The test
As mentioned above there were no problems during installation, nor have there been any in actual use. The fan on the card is always running at full speed, which is slightly annoying, as it is clearly audible. It is, however, far from the noise levels of a Geforce 7800GT at full speed or Geforce 6600GT's - a better comparison would be the stock cooler of a Radeon X800XL, so it's all in all quite reasonable. In idle condition the card is warm, and the temperature increases slightly under load. It's in no way so hot that you risk burning your fingers, but BFG might want consider a larger cooler, both due to the current noise level and to keep the temperature down, as it might become a problem in closed cases with insufficient air flow.
Test setup: CPU: AMD Athlon 64 FX-53, AMD CPU v1.2.2.2 driver. Main board: MSI K8N SLI, 1GHz HyperTransport, 200MHz clock signal, NVIDA nForce 4 v6.70 driver. RAM: 2x512MB Mushkin PC3200 Special DDR RAM @ 200 MHz, CL2.0-2-2-5-1T. Primary graphics card: BFG Geforce 7600GT, PCI-E, 256 MB, NVIDIA ForceWare v84.21 driver. Secondary graphics card: Inno3D Geforce 7800GT, PCI-E, 256 MB, NVIDIA ForceWare v84.21 driver. HDD: Western Digital RE, SATA300, 250 GB, 7200 RPM, 16 MB Cache. OS: Windows XP Professional, SP2. I have chosen to keep the usual reference graphics installed, in order to be able to switch between them and discover if they could be a bottleneck.
The main board has been setup for SLI, and each graphics card has been running at a speed of x8 PCI-E lanes instead of x16.
Test programs: 3DMark2005 build 120. 3DMark2006 build 102. Far Cry - Rebellion timedemo. Half-Life 2 - Canals timedemo. Ghost Recon - Advanced Warfighter. CellFactor. Bet on Soldier: Blood Sport. Far Cry and Half Life 2 have been run at two levels of detail, High and Low. High at a resolution of 1280x1024, 4x AA, 8x Anisotropic filtering, whereas Low was at 640x480 with AA and Anisotropic filtering disabled.
Ghost Recon, CellFactor and Bet on Soldier are all on the AGEIAT list of PhysX supporting games and have been run at 1280x1024 but with Anti-Aliasing and Anisotropic filtering disabled.
3DMark 2005 & 3DMark 2006: 
Neither 3DMark 2005 nor the newer 2006 seems to be able to make use of PhysX.
Half-Life 2 & Far Cry: 
Both Far Cry og Half-Life 2, which as you know have quite a lot of physics, are somewhat boring when used with the PhysX card. There are no new effects visible, and the performance does not change no matter what level of detail you chose.
What does that tell us? It emphasizes one important aspect of the PhysX technology, namely that
PhysX cannot give any advantages in games that are not made for PhysX, neither graphically nor with regards to rendering. PhysX is
not backwards compatible and will be never be, as a separate API has been developed for PhysX which games are just not programmed for.
You can hope for patches to older games, but this is not likely, as it is a huge task and requires a rewrite of large parts of the game code. In the Hexus' review, which is written by a spare time 3D programmer they write the same thing: PhysX support must be included from the beginning. Anyything else will only result in frustrated programmers.
How does it look in games which actually support PhysX?
Bet on Soldier: Blood Sport: 
Bet on Soldier: Blood Sport was the first game I fired up with the PhysX card installed. After having looked in vain for something PhysX-related in the settings I launched the game in campaign mode. At first I didn't notice anything out of the ordinary, but I kept playing to get to know the game before I tried without the PhysX card.
After having removed the card I started the game again and continued my campaign, but with no visual differences whatsoever. Pretty strange - perhaps a patch was needed? This didn't help either, and the effects seen on various videos on the Internet never appeared.
Bet on Soldier: Blood Sport leaves bland taste, which is of course quite understandable when I never managed to activate any kind of PhysX stuff in the game.
