Forums > Jitter

increase frame rate — hardware solution?

August 7, 2008 | 2:05 pm

I’ve got a g5 dual 2ghz 2gig ram running max 5 etc. Any suggestions regarding increased performance in jitter? Specifically I’m grabbing video from 2 cameras via 2 separate 1394a busses (unibraine fire i 640*480 uncompressed, not dv). I’m not doing any reading or writing to disk… it’s all video processing to trigger midi. So its all cpu. If I put in more ram, would that conceivably increase frame rate??

At the moment it’s pretty sluggish (one major factor is that openGL calls get referred back to the cpu rather than being processed on the video card as I have to do other post-openGL matrix processing and apparently mac os x won’t allow this to be done on the card).

or should i get a faster computer (would prefer not to for economic reasons)? How much faster would the frame rate be on an intel mac– any experience in this regard out there?

any advice Ta muchly

tm


August 7, 2008 | 4:13 pm

Hit there,

> [...] via 2 separate 1394a busses [...]
Sure about that? Because I believe there is only one FW bus (even if
more than one FireWire port) on the G5s, unless you add extras
independent ports from PCI cards.

What kind of framerate are you talking about? Have you tried it
without anything else, ie the simplest patch possible with two
jit.qt.grab. To help a bit, just post the patch you’re talking about.

Just for the record, my intel MacBookPro 2 GB will run 3 grabbers
simultaneously@640×480 , 30fps, using the embedded Isight and 2 USB
webcams (one in each port).

Also there is a possibility for me to test your code a G5 Quad with 8
GB of ram, so we can check if you’d get any gain from the memory.
And I believe that the Quad versus Dual wont change anything in what
your trying yo accomplish using Jitter.

Best,
JS

On 7-Aug-08, at 10:05 AM, Terry McDermott wrote:

>
> I’ve got a g5 dual 2ghz 2gig ram running max 5 etc. Any suggestions
> regarding increased performance in jitter? Specifically I’m
> grabbing video from 2 cameras via 2 separate 1394a busses
> (unibraine fire i 640*480 uncompressed, not dv). I’m not doing any
> reading or writing to disk… it’s all video processing to trigger
> midi. So its all cpu. If I put in more ram, would that conceivably
> increase frame rate??
>
> At the moment it’s pretty sluggish (one major factor is that openGL
> calls get referred back to the cpu rather than being processed on
> the video card as I have to do other post-openGL matrix processing
> and apparently mac os x won’t allow this to be done on the card).
>
> or should i get a faster computer (would prefer not to for economic
> reasons)? How much faster would the frame rate be on an intel mac–
> any experience in this regard out there?
>
> any advice Ta muchly
>
> tm


August 7, 2008 | 5:28 pm

> At the moment it’s pretty sluggish (one major factor is that openGL calls get referred back to the cpu rather than being processed on the video card as I have to do other post-openGL matrix processing and apparently mac os x won’t allow this to be done on the card).
>

if i understand you correctly, this is probably one of your main problems. *no* os will let you do matrix processing on the gpu. matrix processing is a cpu task. if you want to use the gpu you need to do your processing with shaders.

again, i may be misinterpreting what you’re saying, but no real way to help you without knowing more of what you’re doing


August 7, 2008 | 11:12 pm

Thanks for the response–

Yes I am using 2 busses; one via a pci-x card that I installed to add a port and one via the fw port on the front of the tower.
Quote: jsrousseau wrote on Fri, 08 August 2008 02:13

> Sure about that? Because I believe there is only one FW bus (even if
> more than one FireWire port) on the G5s, unless you add extras
> independent ports from PCI cards.

I can get 20fps max per camera with both cameras on together with the simplest grabber patch. That seems slow to me and indicates a hardware limitation. My patch is demanding, and things slow down to about 4fps.
>
> What kind of framerate are you talking about? Have you tried it
> without anything else, ie the simplest patch possible with two
> jit.qt.grab. To help a bit, just post the patch you’re talking about.
>

I cant post it right now, (gotta go on site) but will later tonight.

>
> Also there is a possibility for me to test your code a G5 Quad with 8
> GB of ram, so we can check if you’d get any gain from the memory.
This would be great. Much appreciated.

tm


August 7, 2008 | 11:19 pm

i’m using jit.gl objects for some of the processing; they’re basically adaptations of various help files, mainly the videoplane object, but then I have to route the output of these renders into a matrix. I get a message saying "reverting to cpu processing" or something similar.
t
Quote: robtherich wrote on Fri, 08 August 2008 03:28
—————————————————-
> > At the moment it’s pretty sluggish (one major factor is that openGL calls get referred back to the cpu rather than being processed on the video card as I have to do other post-openGL matrix processing and apparently mac os x won’t allow this to be done on the card).
> >
>
> if i understand you correctly, this is probably one of your main problems. *no* os will let you do matrix processing on the gpu. matrix processing is a cpu task. if you want to use the gpu you need to do your processing with shaders.
>
> again, i may be misinterpreting what you’re saying, but no real way to help you without knowing more of what you’re doing
—————————————————-


Viewing 5 posts - 1 through 5 (of 5 total)