AVS Forum banner
Status
Not open for further replies.
1 - 8 of 8 Posts

·
Registered
Joined
·
4,720 Posts
Dave,


I posted this sometime before. Where I work at

Lawrence Livermore National Lab, we have a setup that

is officially known as the Visualization Assessment Theater.

We call it the "Holodeck":
http://www.llnl.gov/str/Quinn.html


Greg
 

·
Registered
Joined
·
4,525 Posts
These are both awesome! Thanks for sharing the links!


My big question is: How is the image splitting performed?
 

·
Registered
Joined
·
4,720 Posts
Mark,


In the case of the "Holodeck" and its 15 projectors,

splitting the image really isn't a problem - it's assembling

the image in the first place.


The movies that are shown in the "Holodeck", like the

one on turbulence in the opening picture in the link - are

all calculated from output from physics simulation codes.


You have the data from millions of zone - that are put

together to form a picture. You never had a single pic.

to begin with.


More that likely, the calculation was run on one of the

massively parallel supercomputers - the typical parallel

paradigm is to give each processor a piece of the problem.


In all likelihood, each graphics processor for each projector is assembling the data from multiple calculational

processors on the the supercomputers. There are 15 projectors, whereas each node of the newest supercomputer

has 16 processors on it. The number of calculational nodes

are greater than the number of graphics processors or

the number of projectors.


So each of the 15 sub-images are assembled from many output

files from many processors.


Greg
 

·
Registered
Joined
·
4,525 Posts
Very cool, Greg. Since the 15 sub-images are formed from separate feeds, is it ever an issue keeping them all in-sync? Or is there some mechanism to keep them synchronized?
 

·
Registered
Joined
·
842 Posts
The real question is how do you glue the multiple

pictures together. The dila projectors have halo's

outside the picture array. The halo's would interfere

with the next projector. Do you have a way of masking

the halo's ???
 

·
Registered
Joined
·
133 Posts
Discussion Starter · #7 ·
The link about the holodeck mentions they are LCD not D-ILA projectors. I don't know if this is still true today or if that halo problem is also present with LCD projectors.


------------------

STOP DFAST NOW!!!
 

·
Registered
Joined
·
4,720 Posts
Mark,


As far as keeping the 15 graphics processors in sync -

yes - there is a mechanism to do that.


We have a similar mechanism in the physics simulation

codes where you have to sync thousands of processors.


Because the parallel paradidm is domain decomposition -

i.e. giving a portion of the problem to each processor -

there are regions of the problem where nothing much is

happening - yet - while a lot of physics is going on in

some other area.


Since it is much easier to calculate if nothing much is

happening - those processors with easy jobs could run away

from those with difficult jobs. However, there are sync

points in the calculation where calls are made to "barrier"

routines of the parallel processing libraries. At these

barriers, all processors are forced to sync before any can

proceed to the next phase of the calculation.


Although I don't program the graphics routines - I'm sure

there is a similar, if not identical, mechanism in place.


Greg
 
1 - 8 of 8 Posts
Status
Not open for further replies.
Top