PDA

View Full Version : realtime smoke, Nvivia Cuda



erikals
03-26-2009, 01:00 PM
Volumetric particle shadows

http://www.youtube.com/watch?v=xh2q_p6hQEo&NR=1
http://www.nvidia.com/object/cuda_get.html

:)

praa
03-29-2009, 05:13 PM
cuda is proprietary

OpenCL is the way to go for that ...

http://www.newtek.com/forums/showthread.php?t=91652

prometheus
04-02-2009, 05:25 PM
since my motherboard or nvidia quadro fx 1100 is giving up on me.(freezing several times a day)

Im gonna check the newer quadro fx series up, or maybe it would be suficient with nvidia geforce gtx 8800
as shown with these from keenancrane ..who developed the solver for nvidia appearantly...now how do we get this speed stuff in core or houdini? :)

http://www.cs.caltech.edu/~keenan/project_fluid.html

http://www.youtube.com/watch?v=ZgoDypGMV50&feature=channel

http://www.youtube.com/watch?v=DER5xp29z6w&feature=channel

http://www.youtube.com/watch?v=WDFNLJeoh6s&feature=channel

http://www.youtube.com/watch?v=lqPa389vi4s&feature=channel

Michael

praa
04-06-2009, 01:08 PM
give it up cuda, openCL will win since it is open

http://www.theinquirer.net/inquirer/news/673/1051673/ati-physics-argument

Red_Oddity
04-06-2009, 02:02 PM
Sjeesh, give it a rest buddy, each to his own.

And if the same group that does the OpenGL standards approval does the one for OpenCL, it could take a very very long time inbetween updates (i mean, just take a look at OpenGL 3.0, that's been delayed for what now, more than a year?).

And don't get me wrong, i'm all for open standards, but if it takes years for the open system to get updates to the latest tech, i'm all for the closed standard as long as it utilises my hardware to the fullest.

Lightwolf
04-06-2009, 02:06 PM
And don't get me wrong, i'm all for open standards, but if it takes years for the open system to get updates to the latest tech, i'm all for the closed standard as long as it utilises my hardware to the fullest.
Ho, again, does CUDA utilise ATI boards? ;) I doubt that anybody will bother porting something "big" to a technology that only works on a fraction of the installed hardware.
And OpenCL does have the major advantage of not being GPU specific as well.

Cheers,
Mike

RollerJesus
04-06-2009, 02:17 PM
Ho, again, does CUDA utilise ATI boards? ;) I doubt that anybody will bother porting something "big" to a technology that only works on a fraction of the installed hardware.
And OpenCL does have the major advantage of not being GPU specific as well.

Cheers,
Mike

Nvidia released PhysX to the masses, so assume for a second that they could open up CUDA.

Which tech is superior then? Anybody know?

Lightwolf
04-06-2009, 02:22 PM
Nvidia released PhysX to the masses, so assume for a second that they could open up CUDA.
They could but they don't...

Which tech is superior then? Anybody know?
There seems to be little difference. OpenCL is quite similar to CUDA but (apparently) has easier memory access. On nVidia boards OpenCL will be based on CUDA anyhow.

CUDA does have the advantage of being cross platform both in OS and (C/G)PU tech, and that's certainly a massive plus.

The downside is that there aren't any public drivers or SDKs... yet. But this should change during the course of this year.

Cheers,
Mike

Red_Oddity
04-06-2009, 02:33 PM
Doesn't Cuda scale down to the CPU since version 2.1?
So when no nvidia board is present, it doesn't mean your app won't work, it will just use the CPU.

Also, and i say this again, OpenCL + Kronos group = slow development.

Lightwolf
04-06-2009, 02:58 PM
Doesn't Cuda scale down to the CPU since version 2.1?
So when no nvidia board is present, it doesn't mean your app won't work, it will just use the CPU.
I don't know... i haven't seen a standalone runtime yet or an app that runs without an nVidia GPU.

Also, and i say this again, OpenCL + Kronos group = slow development.
That's not really an issue. You always have that problem with multi-vendor standards.
One the other hand... for how long has DX 10 been out yet? And how many games fully support it (beyond just patching up their DX 9 pipeline)? Yup, none.

Heck, even OpenGL 2.0+ only software is rare (mainly due to the fact that Intel is the #1 gfx chip vendor...).

Cheers,
Mike

geothefaust
04-10-2009, 02:38 PM
I like CUDA, but OpenCL is at least usable for more hardware setups. While I currently do have a Nvidia card, I'd rather not be tied to a single vendor for my needs.

cagey5
04-10-2009, 03:27 PM
Well I like Cuda, I also like OpenCL, but which is better.... there's only one way to find out ..... FIGHT

Note: This will not mean anything at all to anyone outside of the UK :)

phorne_tca
04-19-2009, 10:12 PM
So, wait, let me get this straight. You guys are saying that if you had the ability to do smoke and interactions with real-time results within Lightwave that you would NOT use the plugin simply because you'd have to have an nVidia card??

I'm just curious because I'm currently working on doing this for both 9.6 and a Python version for CORE, but through using CUDA, not OpenCL. Maybe I should reconsider what I'm doing... I'm all ears as to you guys' thoughts.

erikals
04-19-2009, 10:44 PM
i'd buy a rocket scientist card to be able to do that.
well, maybe not, but you get my drift... ;)

phorne_tca
04-20-2009, 12:38 AM
i'd buy a rocket scientist card to be able to do that.
well, maybe not, but you get my drift... ;)

LOL Well, that's /one/ vote :)

I've got the fluid simulator from CUDA to LW working (for smoke and fire effects at least, the water part is going to take me a bit more time, though the equations are nearly the same). The major downside is the rendering of the voxels in Lightwave. Re-writing a voxel shader is a bit out of my reach at the moment. But I'm working on it. It'll likely be a while before it's an entire package. If anybody has any suggestions, I'm definitely all ears...

phorne_tca
04-20-2009, 12:48 AM
Doesn't Cuda scale down to the CPU since version 2.1?
So when no nvidia board is present, it doesn't mean your app won't work, it will just use the CPU.

Unfortunately, this isn't the case. The memory allocation functions are Cuda specific, although a test function could be written to decide whether a cuda-based board is available for use or not, and if not then the program would be kicked out to the CPU instead.

Also, programs are written /for/ CUDA or /for/ the CPU. If it's not written in the code, it's not going to automatically kick it out to the CPU...again, unfortunately.

prometheus
04-20-2009, 10:39 AM
Well Im in the start blocks to get me some of the quadro fx series with cuda.

I havenīt join in on the core program yet and I will probably not do so for a while since nothing is revealed in terms of a roadmap regarding voxels and fluids for core.
If a plugin shows up for lw 9.6 thou. I would definitly be very interested in getting it thou.

edit..oh and dynamite is soon past on to oblivion.

Otherwise for the moment being and regarding current core development, I will probably getting more into houdini, those new pyro fx tools are sweet along with the rest of the updates they have done to houdini 10.

Michael