

hey Hervé -- are these cryo cans to be consumed or do they go in the super computer ??? or both !

[Hervé]

they'll do both... giving cold to your computer... and energy to the user.. .. but the bad thing is that they'll be made by microsoft.. so some might have ... little problems... go figure..