AMD finally gets back to Physics on the GPU

Discussion in 'PC Gaming' started by almighty, Sep 23, 2009.

  1. Albuquerque

    Albuquerque Red-headed step child
    Veteran

    Joined:
    Jun 17, 2004
    Messages:
    4,309
    Likes Received:
    1,107
    Location:
    35.1415,-90.056
    While I do believe you, I'm not finding anything specific on that website that tells us the difference. "Not supported on PPU" really only says that it doesn't work; why doesn't it work? What changed?
     
  2. neliz

    neliz GIGABYTE Man
    Veteran

    Joined:
    Mar 30, 2005
    Messages:
    4,904
    Likes Received:
    23
    Location:
    In the know
    I have to find the interview by the Ageia engineer where he explained how and what they had to do to get PhysX to run on a GPU. He also said which thing would never see GPU acceleration as well. I posted it somewhere in one of the many physx discussions here as well.
     
    #142 neliz, Mar 10, 2010
    Last edited by a moderator: Mar 10, 2010
  3. Argoon

    Newcomer

    Joined:
    Mar 10, 2010
    Messages:
    40
    Likes Received:
    0
    Ridge body physics, unlike the Ageia PPU the GPU's can't accelerate ridge body physics, the new Nvidia demo for Fermi also runs the ridge body's physics on the CPU so it seams the new Fermi "architecture" can't accelerate them also. Without this capability GPU's will never be able to accelerate gameplay changing physics they will only accelerate eye candy physics like particles, cloth, smoke, liquids, etc, gameplay physics are all physics that have a direct impact on the gameplay, like the HL2 gravity gun, physical objects that can kill or hurt you, anything that needs to communicate with the CPU.
     
  4. Albuquerque

    Albuquerque Red-headed step child
    Veteran

    Joined:
    Jun 17, 2004
    Messages:
    4,309
    Likes Received:
    1,107
    Location:
    35.1415,-90.056
    And out of equal fairness to Neliz, I also believe you, but I'd love to see a link that confirms this stance :)

    Oh, and welcome to the forum!
     
  5. Argoon

    Newcomer

    Joined:
    Mar 10, 2010
    Messages:
    40
    Likes Received:
    0
  6. Silent_Buddha

    Legend

    Joined:
    Mar 13, 2007
    Messages:
    19,426
    Likes Received:
    10,320
    It appears that most of the well funded AAA titles choose to go Havok rather than PhysX. Most but not all obviously. :) PhysX seems REALLY common with less well funded projects (obviously) since it's free and if you toe the Nvidia line, they'll throw in additional help and marketing dollars.

    Regards,
    SB
     
  7. Silent_Buddha

    Legend

    Joined:
    Mar 13, 2007
    Messages:
    19,426
    Likes Received:
    10,320
    Interesting then since Pixelux (DMM2 + Bullet) can now do GPU accelerated rigid body physics through OpenCL/Direct Compute. So in some ways it's already got some inherent advantages besides vendor agnostic GPU acceleration of physics.

    Although, as with Albuquerque, it'd be nice to see it stated somewhere also.

    Regards,
    SB
     
    #147 Silent_Buddha, Mar 10, 2010
    Last edited by a moderator: Mar 11, 2010
  8. Argoon

    Newcomer

    Joined:
    Mar 10, 2010
    Messages:
    40
    Likes Received:
    0
    I already gave a link for the Nvidia PhysX forum were a Dev corroborates what i said, but the reply needs to be reviewed by a moderator first because of the link.
     
  9. neliz

    neliz GIGABYTE Man
    Veteran

    Joined:
    Mar 30, 2005
    Messages:
    4,904
    Likes Received:
    23
    Location:
    In the know
    it took some googlewizardry!

    http://www.geeks3d.com/20080614/physx-on-geforce-using-cuda/

    I have to find his complete interview for all the details.

    Rigid body PhysX on all platforms (Cell, PPC, x86, ARM etc.) is done on the CPU.
     
    #149 neliz, Mar 10, 2010
    Last edited by a moderator: Mar 10, 2010
  10. Albuquerque

    Albuquerque Red-headed step child
    Veteran

    Joined:
    Jun 17, 2004
    Messages:
    4,309
    Likes Received:
    1,107
    Location:
    35.1415,-90.056
    Nice find :)
     
  11. Argoon

    Newcomer

    Joined:
    Mar 10, 2010
    Messages:
    40
    Likes Received:
    0
    Didn't saw anything about it on the Bullet forums, what i saw was a old thread about making bullet work on the GPU and they were talking in making narrowphase physics (eye candy physics) on the GPU and broadphase physics (gameplay physics) on the CPU for the time being but they were certainly thinking in making all work on the GPU.

