Specs for 2 x 1080p

Does anyone have experience with running two heads off of one machine, both at 1920x1080? What kind of specs would you need to have some room to play with?

I’m looking at how best to meet a ridiculously large resolution (13500*1080) would be some combination of video and realtime, so vvvv would be nice to use if possible.

I´ve used a mac pro (8 or 12 cores, i don´t remember) 2,6 ghz, Nvidia GTX 285 in horizontal span mode, win xp 32. With animation stuff, no video.

With 13500 yo need 3,51 machines, could it be better with 3 machines 11520px x 1080?.

good luck

I saw that Lasal, really really nice work!

I think this project will be with four machines, the 13500*1080 is to leave extra room for the edge blends. I’d probably rather use a pc based setup- but that’s just from the perspective of trying to work with what I know (never ventured in to apple land). Is there any benefits with Mac? Issues with boygrouping macs?

I’ve done tonnes of stuff with 6 heads off one computer but that was 6x1024x768 no HD. Can suffer a bit if you try using heavy shaders and high vertex counts. You just need a really decent graphics card and very fast core speed on your processor.

I did a job recently with 4 x 720p outputs from 1 machine and even using some quite heavy content and it handled it well. That was with i7 core running at 4GHz and a GTX 580 3GB which kick ass BTW. Another job, same machine, 5 x 800 x 600 and around 10 render passes, so that’s around 24 million pixels being calculated every frame and still achieved 45fps!

All depends on content. If you’re doing lots of render passes you’ll def need tonnes of GPU memory and high clock speeds. CPU high frequency’s are important at high res’s to reduce CPU bottle necking. 4Ghz if you can.

Consider dropping down from 1080p to 720p if performance an issue, depends on install but most people won’t notice the difference. Also, high anti-aliasing will kill FPS at super high res, use 2 for early passes and 4 for late passes. Again, play with and see if you notice the difference when install live.

Also, develop content patches receiving a master back-buffer size coming from a S node and plumb into the renderer nodes. You can then tweak buffer size to help with performance. I then put a back-buffer multiplier on a S node and multiply this with some of my back-buffers in some render passes of my content. Often lower back-buffers in early passes really aren’t noticeable in final output, eps if using blurs. For instance set the multiplier to 0.7 and watch your framerate increase with out loss of quality.

How far the viewers are from the screen makes a big difference to how much they’ll notice too. There’s this whole thing about relative resolutions depending on viewers distance. Sometimes you can take advantage of this.

For me, it’s all about being about to tweak it quickly and easily for performance increases on site.

Hey everyoneishappy,

i don´t like macs to be used with windows there are a lot of inconveniences (no bios acces and not the last graphic cards), but in this firm they have only macs :P

@ lasal -phew that is a relief to hear :)

@ XD
Thanks for your post some nice tips in there!
So does the GTX 580 3GB run two heads and you are using matrox or similar to extend?

I was just checking out http://www.mviewtech.com/en/index.asp they have a version that apparently does 3 * 1080 @ 50hz. I’ve heard the boxes are actually quite good but I would be dubious about performance, as 1 x 1080 is already a lot of pixels.

yeah gtx 580 has two heads and can span 2 tripleH2Gs from it. You can configure tripleH2Gs to output 2 ports of HD formats. see http://www.matrox.com/graphics/en/products/gxm/th2go/resolution/resolutions.html for all possible. So you can use the 2 heads on the 580 or most nvidia cards to span across 2 tripleH2Gs and get a load of resolutions across either 4 or 6 displays from the card and tripleH2Gs. But yeah HD format do suddenly increase pixel count massively, eps 1080p so watch out. I wouldn’t be confident with 4 x 1080p from 1 PC unless doing some basic stuff.

One trick I used before is capturing realtime content to a video file and then playing that. A decent computer and GPU can output 1080p x 4. I’d recommend a SSD drive though.

I think kalle managed to get 12 x 720p from one machine using eyefinity ATI cards. any info here kalle?

6x HD works ok from one eyefinity 6 card. Editing and playing video at that resolution is a different issue. Core codec now supports decoding of 8kx8k but dont ask how to encode such a thing.

I’m looking at maybe building up 2 playback computers around HD6990 cards. Not sure if that eyefinity 6 card has specific design features for multiscreening that others don’t though. These new ones are higher specs (it’s 2 x 2G) and also say they can do up to five heads.

Sleepytom suggested these specs over on vjforums:

I’d be very curious if anyone has tried a HD6990 with vvvv yet…

VVVV can only address 4GB of ram (on a 64-bit system) so 6GB should be sufficient. See:system-recommendations#64-bit-operating-systems

hmm good point. On the other hand, ram is cheap compared to everything else, and it would be likely that the playback computer(s) will also be used for producing at least some of the content

that and also there are scenarios where running more than one vvvv instance could make sense on a 4-6 core system… (as bjoern said, each instance can adress up to 4gb in theory)

we had a total video resolution of 5120x800 on one machine:

2560x800 per gfx card (2x 1280x800 horizontal span),
2 gfx cards per pc (dual view),
2 videos per gfx card (fading between 2 fullscreen videos)
videos per gfx cards had to be read form different harddrives (10k rpm)

2.6ghz per core (8core), 3gb ram, 275gtx 1024mb vram

i’d say videowise it depends a lot on the codec. with our setup we had to sync the filestreams (because of the dual view) so we needed seekable videofiles.
and since nowadays most pcs are multicore it’s worth the time looking for a multicore decoding codec or gpu decoding thingy.

yes, see here: https://discourse.vvvv.org/

not working mister kalle.

try this one: