My Test of Cell Factor running with and without physX

Lord of Shadows said:
It would be funny if the new patched cellfactor removed the enablephysx flag along with its other updates.
Click to expand...

Yeh I made a comment about this on my home forum (OcUK), will they be working on improving the ppu performance or tweaking it so that it's impossible to play it without a ppu, if they had archieved this in the first place then we would of all been fooled into thinking that is was the ppu doing all of those fancy physics.
 
The flag was to be removed with an updated version which was supposed to fix some things.

As for performance with/without the PPU, it will work fine without it, as has been said... but I would suspect that Cloth physics aren't working right, or if they are, they're intensive on the CPU ( which I know they are)... Liquid Physics will also not be enabled without PPU usage, so if you shoot a barrel that should spew oil or such, it will do nothing.

A majority of the CPU usage is unfortunately from the Renderer/Engine side, than the Physics side. It has to do more so with DirectX rendering to a point.


PS. I played this originally without a PPU and to make it even remotely smooth running, required removing Cloth banner, removing a few objects including the Liquid spewer (barrels), and killing alot of the boxes and such.
 
lowrider007 gets 13fps@640x480 without and 19fps with the PPU.
This is a speed boost of almost 50% percent, which
is pretty much. On the other hand, he even has 1fps less at 1024x768 when using the PPU. This states to be a bad comparison benchmark.

Not only different initial condition degrade the expressiveness of this test, but heavier use of effect physics, when running the game using the PPU, as well.
There are more particles used in HW-Mode, and fluids and cloth is completely missing in the SW-Mode.
I dont know how much difference this actually is, but I estimate it to about 10%.

The last thing is the high per-object cpu costs in the Reality Engine, which drastically reduce the advantage of the PPU over the CPU Physics. Each object needs its state updated, you need to test if shadows are casted or received, DirectX Api calls must be done to update the shadowdepthmaps(for dynamic shadows) and you finally need to render the object.
If the physics update part would take no time, you still wouldnt get more than like 100% more performance. There is a lot of forthcoming optimization work on the Realitiy Engine side, which is going to increase performance in both HW and SW Mode, but which is going to increase the difference between HW and SW Mode performance as well, since the percentage of time spent in physics caculation increases, which lets you feel the difference between ppu and cpu physics performance.

And of course, we may look forward to better PPU drivers, since the current one isn't really optimized.

One thing which is for sure, there are currently no killer apps out, which can't help you purchase a PPU.
 
BoredTiger said:
I didn't buy the PhysX card to get better frame rates.
Click to expand...

Yes you did, the PPU doesn't do anything special when it comes to physics, it's just faster at doing it. Theoretically theres nothing to stop it all being run on the CPU, it would just be very slow.

The frame rate comparison was done to give users a good idea of how much faster the PPU calculate the physics for the scene, and the answer seems to be a 1-2 FPS if you have a decent CPU.

OK the cloth and liquid physics aren't available, but thats not a technical constraint, it's perfectly possible to do that stuff on the CPU it would just be a fair bit slower. Don't get confused between what the PPU is actually doing, it doesn't have some magical chip inside which suddenly allows it to do things a CPU cannot, it's merely an accelerator, it does this stuff faster.

As for PCI bus bandwidth limitations, I don't think so, I'm fairly sure most people are over estimating the amount of data this thing is actually going to send in a game like cellfactor, it's essentially just lots of vector data. It's not rendering the objects, it doesn't need to send textures or complex data about lighting just a point position of each object, it's rotation and current vector of velocity and acceleration, probably a bit more stuff but thats the essentials, not a lot of data at all.

If my internet connection can handle it, then the PCI bus surely can, quote from google search I just did returned

The original PCI bus was designed to operate with a 33MHz clock, to provide data transfer speeds up to 132 Mbytes/sec
Click to expand...

I was playing a 1v1 with no bandwidth problems and I only have 448k upload...
 
yeah i agree with u, there simply isnt an app out there worth the money of the PPU.