    If they were able to make it work them they can be very proud of their physics API because Nvidia PhysX still lacks on that feature.
     
  12. swaaye

    swaaye Entirely Suboptimal
    Legend

    Joined:
    Mar 15, 2003
    Messages:
    9,045
    Likes Received:
    1,119
    Location:
    WI, USA
    Some Physx effects are interactive to limited degrees.

    The only one I'm sure that interacts with the player is Mirror's Edge's ground fog. You can walk through it and it wafts around. It was interesting but it was quite demanding for my 8800GTX.

    The problem, aside from being demanding, is that the effects are all made up of little particle blobs so I wouldn't say that it adds realism to a realistic-looking game. :D
     
    #152 swaaye, Mar 11, 2010
    Last edited by a moderator: Mar 11, 2010
  13. Albuquerque

    Albuquerque Red-headed step child
    Veteran

    Joined:
    Jun 17, 2004
    Messages:
    4,309
    Likes Received:
    1,107
    Location:
    35.1415,-90.056
    Well, glass breaking is similarly interactive, but in the grand scheme of things, those broken shards don't really "do" anything else.

    Rigid body physics would come into play with, say, destroying a building piece-by-piece, and then using those pieces of rubble as a shield to hide from incoming fire or something. Basically, something that represents an entity in the game rather than a simple rendered effect.
     
  14. Argoon

    Newcomer

    Joined:
    Mar 10, 2010
    Messages:
    40
    Likes Received:
    0
    Being interactive is not the same as being gameplay changing, i personally have a understanding that gameplay physics are all physics that can directly influence the game your character and NPC's well being, for example, any physics that can kill, change your gameplay path or style, interactive fog, cloth, particles (ex. glass shards) don't have any of this criteria.
     
  15. aaronspink

    Veteran

    Joined:
    Jun 20, 2003
    Messages:
    2,641
    Likes Received:
    64
    Like all other GPU physics, its one way: your action effects it. It doesn't however affect you nor anything else in the environment. Its effectively the same as AF/AA. A little better eye candy with generally a very high cost factor.

    Which is the problem with most "physics" in games. Really we haven't improved at all from the HL2 gravity gun.
     
  16. aaronspink

    Veteran

    Joined:
    Jun 20, 2003
    Messages:
    2,641
    Likes Received:
    64
    Yep, they should be able to stick into people within their flight path. Nothing more broken then intricate animation of glass that harmlessly pass through objects and npcs.
     
  17. Silent_Buddha

    Legend

    Joined:
    Mar 13, 2007
    Messages:
    19,426
    Likes Received:
    10,320
    Aaah, gotcha. I must have misread that somewhere. Probably assuming that all OCL stuff would automatically use the GPU. When OCL can use either GPU or CPU depending on need and applicabiltiy.

    Regards,
    SB
     
  18. neliz

    neliz GIGABYTE Man
    Veteran

    Joined:
    Mar 30, 2005
    Messages:
    4,904
    Likes Received:
    23
    Location:
    In the know
    That will be an interesting dilemma in the future, as right now for AMD has both. How does an application decide which to use?
     
  19. Silent_Buddha

    Legend

    Joined:
    Mar 13, 2007
    Messages:
    19,426
    Likes Received:
    10,320
    The programmer can explicitly state which device to run on, or for it to run on all devices. As well I believe they can specify what % of work is done on each device.

    Was a lot of interesting information/discoveries in this thread here.

    http://forum.beyond3d.com/showthread.php?t=55913

    It appears that also unless explicitly told not to do so, OCL code when run on CPU will automatically use all available cores. This should come in handy for CPU based physics running on OCL. Programmer can also limit it to certain cores.

    Regards,
    SB
     
  20. Grall

    Grall Invisible Member
    Legend

    Joined:
    Apr 14, 2002
    Messages:
    10,801
    Likes Received:
    2,176
    Location:
    La-la land
    This is what IMO makes GPU physics worthless.

    Until there's proper feedback both ways between the physics engine and the main game logic, APIs like PhysX will be WORTHLESS.

    Nvidia's digging a hole for itself with its proprietary crap, hoping to play the part of 3dfx in the 1990s/early 2000s, which is both ironic and stupid. They really should know better from the way they beat 3dfx all those years ago now.

    Proprietary shit only works up to a point, and then you've essentially painted yourself into a corner by investing all your efforts into an obsolete proprietary solution when that effort could instead have been better spent on the winning, universal solution.
     
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...