on
Recently, George Gueorguiev, a Voodoo technician from Calgary wrote an interesting post on our message boards about the difference in performance between dedicated AGEIA PhysX hardware and PhysX-enabled NVIDIA hardware. Since not everyone who reads the blog reads the boards, I thought I'd re-run his post here. What follows isn't conclusive (George only tested on one config and with synthetic benchmarks), but it's a fun, informative read none the less.
-----------------------------------------
Now that NVIDIA has officially/publicly launched PhysX on GeForce (through their latest driver package: 177.83), many people are probably wondering what this means for performance in gaming, since many of us are using recent NVIDIA cards that now support PhysX technology. And, what's the difference between the standalone AGEIA PhysX PCI-Express card, and the now-integrated GeForce PhysX?
Included in NVIDIA's latest driver package, ForceWare 177.83, is the latest version of the PhysX Software, version 8.08.01. That's why the driver package size has jumped to over 100MBs. This driver package supports GeForce 8-series and higher, allowing these cards to do PhysX acceleration. For those who have both an 8-series or higher card and an AGEIA PhysX card, the software allows you to select which device to use for PhysX: the graphics card or the standalone AGEIA card. Does it make a difference?
I ran a few quick synthetic benchmarks, which yielded some interesting results. I ran all these benchmarks on a test system with an Intel Core 2 QX9770 CPU, an AGEIA PhysX PCI-Express card, and either dual EVGA GeForce GTX 280 or a single EVGA 8600 GTS.
I ran 3DMark Vantage's CPU tests (which primarily test physics) on their "Extreme" setting, using various PhysX settings. First I ran the benchmarks with PhysX acceleration running off the 8600 GTS. Then, I disabled hardware PhysX acceleration and ran it in software mode (work is done by the CPU, QX9770). Afterwards I ran the benchmarks with PhysX acceleration being done on an AGEIA PhysX PCI-Express card. Finally, I let 'er rip on dual GeForce GTX 280s doing the PhysX work. At this time I have not yet run any real-world gaming benchmarks, I've only run synthetic benchmarks.
As you can see, the dual GTX 280s run PhysX (in this test) about 263% faster than a standalone PhysX card... impressive.
Below I ran another synthetic benchmark, oZone3D's PhysX FluidMark. The results below pretty much echo the ones above. I ran all of these tests at a display resolution of 2600x1600.
The results above show that dual GTX 280s run PhysX (in this test) about 405% faster than a standalone PhysX card...
In conclusion, it seems that a couple of video card generations ago, a standalone PhysX card did in fact increase performance. This is shown by the fact that it's able to beat out even a QX9770 in PhysX acceleration. Nowadays however, with current generation NVIDIA cards, a standalone PhysX accelerator is no longer necessary. The latest generation of NVIDIA cards, the 200 series, are more than capable of handling both graphics processing and physics. I'm sure we'll slowly see this type of shift towards video cards handling almost everything gaming requires, and CPUs becoming less and less critical towards game performance.
