Cryostasis: CPU vs. PPU vs. GPU...

Atech

2[H]4U
Joined
Jul 14, 2007
Messages
3,946
Link to techdemo:
http://www.bigdownload.com/games/cryostasis-the-sleep-of-reason/pc/cryostasis-physx-tech-demo/

Thought I'd give people a baseline to go from.

Rig in sig, all stock.
DX10
Quality Presets ingame: All High
No sound
Lastest WHQL, CP set to quality, all app controlled.

CPU PhysX(GPU + PPU PhysX disabled)

PPU PhysX(GPU PhysX disabled)
GPU PhysX(PPU PhysX disabled)

800x600
Min 2.9
Max 83.4
Avg 7.8

Min 9.5
Max 98.6
Avg 23.2

Min 19.6
Max 97.9
Avg 42.2

1024x768
Min 2.8
Max 81.7
Avg 8.2

Min 9.4
Max 79.8
Avg 22.4

Min 18.4
Max 81.7
Avg 34.6

1280x1024
Min 2.9
Max 77.9
Avg 8.1

Min 9.5
Max 70.5
Avg 21.4

Min 15.4
Max 72.0
Avg 25.8

1600x1200
Min 2.8
Max 57.0
Avg 7.9


Min 9.5
Max 56.8
Avg 19.5

Min 12.6
Max 65.6
Avg 20.1

Anomali PPU: All 4 runs zombie fell through floor after killshot.
Anomali PPU: 3 times out of 4 barrel go stuck in railing.
Anomali PPU: 4 times out of 4 tin can for shot had fallen due to "internal energy".
Anomali ALL: Dust in room frozen when moving.

One time for the fun of it:
1600x1200 with sound (Open AL - EAX HD)
Min 12.4
Max 62.6
Avg 20.2


Will do DX9 tomorrow (Dang that demo is a tad long :s *L*)
 
Thanks for those numbers.
Looks like the Q6600 is a no-go...
PPU does a good job, but no better than the 8800GT, so no reason to have it in your system. At low resolutions it seems to limit your GPU more than GPU PhysX does.
 
Thanks for those numbers.
Looks like the Q6600 is a no-go...
PPU does a good job, but no better than the 8800GT, so no reason to have it in your system. At low resolutions it seems to limit your GPU more than GPU PhysX does.

Yup,PPUs are dead..Far more efficient on a GPU..

1920x1080 all setting high 16xAF Sound On..

GTX 280 700/1460/1275 and 9800GT SC(PhysX)..Rig in sig

Total Time 201.929s
Total Frames Count 9821
Average FPS 48.6
Minimum FPS 32.1
Maximum FPS 112.1

Disable 9800GT..PhysX on GTX280

GTX 280 PhysX

Total Time..248.009s
Total Frames Count 9821
Average FPS 39.6
Minimum FPS 24.2
Maximum FPS 95.2
 
It ran slow on my quad up until leaving the lowest water area, then when it did all the effects on their own one by one, first the water, then the barrels being launched into a large canvas then shooting the cloth, that was all running about 100fps.

Theres only several effects throughout that really need physics hardware acceleration, the water drops, the particles hanging in the air and the large complex coth physics, they constitute only a few effects in the scene, everything else is totaly do-able on the CPU, the ridgid body physics the pooled water at the bottom with floating barrels, the water streaming down the walls and surfaces making it all look wet...

Be interesting to see how well that runs on the i7
 
It ran slow on my quad up until leaving the lowest water area, then when it did all the effects on their own one by one, first the water, then the barrels being launched into a large canvas then shooting the cloth, that was all running about 100fps.

Theres only several effects throughout that really need physics hardware acceleration, the water drops, the particles hanging in the air and the large complex coth physics, they constitute only a few effects in the scene, everything else is totaly do-able on the CPU, the ridgid body physics the pooled water at the bottom with floating barrels, the water streaming down the walls and surfaces making it all look wet...

Be interesting to see how well that runs on the i7

Those are also the points for running hardware accelerated physis? :confused:
 
Those are also the points for running hardware accelerated physis? :confused:

Yea exactly... Remove those and you're back at where Half-Life 2 was a few years ago.
I think people still aren't seeing the big picture here.
 
Flying debris, shock waves, flipping cars, smokes and bent metals? Hell yeah, much more than curtains and dripping water. :p

Ironically enough cloth and water are far more complex and time-consuming effects. Which is why those are being shown off most in demos.
 
Ironically enough cloth and water are far more complex and time-consuming effects. Which is why those are being shown off most in demos.

So flying debris, shock waves, flipping cars, smokes and bent metals can be done on the CPU? In that case, I don't think that I need a PhysX accelerator any time soon. :D
 
Ran on the rig in my sig.


