ArmA2 and CUDA

[teach 1]

I guess this is more of a question to BIS or someone really in the know but would or could the ARMA2 engine benefit from incorporating Nvidias CUDA technology into the overall processing power?

CUDA is supposed to allowed the GPU to participate in processing efforts WITH the CPU I think. And since this game is so CPU dependant, maybe it could help.

[<-Falco-> 10]

xD, forgot about that, you need the GPU for their own bussines. CUDA is for aplications, no games.

[Murklor 10]

xD' date=' forgot about that, you need the GPU for their own bussines. CUDA is for aplications, no games.[/quote']

A game isnt an application? :confused:

You can use Cuda for PhysX, which Arma2 would probably benefit from... Imagine walking through a forest with leaves falling down, physically reacting sparks flying away from explosions, etc etc.

[<-Falco-> 10]

You dont want to steal the gpu ...Nvida physics are a shame...xD

[echo1 0]

I'm pretty sure the whole point of CUDA is that you use the Graphics Card's processing power for non-graphics related tasks. But if you are running an application that uses the graphics card for... well, graphics, having CUDA run on it is going to waste processing power that could be used for rendering stuff.

Unless of course you are talking about using "Phys-X" or the like, but I think it would be much more beneficial to use the CPU for that instead. I mean, at the moment, ArmA 2 only uses two discrete threads, and a lot of people out there have Quad cores.

[kRush 10]

AI on the GPU. :pray:

[bushlurker 45]

CUDA is a little unreliable as yet - despite the new version just released...

Kegetys has incorporated it into his triplehead SoftTH (though the x64 version stubbornly refuses to behave so far) where its basically used to hi-speed "blit" chunks of the final rendered screen across the pci-e bus for display via a secondary graphics card...

(main card renders say, 4096x1200 frame - sliced into x3 - main card displays x2 slices via 2x monitor outs and passes 3rd slice via pci-e bus (optionally using CUDA) to a "dumb" secondary card, who's only job is to display it... sync the 3 and you have triplescreen Arma2 - without a Triplehead2go box)

... all very clever - and CUDA can occasionally be persuaded to play its part... however, I'm not sure how this would benefit Arma generally - for one thing, not everyone has an Nvidia card, and therefore CUDA...

Coding the game to only use certain manufacturers graphics cards would be nearly as insane as coding the game to only use certain manufacturers head tracking systems (oh, wait - they already did that!!!)

OK - as insane as coding the game to only use certain manufacturers joysticks then.....

B

[whisper 0]

A game isnt an application? :confused:

You can use Cuda for PhysX, which Arma2 would probably benefit from... Imagine walking through a forest with leaves falling down, physically reacting sparks flying away from explosions, etc etc.

If I remember well, devs stated having tested Physix and found it couldn't cope with the requirements of ArmA, or something like that. It was unusable anyway

[An Fiach 10]

CUDA is a little unreliable as yet - despite the new version just released...

Kegetys has incorporated it into his triplehead SoftTH (though the x64 version stubbornly refuses to behave so far) where its basically used to hi-speed "blit" chunks of the final rendered screen across the pci-e bus for display via a secondary graphics card...

(main card renders say, 4096x1200 frame - sliced into x3 - main card displays x2 slices via 2x monitor outs and passes 3rd slice via pci-e bus (optionally using CUDA) to a "dumb" secondary card, who's only job is to display it... sync the 3 and you have triplescreen Arma2 - without a Triplehead2go box)

... all very clever - and CUDA can occasionally be persuaded to play its part... however, I'm not sure how this would benefit Arma generally - for one thing, not everyone has an Nvidia card, and therefore CUDA...

Coding the game to only use certain manufacturers graphics cards would be nearly as insane as coding the game to only use certain manufacturers head tracking systems (oh, wait - they already did that!!!)

OK - as insane as coding the game to only use certain manufacturers joysticks then.....

B

err, he means include code that can benefit from CUDA not make nvidia only compatible..

but if it doesn't work, well it doesn't work.

[bushlurker 45]

I could be completely wrong and probably am, but I thought that Nvidia CUDA was Nvidia-only ? So only Nvidia owners would benefit from CUDA-related enhancement - ditto Phys-X... or can ATI owners use CUDA and PhysX too???

B

Edited July 8, 2009 by Bushlurker

[echo1 0]

AI on the GPU. :pray:

Or just on the CPU which isn't being used properly at the moment?

I could be completely wrong and probably am, but I thought that Nvidia CUDA was Nvidia-only ? So only Nvidia owners would benefit from CUDA-related enhancement - ditto Phys-X... or can ATI owners use CUDA and PhysX too???

Correct. Im pretty sure the OpenCL libraries are platform agnostic though?

[bushlurker 45]

Im pretty sure the OpenCL libraries are platform agnostic though?

yup! interesting article on that topic here...

Meantime, as you can see from my error logs - CUDA SUCKS... :D

29.141 SoftTH: cudaThread <cudaVmToSys>: Using device GeForce GTX 280

29.219 SoftTH: WARNING! CUDA sucks at line 242, file C:\Util\c++\SoftTH2\cudainterface.cpp, thread 0x0694 <unknown error(30)>

29.219 SoftTH: cudaThread <cudaVmToSys>: Registering surface 0 as ReadOnly

B

Edited July 8, 2009 by Bushlurker