Could someone show me a patch on how to spread processing over 2 GPU's ?

I got this MacPro with 2 graphic cards for that.

We have made our application with 2 gl.render context

By the way if someone knows how to do that I can offer paiement.

I got VRAM performance issue.
Everything goes on one GPU and we need 2
I got this MacPro with 2 graphic cards for that.

We have made our application with 2 gl.render context
but it doesn't spread over the 2 GPU's

urgent-help-how-to-use-2-gpus

here's some info from apple that might help:

https://developer.apple.com/library/mac/technotes/tn2229/_index.html

https://developer.apple.com/library/mac/technotes/tn2335/_index.html

The positions of the 2 jit.window's associated with the 2 render contexts is the crux here. You have 1 or more display outputs on the 1st gfx card, and 1 or more outputs on the 2nd. You have to know which display(s) connect to each gfx card. Check the positioning relative to each other in OSX' display preferences.

When you create your 2 jit.window's make sure they are positioned so that the first only shows on display(s) connected to the first GPU, and the 2nd window only on the 2nd GPU.

You use either the pos and size attributes or the rect attribute of jit.window for that.

If GPU 1 has one display of 1920x1080 connected and GPU 2 as well, display 1 is main display and display 2 is positioned to the right of that, and you want your windows to cover the full displays, then give both windows a size of 1920 1080, window 1 position 0 0 and window 2 position 1920 0.

If GPU 1 has one monitoring display + one output projector connected, and GPU 2 a 2nd projector, all 1920 1080, then window 1 gets position 1920 0 and window 2 position 3840 0.

I 'd put a patch together but don't have time till tomorrow at best.

The positions of the 2 jit.window's associated with the 2 render contexts is the crux here. You have 1 or more display outputs on the 1st gfx card, and 1 or more outputs on the 2nd. You have to know which display(s) connect to each gfx card. Check the positioning relative to each other in OSX' display preferences. 

I 'd put a patch together but don't have time till tomorrow at best.


You can make your window span multiple display outputs on 1 GPU. For example 2 x 1920 1080, give it size 3840 0.

I'm doing a lot of tests and I see that the video memory isn't released when I change the resolution of the jit.window

Is there a way to make GPU release the video memory from Max ?

yes I'm spreading the 4 screens .. see the image

Why are you worried about the memory use? To me it doesn't seem to be your issue.

Did you make a basic patch that just creates your render contexts, the windows and a simple geometry for testing? That will help you a lot. And if you post it here we can help more effectively.

Plus, if it's here on the forum the next person having this issue can find the solution. Hopefully.

Did you make a basic patch that just creates your render contexts, the windows and a simple geometry for testing? That will help you a lot. And if you post it here we can help more effectively. 

Plus, if it's here on the forum the next person having this issue can find the solution. Hopefully.


To avoid misunderstandings: I'm not taking your 'job'. Too many projects on my hands already to give full-on support. I'd gladly look at your patch though, or make a simple multi-window renderer, though it seems you already have that.

I'm worried about the memory release because I'm calling patches at different moment according to data collection and turning them off when not needed so if it builds up the memory and doesn't release it could crash no ?

See here the snapshots .. I have reduced from 7680 x 1080 to 935 x 125

and the memory usage on the second snapshot shows no release even now after 5 minutes

See here the snapshots .. I have reduced from 7680 x 1080 to 935 x 125
and the memory usage on the second snapshot shows no release even now after 5 minutes


https://discussions.apple.com/thread/6483464

The Mac Pro cannot use both graphics card for output.

The second one is only used for computations using OpenCL.

So it seems that we basically have only one card we can use ?

According to this thread: https://discussions.apple.com/thread/6483464
The Mac Pro cannot use both graphics card for output.
The second one is only used for computations using OpenCL.
So it seems that we basically have only one card we can use ?


That's the gist of that Apple forum thread, yes. I'd say look up if other sources confirm that. If it is so then the multi-window stuff is of no concern to you.

About memory use: there's the saying "empty RAM is useless RAM". It might be that it keeps stuff in RAM until it needs that memory location for something else. Did you experience crashes that are clearly related to hitting a maximum of memory use?

Coincidentally, I reported a memory leak in jit.gl.model to C74 just yesterday. When repeatedly loading 3D model files it will crash Max at some point.

About memory use: there's the saying "empty RAM is useless RAM". It might be that it keeps stuff in RAM until it needs that memory location for something else. Did you experience crashes that are clearly related to hitting a maximum of memory use? 

Coincidentally, I reported a memory leak in jit.gl.model to C74 just yesterday. When repeatedly loading 3D model files it will crash Max at some point.


I know I'm repeating myself but creating a simple patch that shows your issue is the most effective way of getting help. You're in a hurry to get a solution but the one thing you haven't done is exactly that...

"Did you experience crashes that are clearly related to hitting a maximum of memory use?"

I'll try making that simple patch ... but what if it is not simple ?

