80 boulders  =  0.8 fps when all Rbs touch the ground   
 
100 boulders = no movement visible...
 
vista64 / 2GB Ram / EAH4670-1GB / P4 3,2GHz
(I think I could double the rate with XP 32bit on my old CPU)
 

80 boulders about 40 fps on my gear in SW mode.
I would thought it suppose to be better, is it physics calculation or internal messaging ?

I want to know what Weapon121 gets on these test.



EDIT: solid 60 fps when RB's are only linked to terrain and not to eachother.

EDIT2: solid 60 fps on 100 RB too when only linked to terrain and gforce and not to eachother
.. this would imply 100 RB = 200 messages compared to 10000 messages when all are interlinked.
.. so I would think rads internal messaging is the bottleneck, not physx.

shadmar, in your tests RBs linked to RBs Vs. Rbs linked to terrain only,are all the RBs in contact with each other, like the image posted by mgear?? or just hitting the ground here and there?

Im asking because, being all in contact with each other, the problem could be more related to the physics thing (too), like that problem we saw once when we couldnt stack more than 4 or 5 rows or bricks without tunning down physics simulation accuracy... AND, in suchcase, it wouldnt be really a big "ooohhh" that them work just fine interacting only with the terrain.

@PsychoWeasel9
 

Quote

You can find 3D Rad v7 in the Start menu. Running 3D Rad by clicking the desktop icon instead will start the version you last launched from the Start menu.

 

Yes, I know, but the first time I clicked the v7 icon it said something like "can't find .exe" and now it only loads v6
I know how to use my own pc, thank you

I know how to use my own pc, thank you

sorry for trying to help, maybe I read it wrong 

opened the same project in v6.49 and:

40 boulders it goes ~58fps
50 boulders, crash..
60 boulders, crash..

Thanx for setting up the test project mgear, and thank you everyone for the feedback!

80 boulders made of 3 spheres each, tested for collision against each other and against a polygonal model (terrain) made of thousands of triangle is quite a massive task to perform 60 times a second 

So, if you get more than 30fps in software mode, on a midrange PC, we can safely conclude that switching to PhysX was a giant leap forward from 3D Rad v6 

There should be no big overhead due to internal 3drad messages in this case because RigidBody objects don't actually exchange messages with other RigidBody objects, run-time.

So, the performance we get should be basically the same that we would get with any other 3d engine which uses PhysX, when performing exactly the same task (exactly the same models, same number of polygons for terrain etc).

Keep in mind that actual performance depends on many factors, including how exactly the rigid-bodies involved collide (all at the same time vs over a longer period, or all around the same location or against the same terrain model versus far apart one another or against different terrain 'sectors' etc).

Also note that PhysX skips collision detection for rigid-bodies that are still (eg resting on ground). You should notice that frame rate gets better, in the mgear demo, after a while, when all objects are still. However note that sleep mode for rigidbody that hit the same terrain rigid-body may only trigger when ALL other rigid-body hitting the same terrain are still.