RealFlow 2013 technology preview at FMX!

[Aniki]

after affects and nuke

Yes, and After Effects is so much easier to use..

[mdonovan]

late here as well ... NODES > solo property pages with lists ... you can see the relationships much better.

[dmeyer]

I know I'm a little late in the conversation but I prefer a node system as oppossed to a list, I can sort out everything better. its like a difference between using after affects and nuke

+1

[dmeyer]

Gus,

Any details on the GPU support in RF 2013?

Is it primarily single or double precision ops? OpenCL or CUDA? Memory an issue? How about scaling across GPUs?

Would be helpful in selecting hardware.

[gus]

Hi Doug,

Here is our currrent public info on RF2013 and GPUs...

http://nvidia.fullviewmedia.com/siggrap ... SB127.html

g-)

[dmeyer]

Gus,

Any details on the GPU support in RF 2013?

Is it primarily single or double precision ops? OpenCL or CUDA? Memory an issue? How about scaling across GPUs?

Would be helpful in selecting hardware.

Ping. Hoping for some details!

[atena]

Hello, despite the fact that most of the GPUs nowadays support double precision we decided to do single precision for compatibility with all GPUs, new ones and old ones. It doesn't affect the simulation at all.

It is OpenCL, in the last slide we gave some implementation bits.

Memory could be an issue but normally is not because we use the GPU as an additional CPU, meaning we split the computational load and only one part is going to the GPU avoiding for the GPU to have in memory all the simulation data. This is mentioned in the slides as well.

Multiples GPUs? The implementation takes care of that, but we haven't tried that in our office as we don't have a machine ready for that.

Cheers,

[dmeyer]

Hello, despite the fact that most of the GPUs nowadays support double precision we decided to do single precision for compatibility with all GPUs, new ones and old ones. It doesn't affect the simulation at all.

It is OpenCL, in the last slide we gave some implementation bits.

Memory could be an issue but normally is not because we use the GPU as an additional CPU, meaning we split the computational load and only one part is going to the GPU avoiding for the GPU to have in memory all the simulation data. This is mentioned in the slides as well.

Multiples GPUs? The implementation takes care of that, but we haven't tried that in our office as we don't have a machine ready for that.

Cheers,

Thanks for the details. Looking forward to trying it. Let me know if need beta tester as we have tons of different hardware.

[LuisMiguel]

My two cents

The nodal system is extremely easy to work now. We don't need to think if I can link a node with another node, simply if the nodes can be connected you'll be able to do it, if not, you won't.

luisM.