Correction: Hardware PhysX acceleration was forced off, not on, in the Great Kulu demo. I've updated the article with an explanation.

These numbers say it all. Using part of the GPU to compute fancy physics effects induces a performance hit, although in this case, that hit was small enough not to seriously affect playability.

Am I reading this right? I mean, the chart says an average of 66.2 fps running standard with no PhysX and an average of 40.1 fps with GPU PhysX, and this is "small enough not to seriously affect playability"...? Well, I mean, I sometimes see that 5 fps (or less) differences in other benchmarks are notable enough to publish and graph--so I'm a little puzzled as to why a frame-rate differential of ~35% would be considered so small as to be negligible.

CPU utilization was paradoxically higher in the hardware physics mode, even though the GPU shouldered the simulation work.

Offloading to the cpu--nothing paradoxical about it, unless you'd expect the GPU to do all the work--OOooops--that's what's supposed to happen, but isn't?...;) It would have been kind of nice if you had shared the cpu utilization numbers here, since apparently you measured them.

We didn't benchmark this particular test, because we somehow couldn't run it with PhysX acceleration disabled.

OK, you explain that for the The Great Kulu demo, but you make no mention of it for the Particle Fluid Demo, but in the PFD results you also do not post any PhysX-disabled numbers. So, can I take it that for the two nVidia supplied demos nVIdia didn't want you to see how either looked or ran sans PhysX acceleration? It's a bit hard to see why you didn't benchmark The Great Kulu for at least the PhysX frame-rates, as it seems that would have been as interesting as the PFD frame-rate results.

My biggest concern in all of this is that the PhysX stuff is going to be used to inflate benchmark scores in popular benchmarks, but for very little else. Hasn't this already happened with FutureMark's Vantage? Maybe I'm just too cynical...;)

Ok... so i buy a physX card today...

http://www.newegg.com/Product/Product.aspx?Item=N82E16814121246

Yea this one will work nicely...

And then i say that i wanna load up some of these maps and such... Do i get the drivers from NVidia now, and do they still force the physics calculations onto the videocard... or will they default them over to the PhysX card which was actually designed to handle these types of things? Sorry if it was already explained.. in which case i fail.

The point i am getting at, is NVidia knows how to push development of certain things into games... and if they are still going to support the dedicated physX hardware, AND be getting game developers to begin actually supporting such a thing on a wide level (as almost any NV videocard will also be able to support physX should you decide dedicated hardware is not for you), then it becomes more plausible that people would buy actual PhysX cards just for the physics acceleration if a wide variety of new games are going to be supporting it on a hardware level. So if that's the case, then dedicated Physics acceleration cards may not be such a dumb thing once more and more games begin to support them. Not saying they will be 100% common in every machine... but it could raise it up into a niche market similar in scope to where the X-Fi sound card is today. Which isn't necessarily a totally bad place to be.

guru3D did a review on it as well , using a gtx 280 as a graphics card and a 9600 gt as a physX card.looks pretty promising....:-)

Nvidia will release new graphics drivers...

So they aren't making it available as a separate driver, at least not initially. That's a shame since WDDM in Vista doesn't really allow two different graphics drivers. I would have considered a secondary lower-end or used (read: cheap) GPU/graphics card for physics once games support it but with the PhysX drivers bundled in to the graphics drivers that is only an option if your main graphics card is NV and I don't like to tie myself down like that.

And why is Havok FX 'apparently defunct'? I know Intel bought them but are they just going to shelve it?

Cmon man.... 35% small ?!? It crippled the card! I prefer to pay for a dedicated physX card to have my 2/3 of my GPU firepower.

At least a nicer demo would take the idea to a new level... but why I would buy in a technology that I had not see working right, that also cripple my gpu?

That is not going to work. Get back to the drawnboard nVidia.

If the GPU will do the video processing and physics what the CPU will do?
A gpu is a video processor ,why put it for performance penalty ?

I like havok because is free and you don't need a second video card to have physics and 100% speed of the GPU .
Is a option but you need to buy the same gpu for sli (1 for physics and 1 for video), Ati has a better option with Crossfire X buy a cheap video for physics and a powerful one for video but not PhysX (for now).

ps put some benchmarks of the 4800 and the 200 series with PhysX!

http://www.youtube.com/watch?v=gLgb9AdnaBI

ATI physics demonstration...from 2006.

I'm assuming nothing much came from it (or ATI didn't feel like it was much of a priority) because game developers are just not interested in bothering with GPU physics.

It's interesting technology, and maybe Nvidia, with a helping hand or two from their marketing and dev relations side, can make this work. But until developers actually start showing a greater interest in this, it's probably going to remain "interesting technology" for a long while yet.

