» 20x1920x1050 on one machine
This site relies heavily on Javascript. You should enable it if you want the full experience. Learn more.

20x1920x1050 on one machine

4 ati graphic cards with 6 outputs each.

performance is ok but drops faster if you use complex geometry.need to make more tests. with fts vlc plugin its also possible to play videos across cards.
advantage compared to boygrouping is that you dont have to sync videos.

need to make more tests to find out about the pros and contras compared to boygrouping.

elektromeier, Wednesday, Dec 21st 2011 Digg | Tweet | Delicious 9 comments  
circuitb 22/12/2011 - 03:48


viktor2 22/12/2011 - 22:53

what mainboard?

xd_nitro 23/12/2011 - 01:48

is this a single span across all the screens? i.e. one full screen renderer?

manuel 23/12/2011 - 19:48

and what ati ?

I think you can make 1 span for each gpu

elliotwoods 23/12/2011 - 23:21

@xd_nitro - if you could span all with 1 renderer (generally possible in window mode, but generally not fullscreen, even with Quadro)
then all renderering would be performed on 1 card, and the output 'texture' would be 'shipped' across the PCI-E bus to each of the other cards

since you'd already be sending a lot of video data up to the graphics cards if you're doing video playback like in this example* you'd saturate your PCI-E bus and get unhappy fps

  • (with current methods, we're shipping uncompressed video from cpu to gpu for video playback outside of DirectShow)
elektromeier 24/12/2011 - 03:36

its 4 eyefinity groups, so its 4 devices=4 renderers.

we used the Radeon HD 6870 Eyefinity 6, 2GB.

as elliot mentioned theres a lot of traffic on the pci-e bus when streaming videos to textures, thats why we choosed this exotic board which can provide pci e 16x on all 4 slots. frame rates are ok if you dont do to complex stuff and surprislingly we can play quite some videos with a good frame rate. on the other hand framerate drops fast with some amount of geometry which i dont really understand as its loaded to the gpu just once.

we tried this as an experiment and we thought its an advantage that we dont have to sync videos over network.
i also experienced not so smooth animations via boygrouping, except if you really damp your values quite alot.

so for our application this solution works nice but it sure have also it drawbacks...

valyard 24/12/2011 - 18:16

1. why 20 monitors not 6x4 = 24?
2. why 4 eyefinity groups? internet says that all 4 cards are used in rendering of one frame. they render square by square like a checkerboard.

jyj2002kr 10/02/2012 - 00:44

Wow~ good!

  • 1

anonymous user login


~7h ago

xd_nitro: will do Joreg. @cat - poss but can't move to 50 anyway.

~12h ago

catweasel: zoom-pan-rotate-touchgestures @XD_nitro, could VL do what you need?

~15h ago

joreg: @xd_nitro: from looking at that patch i don't really understand what it is doing: can you start a forumthread and specify your needs?

~17h ago

xd_nitro: oh it's a contribution not forum. doh! multitouch-stack-revived

~17h ago

xd_nitro: hey y'all - anyone up for doing some sponsored work on multitouch-stack-revived?

~1d ago

tonfilm: @all please update your VVVV.OpenVR pack, important bugfix: vvvv.openvr #htc #vive #oculus #rift #vr

~2d ago

~2d ago

id144: @joreg i'm not sure where to start with the requests, everything i know is wrong and yoghurt

~2d ago

joreg: only 5 requests but already 20 proposals for #node17 workshops. keep'em coming: node17-call-for-workshops #vvvv #vl #visualprogramming

~2d ago

joreg: the 2016 numbers are out: vvvv-in-numbers-2016 thanks everyone for contributing #vvvv