Existing Card Versus Dedicated Physx : Results

[MeatLoaf]

Well i posted this in a different thread but thought id share my findings in a seperate thread as well

My main GFX card is an EVGA 8800GTX ACS3 superclocked jobbie and i recently got a BFG 8600GTS OC to use for physx. Rest of spec in sig, running Vista64. I had thought the GTX + GTS would be the better combo but not judging by the results..

First Run : 8800GTX AC3 doing GFX and BFG 8600GTS OC 256Mb doing physx

CPU Score : 13797
Test 1 : 777.97 Plans/S
Test 2 : 46.69 Steps/S

2nd Run : 8800GTX ACS3 doing GFX and physx ( 8600GTS removed )

CPU Score : 26407
Test 1 : 772.67 Plans/S
Test 2 : 107.36 Steps/S

Now during these tests the GTX isnt really being tested and the higher score could well be the result of the much better capabilities of the GTX.

Is there a benchmark for UT3 that i can use to compare the dedicated physx option as i feel it will be a better indication of what to expect since the GTX will be busy with GFX duties

 

[Blackbadger]

You used the wrong benchmark.

The futuremark tests dont strain the gpu and "ppu" simultaneously, resulting in completely non realistic results..

3dmark's time has gone, it's a pretty show, but useless to determine real world performance of hardware.

 

[MeatLoaf]

You used the wrong benchmark.

The futuremark tests dont strain the gpu and "ppu" simultaneously, resulting in completely non realistic results..

3dmark's time has gone, it's a pretty show, but useless to determine real world performance of hardware.

Thats what im saying but apart from UT3 i dont have any physx games

 

[drunkenmaster]

Thats what im saying but apart from UT3 i dont have any physx games

theres a reason for that. Other than deliberately sabotaged games, its useless, even in those games, its useless but gives a pretty number.

 

[sarge78]

theres a reason for that. Other than deliberately sabotaged games, its useless, even in those games, its useless but gives a pretty number.

So I guess Nvidia, Ati/AMD, Intel, Apple etc are all wrong about performing number crunching on GPU style architecture

I wouldnt expect games which use physx and stress the main gfx to come out for a few years yet.
________
FORD EXPLORER SPECIFICATIONS

 

[Metalface Mark]

cant you just run a timedemo using UT3 PPU levels?

 

[Blackbadger]

Imo gfx cards are not yet powerfull enough todo real time physics.

Sure little fancy bits here and there, but no good fluid dynamics/fire/smoke.

I guess they had to start the ball rolling with the current architecture, which is a good thing.

However until proper physical shaders can be applied to objects, so say a concrete block, has a concrete shader applied, it will break/chip/explode/fall/impact in a realistic way. Properties such as weight, buoyancy, friction, fracture points, stretch, squash,flex, flammability, burn rate, what colour smoke it gives off, how much, etc ... many others I'm no engineer. There are so many variables for a simple block to make it physically accurate, I can imagine it being a gargantuan task to create the initial library.

Of course there are many types of concrete, so multiple shaders will need to exist for different types of object. These shaders will need to be hand made, using architectural/engineering data sets which do exisit to some extent as engineers need to test/simulate construction materials and how they react to cars/planes being driven into them, explosions etc. however the simulations are so complex it currently takes a crazy amount of time to calculate. Gpu/Ppu's would certainly help. But they are not powerful enough to do real-time environmental calculations.

To conclude, Physx is a neat idea, however due to today's technology its not powerful enough to be useful, at least to enable immersive physically realistic environments, which I guess is what we are all after.