Nvidia PhysX dirvers for CUDA enabled GPU's

AMD_Gamer

Fully [H]
Joined
Jan 20, 2002
Messages
18,287
I was just reading this http://www.hardwaresecrets.com/blog/97 and it got me really excited, but i was wondering if you run this on your GPU would it be just like having the PPU card in your system, like could i play cell factor , or the UT3 Physics mod or any of the tech demos and games that support the Physics card would it treat it just like a normal PPU card? or is it going to have serious performance hits? I remember reading about this back when Nvidia announced that they were working on this but i have not read much about how exactly this is going to work.
 
Yes, you can use a 8xxx + series Nvidia card to use as a Physx card. But you will need 2 cards because you can't do physics and graphics on one card (as far as I know but I wish you could). But it's not live yet.
 
Yes, you can use a 8xxx + series Nvidia card to use as a Physx card. But you will need 2 cards because you can't do physics and graphics on one card (as far as I know but I wish you could). But it's not live yet.

well being that the GPU is a heavily parallel array of processors, you could do the physics simulations in some cores while doing the graphics in the rest.
 
Why in the heck did they do that? The whole design of GPU's is inherently multi-threaded! I'm assuming it's so that have to buy and low range card for physics...
 
Yes, you can use a 8xxx + series Nvidia card to use as a Physx card. But you will need 2 cards because you can't do physics and graphics on one card (as far as I know but I wish you could). But it's not live yet.

I'm fairly positive that you can do both CUDA and rendering at the same time, so you should be able to do PhysX + rendering on the same card. Of course, since PhysX will be using up some of the GPU's power, your FPS might suffer a bit...
 
Why in the heck did they do that?? The whole design of GPU's is inherently multi-threaded! So we have to buy and low range card for physics?

no you don't, as GPUs get bigger (more cores) and faster, u would be able to do both physics and graphics on the same GPU at the same time
 
no you don't, as GPUs get bigger (more cores) and faster, u would be able to do both physics and graphics on the same GPU at the same time

That wasn't what I meant - I'm suggesting that you should be able to do both as a GPU is inherenty multi-threaded and that if they're not allowing the user to do so it would be blatant marketing ploy to get us to slap extra dollars down on the table for a second card (and even an Nvidia mobo if they're feeling particularly evil).
 
You don't need a second card, I know with the GTX 280, I'm sure it's the same on other cards as well where it just runs on the single card in the system. It works like having a PhysX card would, possibly a minor FPS hit but that's about it. The 8-series drivers are coming within the next few weeks, this is NOT for the 9-series/GTX 2xx series only.
 
You don't need a second card, I know with the GTX 280, I'm sure it's the same on other cards as well where it just runs on the single card in the system.

yes, the vid card only has to be CUDA compliant, so at least a Geforce 8XXX
 
You don't need a second card, I know with the GTX 280, I'm sure it's the same on other cards as well where it just runs on the single card in the system. It works like having a PhysX card would, possibly a minor FPS hit but that's about it. The 8-series drivers are coming within the next few weeks, this is NOT for the 9-series/GTX 2xx series only.

wow this is really awesome stuff, we will have to wait and see how much of a performance hit it will be , i can't wait to play cell factor and the UT3 physics mod
 
You can do the physics and rendering on the same card but not at the same time. So the card will do physics calcs and return the result to the CPU and the CPU will then deliver the frame for rendering.

The card can only be in either CUDA or graphics mode but not both simultaneously.
 
You can do the physics and rendering on the same card but not at the same time. So the card will do physics calcs and return the result to the CPU and the CPU will then deliver the frame for rendering.

The card can only be in either CUDA or graphics mode but not both simultaneously.

HELLO!! CUDA is an nVidia's program for programming on the 8xxx+ cards. NOT A MODE. There is no switching between MODES. Its not CUDA and GPU. CUDA supplies the GPU with its directions for what is to be done. So yes, both can be done at the sametime if programmed to do so in CUDA. Thanks! ;)
 
You can do the physics and rendering on the same card but not at the same time. So the card will do physics calcs and return the result to the CPU and the CPU will then deliver the frame for rendering.

