A quick PhysX comparison with Mirror's Edge

So, it seems to me that PhysX started out as a bit of useless technology, not because the tech itself wasn't any good...but because of the lack of support. I read tons of reviews back then concerning the BGX PhysX add-in board, and recall being very underwhelmed.

Things have changes significantly since then with the buyout from the green team, and PhysX support seems to be popping up in more and more games these days. A quick check on vNidia's website shows over 80 published titles with PhysX hardware support What's more, all nVidia GPUs that support CUDA, also support PhysX HW acceleration (8800GTX and newer).

What seems to be missing these days are decent reviews (not that I'm claiming this one is decent) showing a more up to date reflection of PhysX implementation. Maybe this will compel you to dig out last years video card, and throw it back into your rig...enjoy

So I recently upgraded my 8800GTX to a 9800GX2 and have been playing away without a thought towards PhysX...that is until I came upon Mirror's Edge, a recent game from DICE. Interesting game, and it proudly wears the PhysX badge. So I decided to throw my 8800GTX back into my machine to see if there was a tangible difference...much to my surprise, there was

My goal here was to compare the performance difference between SW PhysX (CPU handles all PhysX calculations), HW PhysX on primary graphics card, and HW PhysX on dedicated PhysX card.

My system:
Intel Core 2 Extreme QX9650 @ 4.0GHz
Gigabyte GA-X48T-DQ6
4x1GB DDR3 @ 1600 7-7-7-21 1T
eVGA 9800GX2 SSC @ 675/1675/1050
eVGA 8800GTX @ 575/1350/900 (dedicated PhysX processor)
300GB WD Velociraptor HHD
FSP 1050 Everest 1000W PSU
Sharp 32" 1080p monitor @ 1920x1080
Vista Home Premium x64 SP1
nVidia Forceware 182.08 drivers

First I'd like to point out that turning PhysX on or off in the game has a major impact in performance. When switched off, many of the visual treats in the game are completely missing. So the frame-rates stay locked near the 60FPS limit. For instance when watching the video, as I'm running towards the helicopter at the corner of the building, it shoots through the tarp...with PhysX turned off...the tarp is completely missing

30 second segment used for the benchmark sequence

The results are the average of three consecutive runs (no pun intended ).



Switching PhysX on in-game adds all the extra visual goodies into the mix, and can really tax the frame-rate if you only have a CPU to handle the load (1st column). However, when using my 8800GTX to offload all the PhysX calculations (2nd column), the frame-rate is almost as good as with PhysX shut off completely, staying above 50FPS for the entire run. The most interesting portion of my testing was with the 9800GX2 handling both the graphics processing and the PhysX calculations, it seems that adding the PhysX to the mix really kills the performance, resulting in even lower frame-rates than SW PhysX processing.



Increasing the AA to x16Q really puts a hurt on the system here. As you can see, we're becoming very GPU limited and the SW PhysX are almost as fast as the dedicated card. The extra load on the 9800GX2 seems to be too much to bare, and the performance plummets.

Obviously, this is not a comprehensive review, and I'd still like to investigate much more. For instance, what about those of you not running a 4GHz quad...maybe it would be even more beneficial to you. Many more questions...not enough time.

Maybe this will spur more people to find out on their own what is needed, and what helps most in all those other situations. Also, Mirror's Edge is only one of over 80 titles on the market, it's be interesting to know to what degree the other titles benefit from a dedicated PhysX processor.

Anyhow, for now...it get's my thumbs up, and I'll continue using my rig configured this way.

Thanks!

PhysX is so USELESS

Apparently he has no idea what PhysX is and can provide.
Either he runs ATI cards and has never seen it or is still in the world of MS. Pacman