PDA

View Full Version : Nvidia releases PhysX source code on GitHub



erikals
03-05-2015, 07:58 AM
hmm... http://erikalstad.com/backup/misc.php_files/smile.gif
http://www.cgchannel.com/2015/03/nvidia-releases-physx-source-code-on-github



https://www.youtube.com/watch?v=O04ErnJ8USY

vonpietro
03-05-2015, 05:23 PM
amazing

this is not that easy to do in bullet. I keep running into the dreaded - touch a small spot and the entire thing explodes problem. Even with constraints - its a pain.

Surrealist.
03-06-2015, 02:14 AM
Fun stuff.

ianr
03-06-2015, 04:39 AM
All being well we should Bullet 3 sometime this year.?
i hope erwin sorts out the small world size addition.
But back to this Nvidia Gift, we could sure use the fluids
& particles inside our package asap.
A pity the model didn"t get Drums to make up the stone
columns,these toppling away would have sold the shot.

Having said that, if this on Github & in the public domain,
maybe a gifted Waver Programmer could take a peek at
a LW engine. Anyone interested,we could do a subscription
to expenses if a poll could create a stir this time a round
The community could push LW along with this, surely this was
the notional driver of our 'famous plug-ins' in the past?

ianr
03-10-2015, 01:36 PM
Hello Out There, Now thats a Deafing Response Wavers

erikals
03-10-2015, 01:50 PM
the fluids ain't that good, but the RBD sure is fast, it'd be fun to add, the only person i can see adding this at the moment would be Hurley, but he is working on advanced placement for the time being...

NT is probably busy with other stuff, so, might take a while, never the less, cool progress... \ : )

03-10-2015, 05:36 PM
IBounce is what I believe is the present name of Hurley's tool. The example explains a bit of why he has held off: that's a LOT to include in an app for LW. Well, most apps. Hopefully he has been R&Ding this in the background and will share some of his knowledge with us in a beta or ready-to-go app soon.
That does look sexy.

creacon
03-11-2015, 05:07 AM
Or me ;-)

ianr
03-11-2015, 06:17 AM
Well, we have a healthy support of Renderers

a few more (+1) Physic Sim engines would help.

I too, look forward to H's ' i-bounce' launch.

But for now, I cite the speed increase in

certain areas that Syflex Engine is

'Wiping the Floor' with Bullet,

as Ryan's Excellent Tutorial Vid

on Liberty3d can attest too.

http://www.liberty3d.com/2015/02/the-syflex-workflow-guide/

creacon
03-11-2015, 07:02 AM
My tests with PhysX:

Rigid bodies are about 10 times as fast as Bullet (and RB's are CPU only in PhysX too!)
Cloth is realtime where bullet takes 60 seconds, so 100 X or more faster.
Particles/liquids: not available in bullet
Soft bodies: not available in PhysX (since version 3.x)

Still waiting for Flex.

creacon

- - - Updated - - -

I'll take a look, only problem is that my eyes hurt after I've seen one of his vids ;-)


Well, we have a healthy support of Renderers

a few more (+1) Physic Sim engines would help.

I too, look forward to H's ' i-bounce' launch.

But for now, I cite the speed increase in

certain areas that Syflex Engine is

'Wiping the Floor' with Bullet,

as Ryan's Excellent Tutorial Vid

on Liberty3d can attest too.

http://www.liberty3d.com/2015/02/the-syflex-workflow-guide/

creacon
03-11-2015, 07:03 AM
double post

Danner
03-11-2015, 07:12 AM
I'ts not iBounce, it's now called Advanced Placement. http://hurleyworks.com/advanced-placement/

creacon
03-11-2015, 08:14 AM
That's the "sidestep", Steve was working on a full physics implementation originally.
Like I am doing.

creacon


I'ts not iBounce, it's now called Advanced Placement. http://hurleyworks.com/advanced-placement/

03-11-2015, 10:18 AM
As stated, AP came about because of IB.

