Page 1 of 1

Does AutoPano Giga support CUDA or OpenCL?

PostPosted: Wed Jan 09, 2013 4:22 pm
by Track
I'm buying a new laptop and I need to know which GPU rendering technology AutoPano Giga does/will support.

If it's CUDA, I'd have to get an nVidia GPU and if it's OpenCL, I'd have to get an AMD GPU.

Thank you.

PostPosted: Wed Jan 09, 2013 5:15 pm
by gkaefer
Track wrote:I'm buying a new laptop and I need to know which GPU rendering technology AutoPano Giga does/will support.

If it's CUDA, I'd have to get an nVidia GPU and if it's OpenCL, I'd have to get an AMD GPU.

Thank you.

cant give you an answer, but read this:
http://www.kolor.com/forum/t14548-fastest-gpu-for-autopano-gpu-acceleration
and take your conclusion...

Alexandre does say never to go propietary CUDA. Than he does say they're playing(?) with OpenCL currently (so they go propietary AMD road(?)...
Currently OpenGL is used for Preview only. and this works with my nvidia GPU too.
so for rendering/stitching itself it makes no difference, you can stick at any 50€ GPU Card with 2GB Ram. everything else is wasted money.

to also make a personal statement: I will never us an AMD GPU. never ever. For Blender Renderings these are far to slow compared with Nvidia GPUs and I have to use my PC not only for one task.

Georg

PostPosted: Wed Jan 09, 2013 6:28 pm
by Track
Actually, that's OpenCL. And if they use OpenCL, AMD will have the advantage.

Are you saying that if they DO use CUDA, they still won't use it for stitching? Why not? What good can CUDA do besides that?

EDIT:
The thread you linked me to had this,
"We will never go for CUDA, because it is proprietary. If we are going to use more GPU power, it will be through OpenCL for cross-hardware concerns. BTW : we already made test with OpenCL : it is really good."

Is that the word of an actual Kolor employee or just someone on the forums? How can we be sure?

PostPosted: Wed Jan 09, 2013 6:40 pm
by mediavets
Track wrote:The thread you linked me to had this,
"We will never go for CUDA, because it is proprietary. If we are going to use more GPU power, it will be through OpenCL for cross-hardware concerns. BTW : we already made test with OpenCL : it is really good."

Is that the word of an actual Kolor employee or just someone on the forums? How can we be sure?

That's the word of Alexandre Jenny one of the founders of Kolor.

PostPosted: Wed Jan 09, 2013 9:45 pm
by Track
mediavets wrote:
Track wrote:The thread you linked me to had this,
"We will never go for CUDA, because it is proprietary. If we are going to use more GPU power, it will be through OpenCL for cross-hardware concerns. BTW : we already made test with OpenCL : it is really good."

Is that the word of an actual Kolor employee or just someone on the forums? How can we be sure?

That's the word of Alexandre Jenny one of the founders of Kolor.

Ah, thank you. I e-mailed him. Hope he responds.

PostPosted: Thu Jan 10, 2013 2:40 pm
by Track
Still haven't received word from Mr. Jenny.

Just wanted to update that I am still looking for an answer, so I'm leaving this thread active and subscribed.

Thank you to whoever has any input for me.

PostPosted: Fri Jan 11, 2013 9:13 am
by ThomasV
Hi all,

Autopano use GPU only through OpenGL shaders. No OpenCL, no CUDA.
Why ? Mainly for compatibility reasons across architectures (Mac/Linux/Windows) since manufacturers does not provide same support on different archs.

Regards,
Thomas

PostPosted: Fri Jan 11, 2013 1:26 pm
by gkaefer
ThomasV wrote:Hi all,

Autopano use GPU only through OpenGL shaders. No OpenCL, no CUDA.
Why ? Mainly for compatibility reasons across architectures (Mac/Linux/Windows) since manufacturers does not provide same support on different archs.

Regards,
Thomas

so does this mean
1. we can buy our NVIDIAs or our AMD cards according our other preferences (3D blender, gaming, PS, price, whatever) and both are exchangable in quality and in autopano power?
2. Does autopano profit from more GDDR Memory?
3. can autopano profit from 2,3,4 graphic cards in one Computer (SLI by nvidia, Crossfire on Radeons) or does autopano only use the Memory of one graphic card?
4. is it planned for 3.1, 3.2,...,4.0 version that autopano also can use (by choice, automatically, by settings) the graphic card for rendering or other stages except the preview?

PS: why not supporting CUDA and OpenCL ;-)
whichever software I use on my Computer 16GB RAM Win7 64bit (currently boinc with several projects; blender rendering) even my outdated Nvidia GTS 260 card does perform decades better running on CUDA than on my 4+4 CPUs... so why should autopano not profit from these technologies?

thanks,
Georg

PostPosted: Fri Jan 11, 2013 2:00 pm
by ThomasV
Hello Georg,

1. we can buy our NVIDIAs or our AMD cards according our other preferences (3D blender, gaming, PS, price, whatever) and both are exchangable in quality and in autopano power?

Both are supported. I cannot give any opinion about quality of them. The gpu is used the same way for nvidia and amd, so at equivalent power, performance should be the almost equivalent.

2. Does autopano profit from more GDDR Memory?

If I understand well, GDDR is the momery located on the graphics card, and then is used to store textures and other graphics stuff, so yes we benefit from that.

3. can autopano profit from 2,3,4 graphic cards in one Computer (SLI by nvidia, Crossfire on Radeons) or does autopano only use the Memory of one graphic card?

No, Autopano only use 1 gpu.

4. is it planned for 3.1, 3.2,...,4.0 version that autopano also can use (by choice, automatically, by settings) the graphic card for rendering or other stages except the preview?

There are plans to improve performance of Autopano, and then using more the gpu is one way to achieve that.

