NVIDIA PhysX in Games [Review and Tests]

[zalbard]

Source (in Russian): Overclockers.ru
Translated into English (Google): Google Translate link
Somewhat semi-decent translation of the conclusion done by me:

In 10 games out of 25 installing GeForce GTS 250 or activating PhysX for a single GeForce GTX 275 lead to a significant performance boost.
We expected a smaller number of games in which "physics" will work, but the numbers we got quite surprised us.
We would like to note those games with an option to use both hardware and software physics acceleration: Cryostasis, Ghost Recon Advanced Warfighter 2 and Sacred 2: Fallen Angel. You can clearly see that CPU is no match for graphics cards at hardware acceleration, and performance difference can be as big as 400-800%.
We also made one very important discovery: a system with a single GeForce GTX 275 with activated PhysX in NVIDIA Geforce drivers had a very similar or even identical performance to the system with GeForce GTX 275 and GeForce GTS 250. This shows that buying a dedicated GeForce GTS 250 just for physics acceleration is pointless.
This discovery has another important background - the competition with Radeon graphics cards. If there is an option to turn off physical effects then owners of Radeon don't lose anything but immersion to a degree. But if this option is not available in the game then computers based on GeForce graphics cards have an indisputable advantage over competitors.

Discuss!

[trinibwoy]

Why is the GTX 275 + GTS 250 combo showing higher fps in non-PhysX games? Did they enable some sort of bootleg SLI?

[Clairvoyant129]

Why is the GTX 275 + GTS 250 combo showing higher fps in non-PhysX games? Did they enable some sort of bootleg SLI?

Lol, bootleg SLI

[cegras]

Translation mangles whatever useful data I can get out of it.

This sounds way too fishy for me. According to the graphs, having PhysX on somehow increases performance, and having it on incurs no sizable performance penalty.

Comparing it to CPU acceleration is a pretty sneaky marketing trick, if both yield bad framerates.

[ToTTenTranz]

WTF is this "review"?!?

First they try to install two graphics drivers (ATI + nVidia) in Vista, which everyone knows it's impossible..

Then, in page 2 ("PhysX Off") they show what appears to be.. performance numbers of a GTX275 in SLI with a GTS250?!? There's no such thing!
And the Lost Planet benchmark even shows a performance boost, lol.

But the worst is probably the page 3, where they show performance gains on games that don't even support hardware PhysX like Jericho and Damnation.
The cherry on top is where they show that enabling PhysX shows performance boosts up to 86% on games that use Havok for physics like Deadspace and Kane&Linch. -> there's really no beating that.


To put it simply, the article is a complete joke. The numbers are fabricated by ignorant newbies.

[Chickenfeed]

Physx increasing performance? lol much? I've tried using a GTX 260 in tandem with a 8800gt and although performance didn't decrease as much as just using the 260, it still decreased by a measurable amount ( GRAW2, UT3 and Mirrors Edge )

How they get those numbers in none Physx supporting titles is beyond me... Did they like use the force? Perhaps someone fluent in Russian could care to elaborate? Google translate can only do so much :p

[Hornet331]

Insert "In Soviet Russia" joke here.

[Clairvoyant129]

This article is a big pile of crap. PhysX is one big joke... enabling PhysX with 2x GTX 260 Core 216s on my Core i7 system decreases performance in most of those so called, "PhysX enabled games." And adding my old 8800GTX to use as a dedicated PhysX card brings down performance even more.

PhysX is worthless.

[zalbard]

On the page two, you can see results for 275+260 where 260 is acting like a dedicated PhysX processor [PhysX ON in Nvidia CP] vs 275 only [PhysX OFF in Nvidia CP]. So ofc you see a performance improvement in most cases. Not rocket science.
Regarding the page 3 - well, nfc, I haven't played any of the games listed so hard to say whether enabling PhysX actually boosts the game performance or not for a single card. Sounds fishy, as mentioned above.

[Mad1723]

Insert "In Soviet Russia" joke here.

In Soviet Russia, PhysX increases performance

When I read the result, the bottom result should have been with physx On, not the other way around... Honestly, sounds like either it's a big joke or some guys got money from a green giant

[Xoulz]

So adding another $100 to your video sub-sytem, speeds it up...? No way !!