----
anywayz i cant play the demo. the start menu buttons appear blank and doesnt work only the supposed QUIT button works.

bad gpu config?
 
byteasc said:
PS. I played this originally without a PPU and to make it even remotely smooth running, required removing Cloth banner, removing a few objects including the Liquid spewer (barrels), and killing alot of the boxes and such.
Click to expand...

I'm interested to know how you got the game running and have the cloth/liquid physics working while not using a PPU, if you could explain how then we could have proper frame rate comparisons between the PPU and CPU
 
Frosteh said:
I'm interested to know how you got the game running and have the cloth/liquid physics working while not using a PPU, if you could explain how then we could have proper frame rate comparisons between the PPU and CPU
Click to expand...

I didn't have the game running at 100% unfortunately. You use to be able to enable Software Cloth (which the flag may be removed now) and there was potential for "Software" Liquid, but I think that was for preliminary testing, as the Liquid actually does use specific rendering functionality tied to the PPU SDK.


You could easily use Reality Builder/Editor to clean up the Level and remove amounts of boxes to balance it out and make it not as hard on the CPU, but still give a general solid feeling to everything.

Heck, You could do a whole new level fully custom, but it all depends on implementation and time.
 
Not especially interested with the standard pile of rigid bodies all box shaped.

More interested in the cloth and liquid simulations on software, obviously this demo has gone through a few itterations of testing and that stuff is no longer supported in software.

I don't think loads of box shaped primitives is much of a strain even for just a CPU, as i've mentioned on this board before somewhere, when garrys mod was first released for HL2 I spewed out 100's of computer monitors into a river and they were all bobing around and knocking into each other without causing any real frame rate drop.

A high end CPU can deal with a loads of these physics calculations fairly well.
 
Frosteh said:
A high end CPU can deal with a loads of these physics calculations fairly well.
Click to expand...

The physics simulations we see in games are simply fooling your eyes. They're designed to approximate reality by fudging enough parameters of "real" objects to give the illusion of true interaction. They're not simulating reality, they're enacting an artificial model with a limited set of parameters.

While any physics engine is limited to the parameters you can feed into an entity, such as surface density, weight, elasticity, etc, the advantage of the PPU is being able to process a much more detailed and complete set of parameters, allowing for more realistic games.

Take the original Deus Ex, for example. It had a simple physics model that allowed you to, say, bounce basketballs around. The ball entity had a very simple set of rules that governed it, and nothing like elasticity, rebound rate, rotation, etc, to affect the way it behaved. It was most probably box-bounded, too (i.e looked like a ball but acted like a cube).

Later physics models like Havok increased the parameters like rotation around vertex axis', center of gravity, buoyancy, and so on. This is a great beginning, and allows great looking object interaction. Nevertheless, the Havok software engine is neccessarily limited in the amount of parameters it should process per entity, simply because it has to run on a general purpose CPU which has to do other things too. So many "effects" are merely clever hacks and fudges and are not true physics calculations.

The power of the PPU is unquestioned. The problem is that not everyone has one, so game designers have to create "limited", ie. simple, objects in their games, so the PPU isn't neccessarily putting its best foot forward. The PPU is in a bit of a catch-22 situation now that it is launched... games that are designed for it wont sell well, because few people own one. Game designers will create for the biggest slice of the bying pie, and the games that are made, are thus made to run okay on non-PPU hardware, so the PPU is sitting idle, looking like it's a useless piece of shit.

Trust me, if you create an environment with many highly detailed, deeply parameterized and interactive objects, the CPU will slow to a crawl loooong before the PPU will.
 
Frosteh said:
Not especially interested with the standard pile of rigid bodies all box shaped.

More interested in the cloth and liquid simulations on software, obviously this demo has gone through a few itterations of testing and that stuff is no longer supported in software.

I don't think loads of box shaped primitives is much of a strain even for just a CPU, as i've mentioned on this board before somewhere, when garrys mod was first released for HL2 I spewed out 100's of computer monitors into a river and they were all bobing around and knocking into each other without causing any real frame rate drop.

