• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

CPU as Physics Processor

  • Thread starter Deleted member 169154
  • Start date
D

Deleted member 169154

Guest
I know the PhysX card was not too popular because of its cost and taking up a PCI slot, but I have heard devs saying that extra cores in a CPU can be utilized for physics.

This is purely curiosity, but can a single core of a modern CPU replace the entire PhysX card? I know GPUs are good at FLOPS and CPUs are good at whole numbers / MIPS. I am just not sure where physics calculations fit in this.

I would love the have that fourth core be utilized purely for physics.
 
No, it would be a LOT slower.....

but noone wants to shell out extr money for a physx card that only 3 POS games support so we wont see physx card anytime soon, no to mention nvidia stated they are killing them in support of gpu physx using the physx sdk wrapped around the CUDA engine.....
 
Unfortunately a standard CPU, even a multicore one, isn't as good as a GPU (or PPU) at highly parallel number crunching. (Many instructions performed simultaneously).

It's the same reason we moved nearly all the rendering tasks from the CPU to an accelerator card. Physics benefits from massively parallel computations the same as graphics does. Its why a 500MHz GPU can render dozens if not hundreds of times faster than a 4GHz CPU.
 
CPUs are really pretty slow and inefficient things, it's just that they're very versatile. Any chip that's custom-built to do a particular task will almost certainly beat the CPU at it.
 
Well, I just saw the article that after a software upgrade all nvidia 8 series will have PhysX support.

Thanks for answering my question. I figured physics also benefited from processors dedicated to physics, but wasn't sure if the a dedicated core could perform the same.
 
NVidias solution looks to be splitting the load between CPU and GPU.
 
I didn't read specifics. I just saw that Nvidia is acquiring Ageia. So they have the ability to turn an 8 series GPU into a "PhysX" card through software apparently.
 
NVidias cards are gaining a driver to port PhysX commands to CUDA http://en.wikipedia.org/wiki/CUDA.
All drivers run in software and give access to hardware therefore allowing some processing to be performed on the CPU and the rest on a GPU or 2 or 3 :)

The driver itself (not the processing) can effectively run for free at the moment as there is usually spare CPU available.
There is likely to be some hit in graphics performance due to stream processors being used on the GPU but I imagine the amount of Physics you can have will be scalable and user configurable to some degree.
It has been pointed out that part of the GPU that hasnt been used yet is set aside for Physics type operations. Also there is idle time on the stream processors which can be utilised.
So some Physics processing will be free, it remains to be seen how much though and what impact any extra Physics has on graphics performance.
 
You must log in or register to reply here.
Back
Top