[zanzabar]

This article is a big pile of crap. PhysX is one big joke... enabling PhysX with 2x GTX 260 Core 216s on my Core i7 system decreases performance in most of those so called, "PhysX enabled games." And adding my old 8800GTX to use as a dedicated PhysX card brings down performance even more.

PhysX is worthless.

your comparing physX dissabled (its been worthless in everything that ive tried) to having it enabled. with it enabled and gpu physX disabled u will get choked but with things like mirrors edge or UT3 that use it by enabling it there is no real diffrence things just break into little uniform pieces instead of medium uniform pieces. its nothing like havoc that works way better since its designed for the cpu and can do non uniform breaking. the new red faction looks better on a console with geomod that uses havok than anything that ive ever seen on physX

there is a reason why intel bought havok and not agea, they could have bought both but physX is garbage as an api

[saaya]

for the ones who dont want to go through the article for details.

GTX275 + GTS250 as physix card performance improvement over
GTX275 with physix OFF (?)
(basically they claim adding a dedicated physic card not only adds better eye candy but ALSO gives the below mentioned fps boosts)

clive barkers jericho 20%
cryostasis 30%
damnation 60%
dead space 90%
empire total war 25%
ghost recon advanced warfighter 500%
kane and lynch dead men 45%
lost planet colonies 5%
mirrors edge 25%
sacred2 fallen angel 850%

so if you enable physix you expect a performance drop over no physix, right?
and if you add a dedicated physix card or ppu, you expect that drop to be smaller or maybe even zero, right? but how can you get MORE fps by enabling physix than you would bet with it disabled? adding a dedicated card or not, how can enabling a hardware hungry feature result in more performance than having it disabled?

Why is the GTX 275 + GTS 250 combo showing higher fps in non-PhysX games? Did they enable some sort of bootleg SLI?

i was thinking the same thing... dead space supports physix?

To put it simply, the article is a complete joke. The numbers are fabricated by ignorant newbies.

yeah, it seems the number are completely made up...
amazing, who would have thought we see something like this in 2009...

Physx increasing performance? lol much? I've tried using a GTX 260 in tandem with a 8800gt and although performance didn't decrease as much as just using the 260, it still decreased by a measurable amount ( GRAW2, UT3 and Mirrors Edge )

exactly... if they would compare physix on a gtx275 with gtx275 and dedicated gts250, ok... this would have made some sense actually... but they say physix is off on the gtx275 only config...

lets assume its a typo and physix is actually enabled on the gtx275 only config... then those numbers still dont add up... so this seems like a really sloppy attempt at fabricating a review...

Insert "In Soviet Russia" joke here.

in soviet russia, the independant press is sponsored by nvidia ^^

This article is a big pile of crap. PhysX is one big joke... enabling PhysX with 2x GTX 260 Core 216s on my Core i7 system decreases performance in most of those so called, "PhysX enabled games." And adding my old 8800GTX to use as a dedicated PhysX card brings down performance even more.

you didnt mention that all the eye candy and "game immersion" benefits you get though ^^

seriously, its hard to tell in most physix enabled games if physix is on or not
the easiest way is to check the fps

its really sad though... game physics had and still have such a huge potential... and im worried nobody will focus on it cause nvidia got everybody so fed up with their crappy physix implementations and marketing nonsense

sigh...

[RPGWiZaRD]

Err, of course PhysX means games will run slower than not utilizing it at all but what PhysX on GPU's purpose is to do this way faster than CPU would. UT3 gets over 200% performance increase with PhysX enabled on GPU on the PhysX enabled maps, a matter of playable and not playable scenario's on 95%+ of computers.

What I've seen so far is the efficiency of using PhysX on GPU is very good even on today's cards, now the next nvidia chip will most likely have even better PhysX-capabilities.

Would be of course better if you'd be able to efficiently do it with the CPU but I don't think physics on CPU will be a good move before we get Larrabee-like CPUs with really many cpu cores, but even then it will probably be more difficult to maintain a physics code for so many CPU cores, unless they'd come up with a groundbreaking multithreaded code in the API.

