[Paraview] opinion on new machine for visualization

[Moreland, Kenneth]

I'm not much of an expert on hardware performance, but I think you'll get more mileage out of having two graphics boards per node.  That's how one of our clusters is set up (2 GPU's, 4 CPU cores per node), and it works OK.  Our graphics cards are set up so that each uses a different xhost.  It's a little tricky setting up your MPI job to launch ParaView to use different displays on different nodes, but once you get that set up it works fine.

I also sometimes run 4 process per node, having 2 processes sharing each GPU (you need to turn on offscreen rendering for that).  It slows down the rendering, but I prefer this mode because rendering is rarely the bottleneck anymore (at least when visualizing surfaces).  A bigger issue is that I sometimes see artifacts in this mode.  It appears to be a problem with the driver that I have never been able to track down.  For that reason, we turned that feature off for our production vis.

-Ken

> -----Original Message-----
> From: paraview-bounces at paraview.org [mailto:paraview-bounces at paraview.org]
> On Behalf Of Ricardo Reis
> Sent: Thursday, March 27, 2008 4:38 AM
> To: paraview at paraview.org
> Subject: [Paraview] opinion on new machine for visualization
>
>
>   Hi all
>
>   This may step out of the paraview only related stuff so if someone
> considers this noise please reply me personally. My lab as some money to
> buy a new machine, 8 Gb RAM, 2 x dual core or 1x quad core, not decided
> yet. Anyway, I'm thinking, for visualization purposes and using paraview,
> what would be the best option:
>
>   1 graphic board, super-duper state-of-the-art
>
>   or
>
>   2 graphic boards?
>
>   My idea is that if I'm working with very large data sets, there is a
> problem not only rendering but also processing the data and, with only one
> graphic board, lauching parallel paraview isn't very good (at least in my
> experience).
>
>   I wait your thoughts on it.
>
>   thanks for input,
>
>   Ricardo Reis
>
>   'Non Serviam'
>
>   PhD student @ Lasef
>   Computational Fluid Dynamics, High Performance Computing, Turbulence
>   http://www.lasef.ist.utl.pt
>
>   &
>
>   Cultural Instigator @ Rádio Zero
>   http://www.radiozero.pt
>
>   http://www.flickr.com/photos/rreis/