A high end CPU can deal with a loads of these physics calculations fairly well.
Click to expand...
So conclusion wich we can make now is. Software load =/= Hardware load. For CF demo
So benches are disqualified.
Or disable all hardware effects that are hardware exclusive. To make the load equal.
But then again with PPU you only would play full enabled.
Without those exlusive features the software performance say nothing. only about games that would only use those software enabled Physics features. Wich CF is not.
How ever you can weight Physics feature to there gaem value.

Or The PPU can handle a lot more but the game use only 200 á 500? wich a CPU can do to. But P1 is capable to do 20.000 but minus Clooths and liquid physics feature it could handle 5000 in the worst case senario if there is a Physics computing piek with all three or more? physic diciplines in use. And if the game uses a light load like 200 á 500 there isn't much a difference but all physics features add up.

Besides the box Physics demo in the driver controle pannel works with and without PPU and It works well. But this is only óne pure Physics Feature based demo.
 
It will become Cellfactor: Revolution, and the Monster Madness demo will be released as a better PhysX demo.

--------------------------------------------------------------------------------

I can't go into full details yet, but here's general info about some of the new stuff:

- 4x as many interactive objects as CellFactor: Combat Training
- Fully destructable architecture
- 16 player online support, server-side physics density configuration for varying bandwidth capabilities
- Player character classes, including Weapons or Psi-oriented characters
- Extremely powerful new Psi Abilities -- become a living god.
- Plenty of new maps, new weapons, and new vehicles, including a treaded tank and a mech power-suit
- Dual-wielding, trading Psi Power for an extra weapon.
- New core gameplay modes (such as Assault), and physics-oriented sports modes
- Improved bots, supporting all gameplay modes, psi abilities, and vehicles

You've been trained. Now, it's war.

And while this isn't official yet, I think CF:R will be sold for ~$20. That's our goal

Will keep you posted with new media soon!

--------------------------------------------------------------------------------

One of the main misconceptions about using CellFactor: Combat Training as a PhysX benchmarking tool (not our intent with it), is that even though there are several thousand interactive objects in the level, it's rare (or even impossible) that all or too many of them are moving at once. Thus, the ones that are "sleeping" consume no CPU or PPU calculations, and hence the game becomes "more or less" runnable on the CPU, only bogging down during big explosions or gravity-warping events. A game that has more objects in "perpetual motion" (or more fully destructable environments, such as CellFactor: Revolution will have) can more effectively demonstrate the very real performance difference between CPU and PPU.

To that end, we're going to release our old Monster Madness PhysX demo (ala http://www.monster-madness.com/MonsterMadness_PhysX.avi ) which has a couple thousand objects that are almost always in motion within a small environment. There'll you get to see more of a performance spread than is visible in CellFactor: Combat Training (at least with the old PhysX drivers that we developed it on).

-Jeremy (from Artificial Studios forum)
 
Gilhooley said:
I can't go into full details yet, but here's general info about some of the new stuff:

- 4x as many interactive objects as CellFactor: Combat Training
- Fully destructable architecture
- 16 player online support, server-side physics density configuration for varying bandwidth capabilities
Click to expand...


YES!
Now we are talking:
"Fully destructable architecture" :D


One of the main misconceptions about using CellFactor: Combat Training as a PhysX benchmarking tool (not our intent with it), is that even though there are several thousand interactive objects in the level, it's rare (or even impossible) that all or too many of them are moving at once. Thus, the ones that are "sleeping" consume no CPU or PPU calculations, and hence the game becomes "more or less" runnable on the CPU, only bogging down during big explosions or gravity-warping events. A game that has more objects in "perpetual motion" (or more fully destructable environments, such as CellFactor: Revolution will have) can more effectively demonstrate the very real performance difference between CPU and PPU.
Click to expand...

Thanks for clearing up af lot of FUD! :)
Looking forward to seeing it
*send mentally headbutt's to my dealer...ship my PhysX NOW!* :D

To that end, we're going to release our old Monster Madness PhysX demo (ala http://www.monster-madness.com/MonsterMadness_PhysX.avi ) which has a couple thousand objects that are almost always in motion within a small environment. There'll you get to see more of a performance spread than is visible in CellFactor: Combat Training (at least with the old PhysX drivers that we developed it on).
Click to expand...

