100 Million Pixels of Virtual Reality
Roland Piquepaille writes "It's ironic that Iowa State University (ISU) announced a big upgrade of its C6 virtual reality (VR) room the same day as SGI filed for bankruptcy. Back in 2000, this 10x10x10 foot room was powered by SGI Onyx2 computers. The new version of this six-sided VR room will use 96 graphics processing units from Hewlett-Packard. And with its 24 Sony digital projectors, the researchers at ISU will immerse themselves into images of about 100 million pixels in the most realistic VR room in the world. Of course, this upgrade is not cheap. But with this $4 million addition, this new C6 should lead to new advances in urban planning, genetics, engineering or unmanned aerial vehicles."
I can only comment about the API - we're using something that is a standard (for us) and that fills in as nice middleware: VRJuggler. It sits atop (among other things) OpenGL.
There are no head-mounted displays that will deliver anything like the pixel resolution of a system like this. You simply wouldn't get the detail. And, the data infrastructure for this kind of project (where the aim is to visualize complex data) is not possible on "relatively standard class hardware". Another thing: In a cave environment like this, if you turn your head, the graphics view is updated only slightly or not at all. With a head-mount display, the whole scene has to swing round when you turn your head. If there's any latency in head-tracking (likely) or graphics rendering (possible), then the cave is much less unsettling per head-turn than is the head-mounted display. Less nausea. And another thing: you get a much larger peripheral view in a cave, leading to better understanding of context. Undisclaimer: I work for HP. :-)
Your missing a piece of this though.
Its in 3d.
Doing 3d is no big deal for a small screen when the viewer is in a fixed perspective, but when you ware walking around the room the images have to change to keep the proper 3d perspective. Doing all of that for a 6 sided room in high deffinition and on-the-fly takes some serious horse power.
(BTW, I was in it in 1999 when it was 4 sided (floor and 3 walls))
If you're doing it on the cheap and only have three or four projectors, you don't need much of a cluster, just a three or four networked computers. Or, use two dualhead computers.
. asp
You'll have a small amount of lag in the syncronization (network + OS + application software) but with some tweaking of the OS network configuration, or using some insanely fast system rather than a network (shared memory backplane?), you might get it to a few ms?
If you want frame-by-frame synchronization you need some specialized equipment driving the projectors, stuff like this: http://www.es.com/products/image+generators/index
(Anyone making a homebrew CAVE want to try using http://interreality.org/ VOS software in it?)
Don't put advice in your sig.