• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

best card for physx

youdsy

youdsy

youdsy

Soldato
Joined
14 Feb 2011
Posts
2,740
looking at investing into a card that will be dedicated for physx.

looking for something that runs off the PCI express, like the 640 for example.

just wondering what is the best card??

i heard that the 600 series is weaker than the 500 series at physx and the 500 series is weaker than the 400 series. this true?
 
I dont really see a point, you wont gain a huge amount if anything. You will use more power and produce more heat... that being said an old 460 would probably be a solid choice, heck a 450 might even do it.
 
650Ti is pretty much the one to go for if thinking dedicated. Have a google and you'll find a guy that tested the 640, 650 and 650Ti as well as a GTX680.
 
PhysX acceleration on an NVIDIA card is done using CUDA.

So the simple answer is, whatever card is the fastest at processing CUDA will be the fastest at accelerating PhysX.

From personal experience (I use CUDA for numerical simulation daily) the current Kepler cards (6xx series) are far superior to the old Fermi cards as they have been designed from the ground up for CUDA.

So the answer is, whatever the fastest Kepler based GPU you can afford is.

That being said, I'm not convinced you'll see a huge difference, only portions of the PhysX library can be accelerated (and not the important parts to be frank!) so when you enable PhysX in a game, quite often a large portion of what is being calculated is still being done on the CPU, with only very specific portions being done on the GPU.
 
I suspect it's actually a case of occupancy.

I do most of my daily grind on either a GTX 660 or a GTX 680.

Smaller cases that have a lower GPU occupancy in terms of workload often perform faster on the 660 than the 680. It's a simple parallelism problem. The 680 has nearly 600 more "cores" than the 660, so if the computational problem fits better on the 660 than the 680 then the overhead of wasting the 680s 600 odd extra threads means the CUDA application performs worse on the better hardware.

Having spent quite a few years of my life studying PhysX (amongst other physics engines), I have no doubt that in many cases problems presented to the GPU by the library are in no way optimal for the 680.

I could be wrong, but my guess would be that's the case.
 
You must log in or register to reply here.
Back
Top Bottom