1920x1080 everything maxed 8xAF

Total Time 289.198
Total Frames 9,821
Average FPS 34.0
Minimum FPS 21.1
Maximum FPS 93.4

I have a 9600GT sitting here I think I will install and run another test with a dedicated PhysX GPU. I'll post the numbers when finished.
 
So flying debris, shock waves, flipping cars, smokes and bent metals can be done on the CPU?

If you keep the detail level low enough, sure.
Same goes for water and cloth... it's just that the detail would be unacceptably low before you got into playable framerates, so games never used those effects.
 
So flying debris, shock waves, flipping cars, smokes and bent metals can be done on the CPU? In that case, I don't think that I need a PhysX accelerator any time soon. :D


You are really troll...ANYTHING can be done on the CPU, but th CPU soons runs out of horse power, so if you don't want to evovle beyond HL2 physcis...then yes the CPU is enough for you..:rolleyes:
 
I thought that they are just simple collision physics and CPUs are also getting faster with the cores are being doubled for each new CPU release. Btw why can't we just have a discussion without the name calling?
 
CPUs evolve at a much slower rate, and they are far less efficient with parallel tasks.
Doubling the amount of cores gives you far less than double the performance in most cases.

Which is why we've pretty much been at a dead end as far as physics go, for the past few years.
GPGPU finally gives us a breakthrough in performance.
 
I know that hardware accelerated physics is a good thing but for effects like in this video, can a quad core CPU do that kind of effects already or do we need a dedicated accelerator for that. After all, they are just "simple collision physics", not the more complex water and cloth physics.

Too bad that I'm more interested in those kind of effects than the curtains and dripping water.
 
Those are effects more than they are physics. It's mostly just eyecandy. Non-interactive smoke effects and such.
 
New numbers, from the new rig in sig, everything stock:

1600x1200 HQ CP settings 16xAF

Min 21.5
Max 879
Avg 50.4

Using the GTX285 for physics.
 
Well, like I mentioned in another thread, the game doesn't actually seem to bother assigning any CPU power to the physics. When I used my 4830, the game only utilized one CPU core, and that core never went above 80% usage, usually it was around 50-60%. And that was for the entire game, not just the CPU-physics. Running the game with a 9800GX2 now, the CPU usage is actually the same, only now it's spread out over two cores( about 20-30% on each of the two ).

So in short, there doesn't actually seem to *be* any CPU physics. If you don't have a PhysX compatible card, the game won't even try to make up for it by utilizing more of the CPU.
 
Isn't the game just single threaded, only using one cpu core?

For a game today, that is a big and unfortunate waste. For all the hardware hammering this game seems to do, I can't say I'm terribly impressed with what I see.

Seems like they need to tighten up the code a bit.

Guess I won't be playing it anytime soon... No biggie, got plenty of others I can pull out of the cabinet and unwrap.


More Physx goodies on the way to further drag the game down. - http://www.cryostasis-game.com/html/news.php

1C Cryostasis forum - http://forum.1cpublishing.eu/forumdisplay.php?f=90
 
The hottest PhysX config.

http://www.evga.com/forums/tm.asp?m=568129&mpage=2

System was at: Q6600@ 3.81GHz
(1) 295 set to C=702, SH=1548, and M=1188. (Processing graphics)
(1) 280 set to C=756, SH=1512, and M=1296. (Dedicated PhysX)

Average FPS= 78.5, Minimum FPS= 36.2, and Maximum FPS=140.4

I wanted to see exactly what my 280 was doing for my 295...
Answer:

My 280 running in dedicated PhysX mode, is adding:
24.8 FPS onto my Average...
5.5 FPS onto my Minimum...
13.2 onto my Maximum...
over what my Single 295 could provide alone, at 1920x1200 with Max detail settings.

Apparently Tri-280 Systems running withought a dedicated GPU PhysX processor, or Tri-280 Systems with a lesser dedicated PhysX processor, can't produce the frame rates my PhysX rig can. ;) :D
 
Last edited:
Isn't the game just single threaded, only using one cpu core?

For a game today, that is a big and unfortunate waste. For all the hardware hammering this game seems to do, I can't say I'm terribly impressed with what I see.

Seems like they need to tighten up the code a bit.

Guess I won't be playing it anytime soon... No biggie, got plenty of others I can pull out of the cabinet and unwrap.


More Physx goodies on the way to further drag the game down. - http://www.cryostasis-game.com/html/news.php

1C Cryostasis forum - http://forum.1cpublishing.eu/forumdisplay.php?f=90

Let me guess...ATI owner.
 
Let me guess...ATI owner.

Both, 8800GT rig and a HD4870 rig. I'm just a graphics whore, if I can't run a game at near max settings with what I've got, than I'll just wait to play it until I have a rig that can crank up the goodies, and from all reports so far, it looks like I won't even bother trying it.

Why play a game if its not in all its full glory, plenty of other games to keep me occupied until then. It will just stay in my backlog for now.

Am I right about the poor coding, optimization and only using a single core crap?

There is no excuse to release (gamers) game today that does not take advantage of the multiple core CPU's.
 
Both, 8800GT rig and a HD4870 rig. I'm just a graphics whore, if I can't run a game at near max settings with what I've got, than I'll just wait to play it until I have a rig that can crank up the goodies, and from all reports so far, it looks like I won't even bother trying it.

I am a gamer, pretty dead graphics means nothing to me.

Why play a game if its not in all its full glory, plenty of other games to keep me occupied until then. It will just stay in my backlog for now.

That your hardware is not up for the task...well you figure it out.

Am I right about the poor coding, optimization and only using a single core crap?

There is no excuse to release (gamers) game today that does not take advantage of the multiple core CPU's.

You just descibed most games, must be hard disliking PhysX so much that you use it to bash most games...
 
Ok, I'm obviously speaking to an Nvidiot here. I said NOTHING about disliking PhysX, it was your "Let me guess...ATI owner." statement that I was attempting to straighten you out on.

No my rigs would not run the game very well at the setting I demand, but maybe if it used both cores of my processor, I'd stand a better chance.

You just descibed most games, must be hard disliking PhysX so much that you use it to bash most games...
- very weak.


What, was it this comment of mine that got you so hot and bothered? - "More PhysX goodies on the way to further drag the game down."

The point was, if they would spend more time cleaning up the game so more people could run it decently first, than worrying about adding in more PhysX effects. Priorities.
 
I am a gamer, pretty dead graphics means nothing to me.



That your hardware is not up for the task...well you figure it out.



You just descibed most games, must be hard disliking PhysX so much that you use it to bash most games...

Boy did you show your hand..he never said anything negitive. Sounds to me like he likes to play his games cranked, why does that bother you?

Its also funny how you say your a gamer who doesn't care about graphics but put down his rigs...you really shouldn't even be in this thread if graphics mean nothing. Turn it all the way down and have fun.
 
If you're on Windows 7 you could use the 4870 for graphics processing and the 8800 for PhysX. I think that you'd probably get near max performance but I'm not 100% sure about that.
 
Flying debris, shock waves, flipping cars, smokes and bent metals? Hell yeah, much more than curtains and dripping water. :p

The flying debris and flipping cars are trivial rigid body stuff. The shock wave is obviously not physics but a post processing effect. The smoke is also obviously a prebaked effect (since it dissipates quickly and does not interact with the environment).

It seems a lot of people don't understand why real physics simulations are intensive. Things like cloth, fluids and smoke require the simulation of the interaction of tens of thousands of particles. That's where the horsepower comes into play. Having said that, it's very possible for pre-baked effects to look better than simulations and they'll continue to be used where it makes sense (i.e where the constituent fluid dissipates immediately and doesn't interact with anything).
 
The flying debris and flipping cars are trivial rigid body stuff. The shock wave is obviously not physics but a post processing effect. The smoke is also obviously a prebaked effect (since it dissipates quickly and does not interact with the environment).

It seems a lot of people don't understand why real physics simulations are intensive. Things like cloth, fluids and smoke require the simulation of the interaction of tens of thousands of particles. That's where the horsepower comes into play. Having said that, it's very possible for pre-baked effects to look better than simulations and they'll continue to be used where it makes sense (i.e where the constituent fluid dissipates immediately and doesn't interact with anything).

Wait, cloth is simulated with particles? :confused:
 
The flying debris and flipping cars are trivial rigid body stuff. The shock wave is obviously not physics but a post processing effect. The smoke is also obviously a prebaked effect (since it dissipates quickly and does not interact with the environment).

It seems a lot of people don't understand why real physics simulations are intensive. Things like cloth, fluids and smoke require the simulation of the interaction of tens of thousands of particles. That's where the horsepower comes into play. Having said that, it's very possible for pre-baked effects to look better than simulations and they'll continue to be used where it makes sense (i.e where the constituent fluid dissipates immediately and doesn't interact with anything).

Intensive calculation doesn't mean that it will be an impressive effect. I'm sorry that I don't live in the Matrix and when I see curtains or dripping water in games, I don't see them as numbers and calculations, I just see them as what they are: just some curtains and dripping water. When I see flying debris, shock waves, flipping cars, smokes and bent metals, I also don't see them as simple codes using rigid bodies, I see them as immersive effects. If I want to see codes and calculations, I don't need to play a game with PhysX, I can just watch the movie "The Matrix" or I could just use MatLab.
 