So, Creacon, whatchya got up your sleeve(s)?
You should beat others to the table to make the big bucks. Curious as to what you may have.
Thanks for keeping us 'Wavers at the front. Or close to it...

creacon
03-11-2015, 05:22 PM
I have nice things cooking ;-)
If I was interested in the big bucks I wouldn't even consider LW. LOL


As stated, AP came about because of IB.

So, Creacon, whatchya got up your sleeve(s)?
You should beat others to the table to make the big bucks. Curious as to what you may have.
Thanks for keeping us 'Wavers at the front. Or close to it...

03-12-2015, 03:52 AM
True. Too true.

Glad to hear something is cooking.

ianr
03-12-2015, 05:23 AM
Well said Creadon,,

LightWave's charisma is its

devoted support by its afisonado's & creative

additions by people who can & do write in house

tools for cgi houses edge.



Best of luck for your physx project,do keep us all updated.

lightscape
03-12-2015, 06:03 AM
I wondered why newtek focused solely on bullet. Physx was realtime and faster.
Probably miscalculated on this one.

MSherak
03-12-2015, 01:15 PM
I wondered why newtek focused solely on bullet. Physx was realtime and faster.
Probably miscalculated on this one.

Bullet is OpenSource and works on everything (cpu based). PhysX was pay to use and limited depending on GPU/CPU specialization.

-M

kopperdrake
03-12-2015, 03:13 PM
This is nice looking, but notice that none of the fractured elements seem to have a secondary fracture when they collide with, say, the ground. The explosions are always started at the top of the building, then move down. If the supporting structure beneath a roof is blown away then the roof will only fall if there is nothing holding it up, but it will fall as one object, and not break up on contact. You can see a column, for example, at 2.41 on the left, dangling in mid-air after its base is blown away. Does anyone know if this is a limitation of this technology?

jwiede
03-12-2015, 05:55 PM
Bullet is OpenSource and works on everything (cpu based). PhysX was pay to use and limited depending on GPU/CPU specialization.

Right, probably didn't hurt that AMD's involved in OpenCL Bullet for GPU space. The Bullet pkg has undergone substantially broader in-3D-package integration compared to PhysX, as well, so more dev experience with the kinds of problems that arise, and how to fix or work around them.

MSherak
03-12-2015, 10:41 PM
Right, probably didn't hurt that AMD's involved in OpenCL Bullet for GPU space. The Bullet pkg has undergone substantially broader in-3D-package integration compared to PhysX, as well, so more dev experience with the kinds of problems that arise, and how to fix or work around them.

And with Bullet 3.0 getting multithreading it will be even better..

creacon
03-13-2015, 12:53 AM
Because I only have an NVidia card I am not able to test if PhysX really won't work without it. But there is one thing that I was able to test and that is you can force PhysX to use the CPU only. If your card runs out of memory that fallback is even automatic.

Flex seems to be GPU only.

creacon




Bullet is OpenSource and works on everything (cpu based). PhysX was pay to use and limited depending on GPU/CPU specialization.

-M

creacon
03-13-2015, 01:01 AM
You're right as far as DCC packages are concerned, but PhysX is widely used in games, the demands are different, but do you really think that there are more developers working on 3D packages than there are working on games?



Right, probably didn't hurt that AMD's involved in OpenCL Bullet for GPU space. The Bullet pkg has undergone substantially broader in-3D-package integration compared to PhysX, as well, so more dev experience with the kinds of problems that arise, and how to fix or work around them.

ianr
03-13-2015, 05:34 AM
Like I said Copperdrake on page1 of this
Most Greco Columns were made of Stone
drums mounted on one another,then the
flutes where chiselled after the erecting.
Well they where when I lived in Greece,
& toured extensively. So I surmise that
the sim was built by a game modeller
whose never seen the wonders of Hellas.
Look its just lazy modelling & implementation.

Further to this look at the speeds Ryan is getting
with his Syflex trials in his Vid Tute, like up to
50 times faster on Cloth simsolver. Now thats
fast enough for Jeremy Clarkson to go :Whoar!

jwiede
03-13-2015, 05:35 AM
You're right as far as DCC packages are concerned, but PhysX is widely used in games, the demands are different, but do you really think that there are more developers working on 3D packages than there are working on games?