"Did you experience crashes that are clearly related to hitting a maximum of memory use?" 

I'll try making that simple patch ... but what if it is not simple ?
Anyhow I'll try. Thank you so much !


Seems that there might be a solution here

Vade: "If you create a window with Jitter on one monitor, the GPU that the monitor is connected two will ‘own’ the content. Ie: its textures and memory will be resident on that GPU*"

https://cycling74.com/forums/how-to-select-the-gpu-for-opengl-rendering/

Yes that's exactly what we 've been trying to tell you with the window positioning. But according to that Apple forum thread it doesn't come into play in case of your Mac Pro. If that info is correct your 2nd GPU only does openCL (not openGL) processing, which standard Jitter objects don't make use of.

For making a simplified patch what you do is delete things step by step till the crash doesn't occur anymore. The previous object(s) you removed is likely the cause of it.

Thanks for your concern. We are reprogramming the whole structure in Max7

I had gl.slab causing weird vram leaks in the old structure (cc.uyvy2argb.jxs) with 6.1.9.

We have encountered similar problems with graphic card ... and then installed max7.0.3.

It seems that gl.node is "picky". Not behaving as expected all the time. But after the new install

of Max things were better but not running a high enough fps.

So since friday we have added Syphon to the structure with different standalones.

That seems to work beautiful for now. No GPU leaks and CPU on one patch at 48fps running

5 screens 1920 x 1080 crossfading 2 jit.movie (so 10 jit.movie). with a standalone patch

having a gl.mesh with gl.pix feedbacks running at 35fps.

We have encountered similar problems with graphic card ... and then installed max7.0.3.
It seems that gl.node is "picky". Not behaving as expected all the time. But after the new install
of Max things were better but not running a high enough fps.

So since friday we have added Syphon to the structure with different standalones.
That seems to work beautiful for now. No GPU leaks and CPU on one patch at 48fps running
5 screens 1920 x 1080 crossfading 2 jit.movie (so 10 jit.movie). with a standalone patch
having a gl.mesh with gl.pix feedbacks running at 35fps.

I'm getting my head out of the water ;-)


Cool, streamlining complex systems is a whole project in itself...

Now I can work on my content a bit more !

Yes totally ... ouf so much work !
Now I can work on my content a bit more !


@Marie-Helene : i'm not sure it's relevant in your case or useful, but just in case : did you know you can run several instances of the Max application without needing to create a standalone ?

 in the Terminal for each new instance you want

It's very buggy though... i just tried on Yosemite and all the patchers windows have a black background, the inspectors and max console window have a size of 0 0 0 0... with max 6 it seems to work alrighty.

Sorry, after test, this seems less useful than ever.

@Marie-Helene : i'm not sure it's relevant in your case or useful, but just in case : did you know you can run several instances of the Max application without needing to create a standalone ?
you need to type something like open -n /Applications/Max7.app/ in the Terminal for each new instance you want
It's very buggy though... i just tried on Yosemite and all the patchers windows have a black background, the inspectors and max console window have a size of 0 0 0 0... with max 6 it seems to work alrighty.
Sorry, after test, this seems less useful than ever.


@MH: How are you dealing with the 2 GPU's now? Is the 2nd of any use?

And are you submitting bug reports for those mem leaks?

@MH: How are you dealing with the 2 GPU's now? Is the 2nd of any use?
And are you submitting bug reports for those mem leaks?


@vichug: ooh it 'd be great if that could work reliably...

indeed :/... i suppose there are some new Apple idiosyncracies that render that impossible, unless you explicitly code it for your application, what with all the Apple will to move away from everything that is natively Unix-based

We did everything to send to 2 GPUs and used this solution from MJ

But even with OpenGL driver I have no way to analyze if it's spreading or not over 2 GPUs. OpenGL Dirver only shows one GPU. But everything runs smooth right now. We place the windows before starting metro. When we don't all the images are gone and the plane gets yellow ... the video card is clearly having jit.fbo problems. So I guess this is our proof that we are using multiple GPU.

We had another problem of drawing and yesterday we had to make sure all the gl had a @draw_to.

Thanks for the info ... we did that and had no success with our project. But that doesn't mean it wouldn't be a solution for a different project.

We did everything to send to 2 GPUs and used this solution from MJ
https://cycling74.com/forums/how-to-select-the-gpu-for-opengl-rendering/

But even with OpenGL driver I have no way to analyze if it's spreading or not over 2 GPUs. OpenGL Dirver only shows one GPU. But everything runs smooth right now. We place the windows before starting metro. When we don't all the images are gone and the plane gets yellow ... the video card is clearly having jit.fbo problems. So I guess this is our proof that we are using multiple GPU. 

We had another problem of drawing and yesterday we had to make sure all the gl had a @draw_to. 

VICHUG
Thanks for the info ... we did that and had no success with our project. But that doesn't mean it wouldn't be a solution for a different project.


URGENT HELP - How to use 2 GPU's ?