[saaya]

Err, of course PhysX means games will run slower than not utilizing it at all but what PhysX on GPU's purpose is to do this way faster than CPU would. UT3 gets over 200% performance increase with PhysX enabled on GPU on the PhysX enabled maps, a matter of playable and not playable scenario's on 95%+ of computers.

yeah but what these guys are saying is, adding a physix card and enabling physix means you get more fps than with physix disabled!

now lets say its a typo and they mean physix enabled on the gtx275 vs dedicated gts250 for physix...

clive barkers jericho 20% - no big deal
cryostasis 30% - no big deal
damnation 60% - never heard of this game?
dead space 90% - this games uses havok physix so... so i have no idea what they are talking about
empire total war 25% - no big deal
ghost recon advanced warfighter 500% - sounds a bit weird and never heard of this game
kane and lynch dead men 45% - never heard of this game
lost planet colonies 5% - no big deal
mirrors edge 25% - no big deal
sacred2 fallen angel 850% - sounds a bit weird...

1.) most games with physix are second or third class games or even less
2.) even the better ones dont show a notable enhancement with physix enabled
3.) the little cool things that physix brings in those few games could be done on cpus as well (and yes, the physix cpu mode is terribly slow, but that says more about the physix cpu mode than cpu performance of physics effects )

[R101]

CPU does some kind of physics anyhow, it might not be as complex as GPU version, but when you offload all and any kind of physics calculations from CPU it will get you a speedup, I think.

[RPGWiZaRD]

yeah but what these guys are saying is, adding a physix card and enabling physix means you get more fps than with physix disabled!

Well it would make sense if the games that also had physics code made for CPU, if somehow PhysX would be able to run that as-is on the GPU. But that's where it fails, it just doesn't work like that, either it's got PhysX support or not, can't make a HAVOK code run with PhysX just like that.

And no I didn't waste my time reading that article.

[YukonTrooper]

Insert "In Soviet Russia" joke here.

In Soviet Russia, vodka and PhysX review is best summer!

[Talonman]

Needless to say I am a PhysX fan, but I don't know what to make out of this article...

All I do know is in Cryostasis, using a 280 for dedicated PhysX does indeed give you a performance increase over no dedicated PhysX card as expected, but that is with PhysX enabled!

Having a dedicated PhysX card, and turning PhysX off for better performance is just out there...

[ToTTenTranz]

Some people are still trying to rationalize this article?!??? Oh come on!
No one read my previous post at all?

They're showing performance boosts on games that don't even use PhysX!!
Deadspace and Kane&Linch use Havok.

Please, let's not give any more importance to these guys..

[yury2808]

Actually, if you have a fast renderer, and your software is speed-limited by CPU, using PhysX _can_ improve a performance in terms of fps. In all other cases - it _should_ be slower, because of vastly increased polygon count.

ps. For example, if the game do not increase detail level, but rather "offload" the effect calculation to videocard - the fps can be increased by a huge amount, since cpu have more spare time to send all that calculated data to gpu for rendering.

[SKYMTL]

empire total war 25% - no big deal

IMO, yes it is. E:TW has enough issues with Crossfire and SLI setups as it is let alone somehow, some way it is getting a performance boost in the configurations they tested. Especially when you considering that this "25%" they saw could seriously take Empire from a literal slideshow to a smooth-running app in no time.

I hate looking at articles like this because it ends up giving the site hits that it just doesn't deserve.

[saaya]

IMO, yes it is. E:TW has enough issues with Crossfire and SLI setups as it is let alone somehow, some way it is getting a performance boost in the configurations they tested. Especially when you considering that this "25%" they saw could seriously take Empire from a literal slideshow to a smooth-running app in no time.

I hate looking at articles like this because it ends up giving the site hits that it just doesn't deserve.

according to their numbers they went from 21/23 to 24/29 (min/av)
i dont think thats a notable improvement for a game like empire total war where you dont need that many fps... if 21/23 is too slow for you 24/29 will still be too slow for your taste :P

and theres an easy way to watch a sites content without visiting it... google cache

[RaZz!]

1.) most games with physix are second or third class games or even less

just because you never heard of these games they're "second or third class games or even less"?

these games a quite widespread, especially ghost recon.

[doompc]

I think that by Physx OFF they mean that the CPU is in charge of PhysX, when enabled the PhysX is done by the GPU, hence the performance boost.