Slashdot Mirror
Triple Headed Desktop Display for Fast 3D Apps?
Once Was SGI Customer asks: "My group was once a big SGI user. We run a Powerwall display center with 3 large screens, currently driven by an SGI Onyx Infinite Reality (IR) to provide a single desktop with 3D acceleration across all screens. The Onyx is now old and very slow compared to our Nvidia cards, that do a great job at TwinView display, but not 'TripleView'. I'd like to know if there are any PC manufacturers who make a card that can do what the IR can do (in terms of a single desktop across 3D displays with fast 3D acceleration), but for Windows and Red Hat (now Fedora Core) Linux?"
Get a machine with Graphics on the motherboard. Add in a top of the line ATI car for dual monitor support and blammo! 3 monitors ready to go. Works in windows, not sure about *nix though. -Rick
"Most people in the U.S. wouldn't know they live in a tyrannical state if it walked up and grabbed their junk." - MyFirs
http://www.matrox.com/mga/products/qid/home.cfm am i wrong or does this satisfy what you are looking for...
If I knew what I was talking about, there would prolly be more text.
Now that motherboards are supporting multiple video cards, wouldn't it just make more sense to put in two good dualhead cards, instead of struggling to find a single triple head card?
Just get a good PCI Express-based machine, and fit it with twin dualheads. You can even add a fourth monitor if you want it.
Remember, there were no nuclear weapons before women were allowed to vote.
Two widescreen monitors would provide the same amount of screen real estate as three normal monitors.
Simple, get one of the newer motherboards that supports SLI Nvidia cards and plop in a pair of AGP GeForce 6600's or 6800's, instant support for up to 4 monitors if you run in non-SLI mode.
actually http://www.matrox.com/mga/multidisplay/product_cha rt.cfm
this shows a breakdown of all their multidisplay cards.
If I knew what I was talking about, there would prolly be more text.
but just add another graphics card.
You can put in as many NVIDIA 6600 PCI cards as you have slots. Each of those can drive two panels. I've got clients with 6 panel desktops.
You can mix AGP and PCI, but depending on BIOS/MB you may have issues. I've seen mix mode work and not work. Seemed to work fine on Dells.
Often time, the built in MB GPU can not be enabled if you're also using an AGP card (because the mboard on-board GPU is using the AGP bus). So be mindfull of that if you go down that path.
-Malakai
A Dragon Lives in my Garage
you may want to look at the Parahelia line from Matrox
They do triple-head out of the box, nice cards.
Sometimes people just have to learn and adapt to change, it is one of the requirements of being a living thing.
Matrox Parhelia supports 3 monitors on the one card, and works great on Windows.. I'm pretty sure they have Linux drivers as well.
I note that the latest Nvidia drivers for Linux have added 'initial support for Xinerama + OpenGL' - in other words, I gather you can have a single OpenGL context spanned over multiple graphics cards.
:-)
See Appendix V in the drivers README - I haven't tried it, but it sounds like you'll be able to expand to three or more heads, so long as the resulting window is less than 4096 pixels across.
Any use? I've only got experience with OpenGL on a single, dual-head graphics card thanks to Twinview, but I have to admit that works brilliantly for me. Who knows what this new thing is like.
Tedious Bloggy Stuff - hooray?
Get an NForce4 SLI board and run two GeForce PCIe graphics cards (or two ATI PCIe graphics cards). You will be able to support up to four monitors.
The Matrox Parhelia is slow, as are PCI (not PCIe) cards.
SGI machines are built to last. They are built to perform. Even if they aren't the fastest or most powerful computers out there, they are amongst the most reliable and industructable systems built. You might get minor performance boosts from an nVidia-based PC, but don't be surprised if that PC overheats and dies within a year. SGI has military-grade strength and quality. PC hardware does not.
Cyric Zndovzny at your service.
I don't know how well the multi-GPU support works with NVIDIA cards under Linux -- I can only testify that dual monitors from a single card works flawlessly. Others in this thread have indicated it should work well if you can get a new motherboard with dual PCI-E and slap a couple 6800s in it, but I haven't seen it in person.
What I do know is that you can go to a multi-node system and run DMX (Distributed Multiheaded X11), which was designed to run powerwall displays. You're in the unfortunate position of having a three-headed wall, as going to a multi box system feels like overkill for you. That said, three PCs with a gig-E interconnect is going to cost far less than one Onyx IR pipe, DMX scales well, and we've got much larger walls than that (I forget the number -- maybe 2x4) running beautifully on a cluster. It's stood up to demos to VIPs without crashing, and I believe that's the most failure-inducing state for any setup.
By the way, DMX is transparent to your application -- it looks just like a single X server with a single OpenGL context. Thus, it can be used with any existing OpenGL apps.
You could use ModVis to image a single data set across multiple monitors. They distribute the image across a cluster of hosts, each of which have a slightly different view. It might be overkill for just three screens, but works very well if you can't do it in hardware.
