sascha pohflepp

just a quick ping about performance and multiple screens:

we got a 2x2GHz G5 here which unfortunately just has PCI/PCI-X slots. 

we need to run one screen and 4 projections from that machine and 

apparently the only way to go seems to be using relatively slow PCI 

question: how much do the cards influence the overall performance. 

will stuff be unbearably slow once something is on one of the slow 

card's screens (as recently experienced with openGL in processing) or 

are the processors most important part here?

we got a 2x2GHz G5 here which unfortunately just has PCI/PCI-X slots.  
we need to run one screen and 4 projections from that machine and  
apparently the only way to go seems to be using relatively slow PCI  
graphics cards.

question: how much do the cards influence the overall performance.  
will stuff be unbearably slow once something is on one of the slow  
card's screens (as recently experienced with openGL in processing) or  
are the processors most important part here?

         www.blinksandbuttons.net
blog.plugimi.com
AIM + skype: plugimi

multiple-screens-performance

On Oct 17, 2006, at 12:40 PM, sascha pohflepp wrote:

> we got a 2x2GHz G5 here which unfortunately just has PCI/PCI-X 

> slots. we need to run one screen and 4 projections from that 

> machine and apparently the only way to go seems to be using 

> question: how much do the cards influence the overall performance. 

> will stuff be unbearably slow once something is on one of the slow 

> card's screens (as recently experienced with openGL in processing) 

> or are the processors most important part here?

Sending large amounts of data (e.g. textures and huge vertex arrays) 

across the 1x PCI bus as opposed to the faster AGP bus will be slow, 

but the rendering itself should exhibit decent performance. In 

Jitter, I'd recommend using display lists whenever possible for such 

a setup, and don't use monitor spanning windows unless they are both 

driven by the the same GPU. Another thing to keep in mind is that VBL 

sync will degrade your framerate by an integer multiple, one for each 

screen. So either turn off sync or use something like the Matrox 

triple head (this will also let you connect up to 6 displays from one 

dual head AGP card that has enough VRAM). Link to matrox triple head 

below (in the instance of the triple head you really only have two 

monitors, it gets cut up after the fact, and you're better off using 

one large window for each "triple monitor" output).

http://www.matrox.com/graphics/offhome/th2go/home.cfm

More info on this is in the Jitter forum archives. search on "matrox" 

> we got a 2x2GHz G5 here which unfortunately just has PCI/PCI-X  
> slots. we need to run one screen and 4 projections from that  
> machine and apparently the only way to go seems to be using  
> relatively slow PCI graphics cards.
>
> question: how much do the cards influence the overall performance.  
> will stuff be unbearably slow once something is on one of the slow  
> card's screens (as recently experienced with openGL in processing)  
> or are the processors most important part here?

Sending large amounts of data (e.g. textures and huge vertex arrays)  
across the 1x PCI bus as opposed to the faster AGP bus will be slow,  
but the rendering itself should exhibit decent performance. In  
Jitter, I'd recommend using display lists whenever possible for such  
a setup, and don't use monitor spanning windows unless they are both  
driven by the the same GPU. Another thing to keep in mind is that VBL  
sync will degrade your framerate by an integer multiple, one for each  
screen. So either turn off sync or use something like the Matrox  
triple head (this will also let you connect up to 6 displays from one  
dual head AGP card that has enough VRAM). Link to matrox triple head  
below (in the instance of the triple head you really only have two  
monitors, it gets cut up after the fact, and you're better off using  
one large window for each "triple monitor" output).

More info on this is in the Jitter forum archives. search on "matrox"  
or "triplehead"

today i tried out the setup with two ATI radeon 9200 but it's pretty 

slow so far, currently trying to optimize. the size of the matrices 

seems to have the most impact, so it's kinda smooth with 320 x 240, 

which is not really great. any more ideas?

>> we got a 2x2GHz G5 here which unfortunately just has PCI/PCI-X 

>> slots. we need to run one screen and 4 projections from that 

>> machine and apparently the only way to go seems to be using 

>> question: how much do the cards influence the overall performance. 

>> will stuff be unbearably slow once something is on one of the slow 

>> card's screens (as recently experienced with openGL in processing) 

>> or are the processors most important part here?

> Sending large amounts of data (e.g. textures and huge vertex 

> arrays) across the 1x PCI bus as opposed to the faster AGP bus will 

> be slow, but the rendering itself should exhibit decent 

> performance. In Jitter, I'd recommend using display lists whenever 

> possible for such a setup, and don't use monitor spanning windows 

> unless they are both driven by the the same GPU. Another thing to 

> keep in mind is that VBL sync will degrade your framerate by an 

> integer multiple, one for each screen. So either turn off sync or 

> use something like the Matrox triple head (this will also let you 

> connect up to 6 displays from one dual head AGP card that has 

> enough VRAM). Link to matrox triple head below (in the instance of 

> the triple head you really only have two monitors, it gets cut up 

> after the fact, and you're better off using one large window for 

today i tried out the setup with two ATI radeon 9200 but it's pretty  
slow so far, currently trying to optimize. the size of the matrices  
seems to have the most impact, so it's kinda smooth with 320 x 240,  
which is not really great. any more ideas?