Hhmm... looks like my 2X 8800GT will now face real problem running new physX based games... Hope i can stay with them for atleast 1 year more or else i am at big loss here :(

Well I just got my 260, so this news just makes it that much sweeter.

I can't help but wonder why these "tests" were not performed with at least 2 8800GTs? I mean it doesn't seem logical to me that Nvidia intended for this "feature" to be used with just one card.

Guys, I got a question:

How about multicore CPU support?
Does it balance workload between CPU and GPU?
To rephrase: should you run those benchmark on quad core results would be significanlty higher?
What would be results for running the benches on quad core with ATI GPU?

Do you have another video card that you could throw in real quick and let us know whether offloading the PhysX to another gpu releaves the primary one of that stress and hopefully provides framerate back up to where it was before. (or if you had that Ageia PhysX card laying around, that'd be a great comparison too).

The problem with water is that it gets things wet (almost never shown) and it usually looks totally unremarkable. The obsession with making it blue I think hampers designers from getting it to look correct. In small volumes water is clear and plain. I also don't think they do the physics correctly. Water bounces around like it is blobby-rubber instead of being dense.

I'm just affraid devs will go overboard with neat effects before they finally manage to scale down the effects to enhance instead of degrade everything. But oh well, i guess a few games with blobby enemies wont be so bad.

The problem I have with this is it all comes from Nvidia. Like ATi's cinema demo, it's heavily optimized, and doesn't provide any unbiased tests. This entire article is Nvidia FUD marketing to me, no offense to the author.

Where do we download the particle fluid demo? I can't find it on the nvidia site.

http://www.shacknews.com/onearticle.x/54069
Congrats TR 8)

The problem with that water demo is that water while water CAN behave like that, it would only do so in a model six inches tall -- e.g. like those little desktop fountains.

In real life, a full-size implementation of that contraption (assuming the ladder rungs are sized for a human to climb) would have the water exiting the pipe as a boiling froth, and churning up more air as it landed and splashed. Even if there was a big enough pump to achieve laminar flow, it would distort and difract light through the column much more severely than that, and still froth the water below upon landing and splashes would fire out like shrapnel rather than bounce around blobs.

IMO that's why it looks cheesy. It is realistic, but at the wrong scale.

The first I saw of this story was on the front page, so, of course, I stared at the image before reading the title below it. My first thought was
"That is one really strange watercooling mod."
I guess that supports ludi's comment about the level appearing 6 inches tall. Is it also a mark on how many heatsinks are on most modern mainboards?

I will echo a sentiment already stated; it just doesn't seem to have any substantial impact on gameplay. This seems especially so for first person shooters, a realm where many competitive and semi-competitive players will strip out as many distracting visual elements as possible (eliminating fog, foliage, flares, etc.)

Wouldn't it make more sense to test this with any of the quad core CPUs? Using an older core 2 duo at just 2.13 Ghz is not going to do the software renderer any favors.

Quick question, will my 9600gt support physx?

Is Tech Report now removing posts!
That's news to me ...

"Yes, the water flows sort-of-realistically and fills little pools like it's supposed to. But the liquid has a strange, almost jelly-like quality, and you can see circular "water" particles fly around every now and then. Perhaps a greater number of particles would make the effect more believable, or perhaps better-looking shader effects would do the trick. Either solution probably wouldn't improve the demo's already-low frame rates, though"

This raises an interesting point, surely as you ramp up the graphical detail/resolution the GPU physics calculations are going to be robbing you of increasingly valuable rendering horse power.

AMD and Intel better step it up with Havok FX! Nvidia's PhysX seems to be solid and ready for the mass market. It sucks that my HD 4870 won't be able to play in this arena for a while but this will only promote aggressive competition on AMD/Intel's side so no worries. Hmm, with Big Bang II drivers we may be able to get a cheap Nvidia card as a dedicated PhysX accelerator. I can't wait for confirmation of that functionality.

I would love to see an actual Ageia PhysX card run in these tests as well. Well, considering I have one. But Usacomp has a great point about a secondary video card as well.

If you do these types of tests again in the future, it'd be great to also add the PhysX as a "baseline" and a second card to see if that plan pans out for nvidia/ati as well in their quest to sell more cards.

Honestly I'm not very impressed. Is it cool? Yeah, kinda. But flying hail doesn't make a good video game. The tornado level is much more interesting because the tornado and flying objects actually impact the play of the game.

That water, OTOH, needs to look a lot more like water.

PhyxX is still mainly just a technology with a potentially very promising future as far as I'm concerned. I'd like to experience for myself, but if I don't I'm not going to be too disappointed.