Cuda support for algodoo?

[Tank2333]

i just recently read about CUDA (a feature where the graphic cards GPU can be used to calculate other stuff than grafic)
so i read (and saw youtube videos) that a GPU can render more and much faster things like an normal cpu
i saw an youtube video where a landscape has been calculated with cpu it was 1min 35s than he turned on CUDA in the options and it was calculated in 2 seconds

i know that algodoo maybe had to be complete reprogrammed
but maybe it could be an option in futre because the performance boost with an avarage graphic card would be astronomicle... 40000 water particles at 30fps would be possible

so what u think about?

[Physicsguy]

Sorry for bump.

Looks like a good idea actually! I might look into these GPUs.

[Rhett97]

Wow when i get my pc i was gonna get a CUDA gpu so now i have more decsion to back up my decision and to help convince my mother

[gradyfitz]

I'm pretty sure I read somewhere it was being implemented, and something along the lines of 200,000 particles at an interactive framerate[source].

Using the GPU to calculate things that CPUs normally did is quite a powerful technology , just take a look at this: Client statistics by OS on Folding@Home, about 125.25x the power. The biggest downside of using the GPU to calculate CPU calculations is that GPUs don't have the same built in commands (or so I've read).

[KarateBrot]

i just recently read about CUDA

oh, so you're not a gamer, right?

[admin]

We have already implemented some of the algorithms in CUDA, but most of what we do now is done in OpenCL. OpenCL is an open standard for parallel programming both for multicore processors and for graphics processors. It was originally proposed by Apple to the Khronos consortium and is now supported by all the Khronos members, which also include Intel, Nvidia, AMD, and many many more. Umeå University (where Phun/Algodoo was originally developed, and where all the simulation methods come from, and where several of us at Algoryx work part time) is also a Khronos member, so we are rather active in this area.

We have 120.000 incompressible fluid particles running at 20Hz in 3D (on an Nvidia 280 GTX), and should be able to do 200-300k particles in 2D if not more. We'll publish some videos of this in a few weeks.

We will also be publishing several scientific papers in the coming months where Algodoo is used

[thebeech]

you are not using OpenCL yet though are you ? because it's a waste of my quad-core but yer OpenGl would be nice ive seen it in use before and it vastly improves performance

[Tank2333]

We have already implemented some of the algorithms in CUDA, but most of what we do now is done in OpenCL. OpenCL is an open standard for parallel programming both for multicore processors and for graphics processors. It was originally proposed by Apple to the Khronos consortium and is now supported by all the Khronos members, which also include Intel, Nvidia, AMD, and many many more. Umeå University (where Phun/Algodoo was originally developed, and where all the simulation methods come from, and where several of us at Algoryx work part time) is also a Khronos member, so we are rather active in this area.

We have 120.000 incompressible fluid particles running at 20Hz in 3D (on an Nvidia 280 GTX), and should be able to do 200-300k particles in 2D if not more. We'll publish some videos of this in a few weeks.

We will also be publishing several scientific papers in the coming months where Algodoo is used

wow that is impressive
but does that work on normal objects too? because my 2.4ghz core 2 duo can handle about 2000 particles at moderate fps and can handle 500 boxes of the same size when they collide
so when openCL is used does that mean a gtx280 can handle like 5000 bodys? or more?

but if the water just would be GPU calculated that it could be actual used to do something with water^^

ps:@ karatebrot ^^ im not a hardcore one and i did know of it earlier but just recently informed my self about it and what it is ^^

[thebeech]

what about if people dont have graphics cards though and just use onboard graphics like in laptops ? becuse that isnt over powerful. like my laptop has something like 128mb of graphics memory and the graphics output isnt very good, although my desktop has got a decent one.

[Tank2333]

i think when it is going to be included i think it will be an option
so that u can turn it off or on
and therefore i dont think that an onboard graphiccard will be support openCL or CUDA

[immibis]

What about using multiple cores (or CPUs, although not many people have multiple CPUs)? More people have multiple cores than have CUDA graphics cards...

[KarateBrot]

What about using multiple cores (or CPUs, although not many people have multiple CPUs)? More people have multiple cores than have CUDA graphics cards...

Read the admin post (i made a quote of it)
and every pc that's using a NVidia 8 or higher Series has got CUDA support. even if your PC is more than 3 years old it could have CUDA. you just need the newest drivers. since NVidia is owner of Ageia they implement GPU physics support onto their graphics chips.

We have already implemented some of the algorithms in CUDA, but most of what we do now is done in OpenCL. OpenCL is an open standard for parallel programming both for multicore processors and for graphics processors.

[Tank2333]

oh i dont know if u know this but take a look at the config.cfg there is an entry "threading =1" if u turn this to the cores u have than i thin it sppeds up phun about 20% sometimes
i think its still beta so they dont put it on options

i didnt test it too much buzt algodoo uses liek 100% core1 and like 10%-20% from the other core but not constantly

[Mystery]

Gpu calculations are awesome and will certainly be the future for computer but at the moment it hasn't been used on a large scale product, the whole idea is still fairly experimental. Before somting like this is implemented support for multiple Cpu is needed. Most computers are dual core if not more and this could greatly improve the speed of algodoo.

[KarateBrot]

it has been used on large scale products right now for simulations and games like unreal tournament 3 and many others.
for physics simulations the gpu can give WAY more performance than the cpu that's a big advantage. with a good gpu double and triple speed compared to a cpu.