Wait, cloth is simulated with particles? :confused:

Yep, how else would it be done? The cloth is simulated as a mesh of control points (aka particles) that react to each other according to the properties of the material and external forces (wind, gravity etc).

http://www.youtube.com/watch?v=iuWxg9K8Mdg&feature=channel_page

http://www.jrc313.com/processing/cloth/

Intensive calculation doesn't mean that it will be an impressive effect. I'm sorry that I don't live in the Matrix and when I see curtains or dripping water in games, I don't see them as numbers and calculations, I just see them as what they are: just some curtains and dripping water. When I see flying debris, shock waves, flipping cars, smokes and bent metals, I also don't see them as simple codes using rigid bodies, I see them as immersive effects.

No argument there. As I said above fakes are often more visually appealing than simulations that require more horsepower than is available today. But fakes have a big disadvantage - you can't interact with them.
 
I think games with lots of PhysX, can make good use of as much dedicated PhysX processor as you can give them...

The numbers say a 280 running in dedicated PhysX mode helps more than an 9800 does...

This tells me there is some heavy calculations being done on them...
If not, a 295 running with a 9800 operating in dedicated PhysX mode, should produce the same FPS as a system with a 295, and 280 in PhysX mode.

They do not. The extra speed the 280 has, does move your FPS higher. ;)
Feed the monster!

Cryostasis TechDemo:
So in summary at 1920x1200 res, this is what we have:

walterman --- Xeon 3350 @ 3.6GHz - Single 8800GTX
Average FPS= 22, Minimum FPS= 13

Roliath --- Q6600 @ 3.2GHz 2x 260's in SLI Mode
Average FPS= 31.6, Minimum FPS= 18.3, and Maximum FPS=75.3

Talonman ------------ Single 280 @ 756/1512/1269 --- Q6600 @ 3.81GHz
Average FPS= 35.9, Minimum FPS= 21.1, and Maximum FPS=109.3

Extreme Bench ------------ Single 285, no dedicated PhysX GPU, QX 9650 @ 4.20GHz
Average FPS= 36.9, Minimum FPS= 23.2

StretchMaK ------------ 2x 260's SLI Mode --- E4800 @ 3.6GHz
Average FPS= 38.3, Minimum FPS= 22.1, and Maximum FPS=96.7

Extreme Bench ------------ GTX 285 + GTX 285 for PhysX, QX 9650 @ 4.20GHz
Average FPS= 44.7, Minimum FPS= 26.3

Shamino ------------ Tri-280's, no dedicated PhysX GPU, i7 @ 2.8GHz
Average FPS= 46.0, Minimum FPS= 17.9, and Maximum FPS=109.2

slaWter ---i7 965 Extreme @ 3.84GHz - 2x 280's in SLI Mode
Average FPS= 50.0, Minimum FPS= 29.8, and Maximum FPS=150.4

Extreme Bench ------------ GTX 285 Dual SLI, no dedicated PhysX GPU, QX 9650 @ 4.20GHz
Average FPS= 51.5, Minimum FPS= 29.9

Talonman ------------ Single 295 @ 648/1512/1188 --- Q6600 @ 3.81GHz
Average FPS= 53.7, Minimum FPS= 30.7, and Maximum FPS=127.2

Extreme Bench ------------ GTX 285 Tri SLI, no dedicated PhysX GPU, QX 9650 @ 4.20GHz
Average FPS= 58.0, Minimum FPS= 31.0

Shamino ------------ Tri-280's, 9800GT dedicated PhysX GPU, i7 @ 2.8GHz
Average FPS= 58.6, Minimum FPS= 24.9, and Maximum FPS=132.1

Dentlord with 3x GTX 280 @ 685/1475/1220...air...none...qx9650 @ 3.6GHz...water produced:
59.6 Average FPS.

Dentlord with 3x GTX 280 @ 685/1475/1220...air...none...qx9650 @ 4.2GHz...water produced:
62.5 Average FPS.

Krog with 3x GTX 280 @ 720/1440/1300 ...water...none... i7 920 @ 3.9GHz...water produced:
64.4 Average FPS.

freakysqeeky: System was at: Q9550 @ 4.0GHz
(1) 295 set to C=660, SH=1423, and M=1210. (Processing Graphics)
(1) 9800GTX+ set to C=790, SH=1977, and M=1160. (Dedicated PhysX)
Average FPS= 65.5, Minimum FPS= 32.2, and Maximum FPS=135.4

Extreme Bench ------------ GTX 285 Dual SLI + GTX 285 for PhysX, QX 9650 @ 4.20GHz
Average FPS= 75.1, Minimum FPS= 32.4

Talonman: System was at: Q6600@ 3.81GHz
(1) 295 set to C=702, SH=1548, and M=1188. (Processing Graphics)
(1) 280 set to C=756, SH=1512, and M=1296. (Dedicated PhysX)
Average FPS= 78.5, Minimum FPS= 36.2, and Maximum FPS=140.4
 
Last edited:
Back
Top