Regards,
Thomas

PostPosted: Fri Jan 11, 2013 5:48 pm
by gkaefer
Hi Thomas,

thanks for your information. good to know point 3. so investing currently money in one card with more memory is not "wasted" money, whereas buying two lets say 4GB cards is wasted...
and I really hope this is on the todo/feature/whish list for a future version. One 8GB card for 4000euro or using 2x4GB cards for 2x500euro is not only a technical issue ;)

Liebe Gruesse,
Georg

PostPosted: Sun Jan 27, 2013 6:36 pm
by HansKeesom
gkaefer wrote:Hi Thomas,

thanks for your information. good to know point 3. so investing currently money in one card with more memory is not "wasted" money, whereas buying two lets say 4GB cards is wasted...
and I really hope this is on the todo/feature/whish list for a future version. One 8GB card for 4000euro or using 2x4GB cards for 2x500euro is not only a technical issue ;)

Liebe Gruesse,
Georg

So glad that I bought two 85 euro cards to find that out ;-) Crossfire did kick in but APG did nothing with the second card, not even when having two editors open, as situation where it should not be to difficult to use the second card for the second editor.........

i can understand using two cards for one editor is a bit hard, but alternating each next openen editor to another card should not be that much a trouble.

When editing lots of panorama's it is nice to alt-tab to another editor while waiting for one preview to finish. Just when I open up four ore more editors that one GPU is just not fast enough....

PostPosted: Sun Mar 24, 2013 12:23 pm
by HansKeesom
Hi Thomas,

Any idea when we can see crossfire/multiple GPU being used by AutoPano Giga?

It would really speed up my workflow.

PostPosted: Sun Mar 24, 2013 12:57 pm
by klausesser
HansKeesom wrote:It would really speed up my workflow.

It would speed up your editor. Nothing else afaik.

best, Klaus

PostPosted: Sun Mar 24, 2013 3:07 pm
by HansKeesom
klausesser wrote:
HansKeesom wrote:It would really speed up my workflow.

It would speed up your editor. Nothing else afaik.

best, Klaus

That is all I need.....plus a preview that can show me exactly how the end result of the rendering will look.

PostPosted: Sun Mar 24, 2013 11:28 pm
by SimDC
I recommend having a look at this video :

https://developer.nvidia.com/content/nvidia-waveworks-introduced-during-gtc-keynote

(2h30mn but very instructive !)

PostPosted: Tue May 28, 2013 11:34 am
by HansKeesom
klausesser wrote:
HansKeesom wrote:It would really speed up my workflow.

It would speed up your editor. Nothing else afaik.

best, Klaus

I am now working with a two machine setup. The fast and big one is my editing machine, the other one does the rendering (at low priority) and panotour work which is done remotely by my photographers.
This setup is nicely keeping us out of each others way as I can work at much higher speed. Still I would like to use crossfire. As George pointed out "One 8GB card for 4000euro or using 2x4GB cards for 2x500euro is not only a technical issue " and I would add, 3 or 4 cards of 85 euro would even be nicer ;-)

PostPosted: Tue May 28, 2013 1:15 pm
by klausesser
HansKeesom wrote:
klausesser wrote:
HansKeesom wrote:It would really speed up my workflow.

It would speed up your editor. Nothing else afaik.

best, Klaus

I am now working with a two machine setup. The fast and big one is my editing machine, the other one does the rendering (at low priority) and panotour work which is done remotely by my photographers.
This setup is nicely keeping us out of each others way as I can work at much higher speed. Still I would like to use crossfire. As George pointed out "One 8GB card for 4000euro or using 2x4GB cards for 2x500euro is not only a technical issue " and I would add, 3 or 4 cards of 85 euro would even be nicer ;-)

Only render-engines which are designed for that can have a boost from multiple graphic cards. APG isn´t - and i doubt a stitcher can be designed for it - beyond it´s (geometrical) editor-tasks. And even here i can´t see stacked cards working (two Quadro).

The best for APG afaik seems to be very much processor-speed and lots of RAM.
Render-farming or stacking multiple graphic cards will not get any advantage with APG as i understand it. Even my aged MacPro - Dual Xeons à  3Ghz, 8 cores, 32GB RAM is very fast with rendering Gigapixels. 4Gigapixels rendered as .PSB takes about 15-20min.
It´s the editor-speed which is very slow on my graphic-card resp. without the aged GPU. Here PTGui´s editor is MUCH faster using the same card . . .
On the other hand PTGui renders a bit slower (but i didn´t compare the different interpolators).

So i think that for *rendering* you better have multi-processors and multi-cores and as much RAM as you can put into your machine.

My PC with a GForce and two Quadro-cards (one is rented for testing), i7/ 8 cores, actually 64GB RAM is much faster in the *editor* . . . :cool:
I´m going to upgrade to a multi-processor and 128 GB RAM this year. But only after Apple has shown the upcoming MacPro . . . :D

best, Klaus

PostPosted: Tue May 28, 2013 11:19 pm
by HansKeesom
I agree that if you start from zero the best thing to do is to put all your money in one superfast machine.

But if you already have a machine, like you and me, and get to the point where there is no good way to expand it anymore, one can consider keeping that machine as a render server and buy a new machine and use that one for fast editing. Some parts can be moved to the new machine, I move a PCI-SSD to the editing machine, the old one is left behind with a sata-ssd which is of course slower.

Of course the fast machine can do rendering if you want to but if you want to editing at the same time it will slow you down.

With the right syncing software it is easy to keep the two machines in sync.

Interested to see prices of new macpro's. I don't think they will cost as little as the 1500 euro I invested in an upgrade kit which now runs a nice score in geekbench http://browser.primatelabs.com/geekbench2/1995060