That sounds awesome...perhaps then we can cut through the PR-Bull and get the men and boys seperated...
(read get the REAL PPU seperated form the eye-candy GPU's)

Terra - Thanks for saving my day :)
 
Terra said:
YES! Fully destructable architecture :D
Click to expand...

I wouldnt get my hopes up until we know a bit more about it, ;o) which is probably one of those things we wont hear about for a while.
(/me thinks this refers to the boxes breaking apart etc)
 
When hardware and games are designed in conjuctions you get a performance boost, such was the case with the 6800s and Doom3. However, this is not the case. I personally believe that the card does absolutely nothing and WILL do absolutely nothing in the future. Its a marketing gimmick just like when you buy a Honda with VTEC thinking it would be faster and better than any other car, but in reality...its just regular ol' VVT like on any other car...marketing gimmick.
 
Endurancevm said:
Its a marketing gimmick just like when you buy a Honda with VTEC thinking it would be faster and better than any other car, but in reality...its just regular ol' VVT like on any other car...marketing gimmick.
Click to expand...

To paraphrase the following website:

"For straight power output, the V-TEC system craps all over the VVT system". Wow... didn't know car buffs were into that stuff, but hey - whatever floats your boat.

Anyways, have a look-see:

http://www.billzilla.org/vvtvtec.htm

Getting back on topic, however...

Generally speaking, doing something in hardware with a purpose driven chip like a PPU should be loads faster than handling something with a general purpose chip like even the most modern x86 CPU. (... with its associated x87 FPU and (hopefully IEEE compliant) SSE/SSE2 implementation.) Think about the days of yore when a DVD took 50+% of your CPU to decode. Your CPU was big, got hot, and used quite a lot more power than the dedicated MPEG-2 decoder chips. The dedicated chips used fewer transistors, and were much simpler to design, but they sure do the job.

I imagine the PPU will be similar once it catches on. Maybe it is just a bit ahead of its time...
 
Oh offtopic.

It totaly doesn't surprise me. That Vtec is better then VVT Performance wise. Because Altho VTEC can be use also in a economical way.

VVT is like varicam a Variable valve timing solution.
Where VTEC isn't variable but a two fix set solution. The pin's. But most importantly a different Valve lifting. It gives a normal setting and a race setting. For timing. But also valve lifting. Well it's RPM more matches a Motor cycle engine. But to get the most out of it you must like to ride with high RPM. I am not that kind of guy. Cruising on torque not PS/ PK /KW.

On topic:
As for PPU. Well if CPU have twice the power of PPU.
CPU is a non dedicated PU wich run a lot of task. If it would be a quater for Physics, the PPU is as twice as fast. So wenn a PPU is 5 time as fast as in Gigaflops to a CPU it's dedicated so 20 times more Physics.

Then there is a difference in efficenty depending on code mix. A PPU will handle Physic related code very well.
 
OT:

IMO please don't make statements about cars if you don't know much about them. VTEC and VVT both make a difference, and often times they help out a lot. VTEC has the two different profiles that switch at a given RPM, and VVT from what I've seen helps to flatten out the torque curve by advancing and retarding the cams where the computer sees fit. VTEC and VVT aren't just marketing hype, they do something.

on-topic - You need averages and min/max to make a reliable benchmark, and even then it's kind of iffy in CF as it's hard to reproduce exactly the same scene. This is just my opinion though :)
 
You're telling me VTEC does nothing?

Have you ridden in my RSX Type-S when the tach crosses 4600RPM at >90kPa?


Do you even know that VTEC does not only alter the valve timing but physically rotates the camshaft to change lobe profiles? Would I REALLY want to rev just shy of 9000RPM on my low cam? HELL NO
 
kent said:
You're telling me VTEC does nothing?

Have you ridden in my RSX Type-S when the tach crosses 4600RPM at >90kPa?


Do you even know that VTEC does not only alter the valve timing but physically rotates the camshaft to change lobe profiles? Would I REALLY want to rev just shy of 9000RPM on my low cam? HELL NO
Click to expand...

Alrighty this is getting way OT!

and yeah, VTEC does have an effect...especially in my S2000 once I hit 6000 :D
 
Hey I got an idea, let’s all talk about random car crap in a PPU thread.