The difference in usage is tremendous -- particularly in efficiently translating DCC pkg UX elements and parameters into efficiently-constructed simulations on the fly. DCC pkgs animation systems also tend to be much more varied and complex than those driving games. Games are basically a "fixed" simulation proposition (or rather, sets of them), which makes the task more about efficiently feeding fixed templates for simulation. DCC packages' integration as a dev task represents a much broader potential domain of inputs and execution paths, with the added complexity of then integrating the simulation output into the DCC pkgs' more complex animation systems, baking systems, etc. once computed (and on top of all that, also offering scripting and API access to dynamics within DCC pkgs).

The difference is akin to efficiently building a house versus building a user-friendly machine that itself can efficiently build houses.

lightscape
03-13-2015, 07:38 AM
Bullet is OpenSource and works on everything (cpu based). PhysX was pay to use and limited depending on GPU/CPU specialization.

-M

Ageia physx and all its iterations can be used on cpu only since before.
If newtek integrated physx before then we could have gotten a usable fluid simulator since lightwave 9.
Dstorm had a decent fluid sim that has been neglected for some time, too bad.

creacon
03-13-2015, 08:58 AM
The difference in usage may be big, but the difference in technology isn't. If you add an object to a simulation everything has to be recomputed anyway. Bullet in LW does this all the time.
The complexity of the host app has nothing to do with it, you feed a position into the sim, the sim does its work and gives back the new position/rotation etc. this has to be applied to the original object. In a game environment this is the same, only you will probably feed it directly into the renderer.
In my case a rigid needs a motion plugin and a custom object (for display). Because it is a motion plugin you could use LW's motion baking on top of that.
A cloth needs a displacement plugin and a custom object (for display). You can use LW's mdd export on top of that.
A particles system only has a custom object. And everything is handled by a master.


The difference in usage is tremendous -- particularly in efficiently translating DCC pkg UX elements and parameters into efficiently-constructed simulations on the fly. DCC pkgs animation systems also tend to be much more varied and complex than those driving games. Games are basically a "fixed" simulation proposition (or rather, sets of them), which makes the task more about efficiently feeding fixed templates for simulation. DCC packages' integration as a dev task represents a much broader potential domain of inputs and execution paths, with the added complexity of then integrating the simulation output into the DCC pkgs' more complex animation systems, baking systems, etc. once computed (and on top of all that, also offering scripting and API access to dynamics within DCC pkgs).

The difference is akin to efficiently building a house versus building a user-friendly machine that itself can efficiently build houses.

MSherak
03-13-2015, 09:03 AM
Ageia physx and all its iterations can be used on cpu only since before.
If newtek integrated physx before then we could have gotten a usable fluid simulator since lightwave 9.
Dstorm had a decent fluid sim that has been neglected for some time, too bad.

I think the DStorm plugin required an Nvidia card.

creacon
03-13-2015, 09:03 AM
DStorm made a liquid simulator when PhysX wasn't ready for it. It was buggy and the number of particles was limited to 32768 (PhysX fault, not DStorm's). The current version of PhysX supports about 2M particles on my 2Gb card and falls back to cpu afterwards, slower but still usable. I tested 10M and it still holds up. (CPU).

creacon


Ageia physx and all its iterations can be used on cpu only since before.
If newtek integrated physx before then we could have gotten a usable fluid simulator since lightwave 9.
Dstorm had a decent fluid sim that has been neglected for some time, too bad.

erikals
03-13-2015, 12:54 PM
limited to 32768 (PhysX fault, not DStorm's)
ah, thank you, always thought it was my video card's fault... http://erikalstad.com/backup/misc.php_files/smile.gif

lightscape
03-13-2015, 09:45 PM
I think the DStorm plugin required an Nvidia card.

Nope it worked with non-nvidia equipped laptop. Slower but worked fine.