>> we got a 2x2GHz G5 here which unfortunately just has PCI/PCI-X  
>> slots. we need to run one screen and 4 projections from that  
>> machine and apparently the only way to go seems to be using  
>> relatively slow PCI graphics cards.
>>
>> question: how much do the cards influence the overall performance.  
>> will stuff be unbearably slow once something is on one of the slow  
>> card's screens (as recently experienced with openGL in processing)  
>> or are the processors most important part here?
>
> Sending large amounts of data (e.g. textures and huge vertex  
> arrays) across the 1x PCI bus as opposed to the faster AGP bus will  
> be slow, but the rendering itself should exhibit decent  
> performance. In Jitter, I'd recommend using display lists whenever  
> possible for such a setup, and don't use monitor spanning windows  
> unless they are both driven by the the same GPU. Another thing to  
> keep in mind is that VBL sync will degrade your framerate by an  
> integer multiple, one for each screen. So either turn off sync or  
> use something like the Matrox triple head (this will also let you  
> connect up to 6 displays from one dual head AGP card that has  
> enough VRAM). Link to matrox triple head below (in the instance of  
> the triple head you really only have two monitors, it gets cut up  
> after the fact, and you're better off using one large window for  
> each "triple monitor" output).

I have a setup that drives 3 screens from one graphics card using the Matrox Triple Head 2Go

http://www.matrox.com/graphics/en/gxm/products/th2go/home.php

I create one wide matrix (3072 x 768) that I populate within Jitter and then feed to the Triple

Head and it splits it into three 1024 x 768 vga monitor feeds. Works great and allows you to move

items accross the large internal matrix with impunity and still have them stitch together on the 3

> today i tried out the setup with two ATI radeon 9200 but it's pretty 

> slow so far, currently trying to optimize. the size of the matrices 

> seems to have the most impact, so it's kinda smooth with 320 x 240, 

> which is not really great. any more ideas?

> >> we got a 2x2GHz G5 here which unfortunately just has PCI/PCI-X 

> >> slots. we need to run one screen and 4 projections from that 

> >> machine and apparently the only way to go seems to be using 

> >> question: how much do the cards influence the overall performance. 

> >> will stuff be unbearably slow once something is on one of the slow 

> >> card's screens (as recently experienced with openGL in processing) 

> >> or are the processors most important part here?

> > Sending large amounts of data (e.g. textures and huge vertex 

> > arrays) across the 1x PCI bus as opposed to the faster AGP bus will 

> > be slow, but the rendering itself should exhibit decent 

> > performance. In Jitter, I'd recommend using display lists whenever 

> > possible for such a setup, and don't use monitor spanning windows 

> > unless they are both driven by the the same GPU. Another thing to 

> > keep in mind is that VBL sync will degrade your framerate by an 

> > integer multiple, one for each screen. So either turn off sync or 

> > use something like the Matrox triple head (this will also let you 

> > connect up to 6 displays from one dual head AGP card that has 

> > enough VRAM). Link to matrox triple head below (in the instance of 

> > the triple head you really only have two monitors, it gets cut up 

> > after the fact, and you're better off using one large window for 

I have a setup that drives 3 screens from one graphics card using the Matrox Triple Head 2Go
splitter. 

I create one wide matrix (3072 x 768) that I populate within Jitter and then feed to the Triple
Head and it splits it into three 1024 x 768 vga monitor feeds. Works great and allows you to move
items accross the large internal matrix with impunity and still have them stitch together on the 3
external screens.

> today i tried out the setup with two ATI radeon 9200 but it's pretty  
> slow so far, currently trying to optimize. the size of the matrices  
> seems to have the most impact, so it's kinda smooth with 320 x 240,  
> which is not really great. any more ideas?
>
> what are display lists?
>
>
> cheers
> sascha
>
>
>
> >> we got a 2x2GHz G5 here which unfortunately just has PCI/PCI-X  
> >> slots. we need to run one screen and 4 projections from that  
> >> machine and apparently the only way to go seems to be using  
> >> relatively slow PCI graphics cards.
> >>
> >> question: how much do the cards influence the overall performance.  
> >> will stuff be unbearably slow once something is on one of the slow  
> >> card's screens (as recently experienced with openGL in processing)  
> >> or are the processors most important part here?
> >
> > Sending large amounts of data (e.g. textures and huge vertex  
> > arrays) across the 1x PCI bus as opposed to the faster AGP bus will  
> > be slow, but the rendering itself should exhibit decent  
> > performance. In Jitter, I'd recommend using display lists whenever  
> > possible for such a setup, and don't use monitor spanning windows  
> > unless they are both driven by the the same GPU. Another thing to  
> > keep in mind is that VBL sync will degrade your framerate by an  
> > integer multiple, one for each screen. So either turn off sync or  
> > use something like the Matrox triple head (this will also let you  
> > connect up to 6 displays from one dual head AGP card that has  
> > enough VRAM). Link to matrox triple head below (in the instance of  
> > the triple head you really only have two monitors, it gets cut up  
> > after the fact, and you're better off using one large window for  
> > each "triple monitor" output).
>
>
>
>
>
>
>          www.blinksandbuttons.net
> blog.plugimi.com
> AIM + skype: plugimi
>
>
>

Keith McMillen 
BEAM Foundation 
http://www.beamfoundation.org/ 
510.502.5310


I'm pretty sure it has come up a few times before but couldn't find it.

I have a patch that needs to be split to two beamers, each on a resolution of 1280x768 making a 32/9 screen I don't have a triplehead from matrox and when I order it will take some time to get here do that's not an option at the moment.

I was wondering if my plain G5 with build-in AGP card (64mb I'm afraid) would allow spanning to two screens. I know you shouldn't span on diffrent graphic cards and trying on a macbook (maybe dumb to do since it hasn't got any real GPU) it doesn't work. I can't test it on the G5 but need to know if this will work. Can anyone lead me to an aproach that will work?

multiple screens performance