Brilliant!
 
That's that and PPU do make a difference. But enough to pay for? Most will think not, but I do and some others..
 
fluid effect off and cloth effect off
you can play cellfactor without physX.. what a descovers !!!

Without PPU 1280x1024 8xAF HDR Using High Graphics Shortcut : 13 FPS
Without PPU 1024x768 8xAF HDR Using High Graphics Shortcut : 13 FPS
Without PPU 800x600 8xAF HDR Using High Graphics Shortcut : 13 FPS

the graphique card is what ? a 9999 GT.. mdr magic systeme... ridiculous

now then watch this.. a real test

cell factor test with cloth and fluid with and without PPU. with a core 2 duo e 6600

and STOP FAKES TEST
 
same video but for lowest connexion... little cut and better compressed

i chose 640 480 for record with "fraps", not to kill the pross when it is calculing phisX effect
and record at the same time....

core 2 duo e 6600 can make both of them but i'don't want to slow down the game more than it is without PhysX ^^ like you see

when HDR fluid and cloth effect ON with the "enablephysiX=false" it's only possible
to play when you don't look at the clothes ^^.... so PhysX card really works like it must work..

Tomorow for example the clothes: will be used on all the characters on them head to make hair, on their clothes and also to make grass and to make leaves and so on...


The Video card that don't change FPS when resolution change should be on Guinness World Records.... i wanna buy it to put it on my wall of bedroom.

this item should'nt not even exist... PEOPLE DONT EVEN RESPECT work of other people working hard to create a card that works for good. This represents 2 or 3 years of really hard work. DON'T FORGET IT !!!
 
I did not understand your response. the test is silly because it tell us nothing. no one plays at level resolution. it is not a fair test of anything because more of the cpu would be used at a higher resolution, so we it does not tell us any useful information, about either the effectiveness or non-effectivness of physics acceleration.
 
when your are recording a movie... pross work little and hard drive work sérious.

If the movie is big.. (the resolution is BIG) 1024/768.. pross work much and hard drive die.
BUT if the movie recording it's small (small resolution 640/480) pross work very little.
and hard drive work little...

So when you play cellfactor at 1024/768 with PPU and your are recording on the same
time.. you lose 4 FPS
when you play cellfactor at 640/480 with PPU and your are recording on the same
time.. you lose 1 FPS

when you play cellfactor at 1024/768 without PPU and your are recording on the same
time.. you lose 6 FPS
when you play cellfactor at 640/480 with PPU and your are recording on the same
time.. you lose 3 FPS

(with my matérial)

THAT WHY i chose 640/480 for the movie.. for the recording don't change (or just a little)
the REAL FRAMERATE.... if you don't anderstand any more go play PS2. and nerver talk something about computeur, silly mind.

SO you can see, even in low resolution.. pross DIE with physX effect activate
and in big resolution pross DOUBLE DIE ^^
 
I understand computers just fine. I am having difficulty understanding the structure of your sentences.
 
@ [VodK@]KYO

Either way you should use the search function more... What your doing is useless as a comparison. If you want to compare software and hardware in Cellfactor you should use Realitymark and read the whitepaper.
 
------------------------
I ran the game at 640/480 and FRAPS recorded at 320/240... sorry for my bad english.
If i ran the game with bigest resolution. the test could be corrupted by the record.
------------------------

REALITY MARK is good but it don't use BI PROSS and multi threading...
so, some FPS should be lose in SOFTWARE mode if you have a BIG pross like mine.

REALITY MARK give me 2...3 fps for my core 2duo e6600...
If multi threading is activate (like in the game cellfactor) i think core
2 duo can make 5 or 6 fps :p yes it's not a big difrent but... LOL ;)
 
TTT EPIC EDITION!!!!!!!!!!!!!!!

Anyways, has anyone been able to get this game to work with the new PhysX pack yet?

Aegia 8.08.01 and Forceware 177.83?

*edit* Doh, even with search, someone beat me with another thread. Sorry :(
 
You must log in or register to reply here.
Back
Top