I can't help but think that this is a troll for people to astroturf Matrox's triple head cards. As a Montrealer I like Matrox but they dropped the 3D ball a few years ago.
Most Silicon Graphics systems are indeed built with military-grade strength and quality, especially the big old black Onyx beast and other Challenge L and XL machines from that era.
But iron like that is old. Really, really Old. The host system architecture dates back to 1993 and the original InfiniteReality 1 graphics date back to about 1995. In those days a wicked cool PC had, at best, a 166 MHz Pentium and a 8 MB Matrox PCI card. Most people were still using Windows 3.1 in VGA resolution on their 486's in 1995.
Times have changed. SGI is a has-been. Cray beats them on raw CPU and bandwidth performance (the version of UNICOS that runs on the Cray X1 is actually based on SGI's IRIX 6.5). SGI's graphics these days are just a bunch of ATI FireGL AGP cards.
For most people, a dual or quad opteron with a pair of PCI-E graphics cards (driving 2 monitors each for 4 monitors total) is plenty of power. You can even use SGI's wonderful OpenGL Perfomer software on Linux or Windows to ease the development pains of your multi-CPU, multi-headed graphics applications.
If you need a *single system* with 16 or more CPUs and a dozen or so graphics cards, then you may be a good candidate for one of SGI's Itanium2/FireGL powered Prisim monsters. Just don't expect it to be any more stable than a PC.
For anyone looking to do this with free software.. it has been around for YEARS. Infact, it was around before XFree86 supported Xinerama! Unfortunately, it does NOT do 3d acceleration; however, I'm providing this information for anyone else still interested.
j pg
Take a look at GGI + XGGI. Run an X11 server on each host and run XGGI, targeting each.
Take a look at the following screenshot:
http://www.ggi-project.org/resources/images/doom.
Now, imagine that each of those tiles was displayed on a separate physical display, each running X11. You can do it.
If your currently using SGI, then I would assume that your not looking for a gaming card. If your doing 3D design work, then you might look into The Wildcat Multiview Card. It only works in conjunction with the Wildcat Realizm cards. Not cheap, at $825 for a midrange card, and nearly $3,000 for the top-of-the-line card and another $420 for the multiview card, but then they are professional cards, not intended for hobbiest or gaming machines.
Famous Last Words: "hmm...wikipedia says it's edible"
Speaking as an SGI admin (Onyx3400 IR2 probably the closest to what you're using) I know where you're coming from.
a) Don't buy a Prism. It's just got 1 generation old ATI cards in, and the performance is... disappointing.
b) I reckon to not bother with a single card solution. We've used a Matrox Parhelia under windows, and mostly due to driver shoddiness it wasn't that great.
c) I'd be tempted (and will be testing a cheaper varient of this out soon to run an IBM 3840x2400 screen) to try a twin Quadro FX 4400 on one of the Nvidia Pro based boards (Tyan seem to be the leader with this at the moment) with twin 16x PCI-E. Nvidia have recently changed their drivers to allow you to use Xinerama and OpenGL across multiple identical cards. Seeing as you've come from SGI, this should be easy to sell financially. The performance is cracking on the FX4400 too, blowing everything else we've got out of the water. You'd manage to put together a dual Xeon 3.6 4Gb machine for something under 8,000 UK pounds.
jh
http://chromium.sourceforge.net/
Support Eachother, Copy Dutch Property!
The very latest NV drivers let you do full OpenGL across multiple heads/cards with X+Xinerama. I'm using this now across three separate cards and it seems to work great (at least for funky screensavers). So you might want to give that a go.
There are many graphics cards with Twin simultaneous monitor outputs designed for running two monitors at once now. They are nvidia and ATI and relatively cheap - especially on ebay. If you get two - one PCI and one AGP they will BOTH work simultaneously and you can usually get three or even four monitors going at once.
I don't have a single desktop in this configuration, as in Xinerama-style single X screen, however I had at some point single kayboard and mouse, and the same computer running applications displayed on three physical screens, mapped to two or three X screens. This is how it looked (laptop's keyboard and trackpad work but aren't used), and this is how it was done. Two monitors are handled by a dual-monitor nvidia card, and can be configured either as two or a single screen spanning both.
This configuration doesn't allow windows to span between "local" and "remote" screens, however for many purposes this is useful -- in a different setup I often run 3D CAD on one screen, and 2D CAD an the other one, and obviously there is little point in mixing those two. Separate virtual desktops on all monitors also help -- I usually have 4x3 viewports in Sawfish, so with three monitors that would be 36 semi-independent viewports.
Contrary to the popular belief, there indeed is no God.
While we are on the subject, are there any people who could recommend displays that are tiled in such a way tht the LCD panels are either directly next to each other or at the least have a very thin edge between them?
The dangers of excessive individualism are nothing compared to the oppressiveness of excessive collectivism