VR with a single computer

[Update 19 Nov 2009: Added link to ATI Eyefinity card]

As an exercise, I was thinking about all the VR systems you could create using only one single computer, ie no cluster.

I will not be able to cover all the possible combinations, but I will expose several issues that need to be addressed.

I’ll also try to update this page based on up-to-date information.

• Introduction

Several limitations or bottleneck need to be considered :

On a purely physical aspect :

- Number of graphic outputs : How many projectors / screens can you plug ?

- Numberof input ports : How many serial / usb / etc ports are available ?

Regarding computing power :

- Number of CPUs

- Number of GPUs

- Buses bandwidths : Texture upload, tracking data …

On a software level :

- Is your pipeline multi-thread (support of multiple CPUs) ?

- Is your 3d engine multi-pipe (support of multiple GPUs ) ?

- What is the bottleneck of your application ? Is it CPU bound ? GPU bound ? … This aspect is the most overlooked by clients

• Systems using only one graphic output

The following system only use one VGA or DVI plug.

Fishtank VR

Fishtank VR is VR with a small screen. It can be an old CRT monitor with active glasses, a new Zalman screen with passive glasses, a DLP 3D screen, the new nvidia 3D Vision system using LCD screens and active glasses …

Single projector wall

With only one graphics output, the wall will be in mono or using active stereo (OpenGL/Quadro only, or both OGL&DirectX with Geforce (and some limitations) if you use nVidia 3D Vision system).

Using two graphics outputs the wall can use passive stereo that use polarized glasses, See “With two graphics outputs”

HMDs

Head mounted displays, also known as VR goggles, require most of the time either one video input (with active stereo, ie alternate frame rendering), or two video inputs (one for each eye).

J-Dome

The J-Dome only requires one projector :

R-Screen

The Clarte’s R-Screen only requires one projector :

Panoscope

The Panoscope only requires one projector :

Panoscope (c) Laval Virtual

Virtual Showcase

Esiea’s Virtual Showcase only requires one projector :

• With two graphics outputs

The two graphics outputs can be on the same graphics card or on two different graphics card. Choosing between one or the other is a matter of computing power, budget and ease of programming. Using two graphics cards will improve the performances if the GPU is the bottleneck, but it can also add some coding complexity. If you’re using two GPUs it would also be better to multithread your 3d pipeline and use one CPU per GPU.

Holostage

The holostage is a L shaped wall. It requires two stereoscopic projectors, ie two active stereo outputs.

An HoloStage © Arnaud Muthelet

Workbench

The workbench looks a lot like the holostage, but is much smaller :

Holobench – Photos © PSA Peugeot Citroën Mediatheque.

Passive stereo wall

A passive stereo wall uses two projectors (one for each eye) with polarized filters and matching polarized glasses.

Stereo HMDs  with two inputs

Those HMDs have two video inputs, one for each eye.

Passive stereo screens

Planar 3D SD2620W 26″ Widescreen Stereo/3D Monitor

Using a semi-transparent, polarized mirror, this kind of screen is used with passive glasses.

• More than two graphics outputs

If you have two graphics cards theoretically you have four outputs. Several solution exist to extend that number :

- The Nvidia Quadroplex systems can have up to 4 GPUs (and 8 outputs) to be used by one computer.

- The ATI Eyefinity card has 6 outputs.

- Cyviz xpo.3 image splitter allows to take one very high resolution image and split it to several projectors.

• Driving a Cave with a single computer ?

With the increase in CPU cores and using several Quadroplex it seems reasonable to think that we should be able to drive a small (in projector number) CAVE with one computer in the near future although there is some software complexity that still needs to be adressed.