How PhysX runs on Nvidia and AMD (ATI) systems

[Goku]

I have been reading through these forums for hours upon hours, since PlanetSide 2 was formerly and I have been noticing many misconceptions regarding PhysX. Most of the misconceptions is in regards to how PhysX runs on Nvidia and AMD (ATI) systems. The purpose of this thread is to clear up these misconceptions entirely. I know there is threads floating around with information pertaining to PhysX use, but all of that info is scattered and has useless bickering.

PhysX software was originally introduced by Ageia and ran on a PPU (no longer exists) or CPU. Nvidia bought out Ageia in order to make use of PhysX themselves. With the Nvidia buyout you had more developers taking advantage of PhysX and this allowed Nvidia to put the software on their GPUs. A comparable Physics engine is Havoc that is only used on the CPU and in known for being used in the Source engine (HL2, L4D, and TF2).

There is two basic kinds of PhysX. First there is one for Physics that just uses CPU and has no attachment to using the GPU this being the more commonly used one. Second is accelerated GPU PhysX meant for specific types of Physics and this is the less used version being only in under 30 games.

No idea on what kind of PhysX PS2 is going to be using.

How accelerated GPU PhysX runs on hardware is different when running systems with either Nvidia or AMD GPUs. With Nvidia you can run PhysX on the GPU or CPU. On AMD you can only run PhysX on the CPU. There is no way to run PhysX on a AMD GPU.

Unfortunately there is a performance gap between running accelerated GPU PhysX on a GPU vs CPU. Running PhysX on a GPU allows for significantly higher performance then a CPU. If you have an AMD GPU you are going to be suffering a performance loss as result. Though there is a work around for this issue if you have an AMD card. You can get a hacked PhysX driver allowing for a Nvidia card to be used as dedicated card for handling the physics calculations. You can do this normally with a Nvidia card as well to get even higher performance then just making one GPU render and do physics calculations.

Here is a comparison from a review site:




This review used a 980X as the CPU. Even though the 5870 is a stronger card then this weaker version of the GTX 460 by running PhysX on the GPU it gets 2.5 times higher performance with PhysX on high. The 980X is the most powerful desktop processor out now and even that cannot out do PhysX being used on a GPU.



This here shows the GTX 275 being use as a dedicated card for PhysX calculations. You can see the the 5870 with the GTX 275 pairing allowed for a 2.5 times increase in performance with performance being nearly the same as the GTX 460.

What can we gather from all this? Like I mentioned before PhysX at the moment is nearly unable on the CPU. A Nvidia card must be used in someway in order to get proper performance if wanting to use PhysX. Be it using Nvidia as your primary card or having an AMD one plus having a Nvidia card for dedicated PhysX processing.

This is what the current accelerated GPU PhysX looks like. Nvidia is just recently released PhysX 3.0 to developers last month. I do not know if PS2 is being developed with this version. The new version promises greater performance on the CPU, but there is no games using this yet. With no games there is no to know the true performance yet in game with this new version when it comes to GPU vs CPU. There is NO mention of accelerated PhysX being used on AMD cards in this version. If it was it would of been mentioned, so for the time being do not think this version will allow it. Accelerated GPU PhysX still is a way for Nvidia to pull in more GPU sales instead of people going to AMD never forget that. Many developers use it as a way to get funds from Nvidia for helping develop the entire game like the TWIMTBP.

EDIT: Corrected the post as per feedback from Atranox's post.