The card can only be in either CUDA or graphics mode but not both simultaneously.

Since running CUDA doesn't make the video output stop, I think you are wrong. I'm near positive that CUDA and regular GPU work can be done simultaneously (as CUDA isn't another mode, just different instructions sent to the GPU similar to that of a shader). However, everyone must remember that the GPUs resources are fixed - the more CUDA work its doing, the less shader work it can be doing... :)
 
So, anybody know if we could pick up a nice cheap 8400GS and let it do the physics while our amazing new graphics cards concentrate on the graphics?
 
IF you can do it simultaneously on the same card it would be great for people with CPU bottlenecks, video card isnt being used fully so it could take some load off of CPU
 
http://forums.nvidia.com/index.php?showtopic=36286 (aka the CUDA FAQ):


Is it possible to run multiple CUDA applications and graphics applications at the same time?

CUDA is a client of the GPU in the same way as the OpenGL and Direct3D drivers are - it shares the GPU via time slicing. It is possible to run multiple graphics and CUDA applications at the same time, although currently CUDA only switches at the boundaries between kernel executions.

The cost of context switching between CUDA and the graphics API is roughly the same as switching graphics contexts. This isn't something you'd want to do more than a few times a frame, but is certainly fast enough to make it practical for use in games.
 
So, anybody know if we could pick up a nice cheap 8400GS and let it do the physics while our amazing new graphics cards concentrate on the graphics?

There isn't any technical reason you couldn't. CUDA doesn't rely on SLI or anything like that, but this is nvidia we are talking about. They seem to love imposing tons of arbitrary restrictions on things. :eek:
 
I was just reading this article http://www.eurogamer.net/article.php?article_id=155719 and i was about to come make a new thread here but looks like it's been taken care of :p

I wonder how this will work with say, my sli 8800GTS cards + a new high end nvidia card on a 3x slot pci-e 16x motherboard. Can i run my 8800GTS sli setup and then have a gtx 280 going solo? I would think this could see a nice performance increase and at no addition cost to those who have the old cards already. Hopefully nvidia can elaborate further on this in the future, it sounds really cool.
 
I don't see the point of learning CUDA to use GPU for physics and other calculations, you can always write shaders for that which isnt hard to learn, better than learning CUDA if you already learned shader programming. It's just my opinion though, I haven't used CUDA yet.
 
I don't see the point of learning CUDA to use GPU for physics and other calculations, you can always write shaders for that which isnt hard to learn, better than learning CUDA if you already learned shader programming. It's just my opinion though, I haven't used CUDA yet.

I promise you more people know how to write C code than shaders. Thats what CUDA is, after all, a C compiler for nvidia cards.
 
CUDA isnt that EASY, I attended a presentation about CUDA and it was too technical to me, kernels and the stuff, all above the head.
 
I was just reading this article http://www.eurogamer.net/article.php?article_id=155719 and i was about to come make a new thread here but looks like it's been taken care of :p

I wonder how this will work with say, my sli 8800GTS cards + a new high end nvidia card on a 3x slot pci-e 16x motherboard. Can i run my 8800GTS sli setup and then have a gtx 280 going solo? I would think this could see a nice performance increase and at no addition cost to those who have the old cards already. Hopefully nvidia can elaborate further on this in the future, it sounds really cool.

great article
 
CUDA isnt that EASY, I attended a presentation about CUDA and it was too technical to me, kernels and the stuff, all above the head.

Meh, I just looked through some of their samples and it looks like normal C code to me. There were some odd calls and naming conventions, but that is just par for course when it comes to C. The actual computational part (where the brunt of the work was done) was just normal C (at least in the few examples I looked at, I haven't messed around with it too much)
 
Back
Top