My guess is that another game version has been used to make those videos
Ghost Recon: Advanced Warfighter: 
After the Bet on Soldier disappointment I continued with Ghost Recon to see if this game could show off something. After installing the game I once again looked for PhysX settings, but again in vain. Very well, I went on and started a campaign. After spending 2 minutes going down in a parachute I could finally start checking. A few random potshots hitting roads and walls revealed that the number of flying debris was larger than you'd normally expect. Shooting some car tires punctured them, but was that really all PhysX could deliver? At first this didn't seem very impressive - until you fired your first grenade.when you sit there in front of the monitor you sort of just sit stunned for a bit, wondering what just happened. It is all very impressive, but how does it feel when the first rush of excitement fades away?
Personally I don't think the card lives up to the hype there has been - at least not in Ghost Recon. Yes, there is a very noticeable improvement compared to before, but spending around 2000 DKK for a PhysX card just to play Ghost Recon with a few extra physics effects is a bit too much in my book.
This is of course just my personal opinion, so judge for yourself:
Ghost Recon: Advanced Warfighter with PhysX:
http://www.hwt.dk/images/litt2/download/Physx/GRAW%20PhysX.avi (right click and chose save as - should be faster than streaming.)
Ghost Recon: Advanced Warfighter without PhysX:
http://www.hwt.dk/images/litt2/download/Physx/GRAW%20No%20PhysX.avi (right click and chose save as - should be faster than streaming.)
CellFactor: 
I got hold of CellFactor near the end of the test, and at this time my expectations were far from satisfied. A possible conclusion could have been that we would probably have to wait and see what kind of effects future games such as Unreal Tournament 2007 could show off, but CellFactor quickly changed this.
CellFactor is one of those games that requires a PhysX card, and the AGEIA review guide specifically explains how to play the game using a PhysX card - and how to do it without a PhysX card - I quote: "It's simply not possible". Well, ok, that should clear up any misunderstandings ;).
Playing the game using a PhysX card is really an extraordinary experience. From the moment you fire your first grenade and until the end credits you just sit there gawking.
If Ghost Recon sparked a "wow", CellFactor sparks a "wow" x10.
It is simply beyond all expectations, with the game play itself built upon the new physics possibilities. I will go so far as to claim that this is only the top of the iceberg.
It does not cause the same tingling in your fingers as when you first fired up Quake 2 using 3dfx Glide using a Voodoo 2, but it is very close. The excitement do of course level off over time in CellFactor, but in no way as much as in Ghost Recon.
I was in doubt before trying CellFactor, but afterwards I am convinced. The enormous scale in CellFactor makes it impossible not to get carried away.
If more games are released supporting PhysX as well as CellFactor does, those 2000 DKK are clearly worth spending. Again, it's a matter of taste, but I will insist that you should try PhysX in something like CellFactor before deciding.
CellFactor with PhysX:
http://www.hwt.dk/images/litt2/download/Physx/CellFactor.avi (right click and chose save as - should be faster than streaming.)
CellFactor without PhysX:
"It's simply not possible" ;).
Regarding the performance of the PhysX card I must refer to the reviews from Hexus.net and Anandtech, as I was not able to run a time demo in Ghost Recon, and hence I do not have a basis for judging it.
Both Anandtech and Hexus.net concludes that you actually get lower FPS in Ghost Recon when using PhysX. This may be caused by an increased load on the CPU, the code or perhaps the PhysX card itself. It should be noted that Anandtech does not think that this have a negative impact on the gaming experience, but they still leave it to the reader to decide if it worth it.
In a way it seems stupid that using an extra card to take care of the physics calculations actually lowers the performance, keeping in mind that there are more and more impressive effects. The goal is clearly to get better performance by taking some of the load away from the CPU, or at least keep the same level with the extra effects enabled.
We will have to see if this will be the case as time goes on.
Links to the reviews from Hexus.net and Anandtech:
Hexus.net: http://www.hexus.net/content/item.php?item=5492
Anandtech: http://www.anandtech.com/video/showdoc.aspx?i=2751
