Hypothetical ATI + Nvidia + PhysX

Discussion in 'Video Cards & Monitors' started by T-O-D, Jul 7, 2008.

  1. T-O-D

    T-O-D Member

    Joined:
    Oct 23, 2003
    Messages:
    1,317
    Location:
    Brisvegas!
    Evening all :)

    Just wondering off into fairy land as i sometimes do, and thinking about the developments with CUDA and PhysX.

    Is it actually possible to have both ATI and Nvidia drivers installed at once on the one machine?

    My theory is this:
    A lot of people are currently purchasing the 4800 cards from ATI. This kinda leaves them in the dark when it comes to PhysX. Wouldnt they be able to pick up some cheaper Nvidia card (not sure what the minimum spec for PhysX is) or even use an existing Nvidia card that they have, and install the drivers, so that they can run the 3D from the ATI card (Bang for buck at the moment it seems), but also run PhysX from their Nvidia card?

    Would be really great for those that are upgrading from the 8800's to a 4870, but also want PhysX!

    If you cant actually install both sets of drivers at once, Why not? (and please dont say "cause you cant" :p )
     
  2. BlackSniper87

    BlackSniper87 Member

    Joined:
    Jun 1, 2008
    Messages:
    1,848
    Location:
    Hamilton, Brisbane
    from what i understand vista does not let you for unknown reasons not sure bout xp though.

    Great idea though
     
  3. hello_goodbye

    hello_goodbye New Member

    Joined:
    May 11, 2008
    Messages:
    721
  4. stevo4

    stevo4 Member

    Joined:
    Jun 27, 2001
    Messages:
    10,629
    Location:
    Melbourne
    With nVidia's PhysX drivers for running PhysX, incorporated into the driver for their video cards, atm it doesnt seem possible,
    as you might have issues with getting the nVidia GPU just to act as PPU.

    I cant see atm why, what you are suggesting, should not be possible, but I think it will be upto nVidia to put out drivers that would enable this.
     
  5. Silenius

    Silenius Member

    Joined:
    May 25, 2002
    Messages:
    817
    There's no reason why nVidia would want it to work - in fact, they'd probably find some way of disabling it in software or hardware - because they want to lock people into using just nVidia cards... rather than their competitors'.. :)
     
  6. stevo4

    stevo4 Member

    Joined:
    Jun 27, 2001
    Messages:
    10,629
    Location:
    Melbourne
    I dont think nVidia would care if it sells their GPU's.
    Better one sale than none.

    Try reading some of the other threads on this forum about PhysX.

    The guys at ngohq.com, that reported that they have modified ati drivers for a 3850 to incorporate PhysX, have reported that they are getting help from nVidia now.
     
  7. mumblyjoe

    mumblyjoe Member

    Joined:
    Jul 18, 2002
    Messages:
    448
    Location:
    Perth, WA
    That's correct. Vista only allows one display drivers to be installed at any one time
     
  8. Reanimated35

    Reanimated35 Member

    Joined:
    Sep 17, 2007
    Messages:
    3,624
    Location:
    Brisbane
    cant u still buy seperate physx cards? or do they use nvidia drivers?
     
  9. mumblyjoe

    mumblyjoe Member

    Joined:
    Jul 18, 2002
    Messages:
    448
    Location:
    Perth, WA
    You can but I've read a thread somewhere that shows the PhysX card doesn't give anywhere near the gains in performance as an nVidia card does
     
  10. me2

    me2 Member

    Joined:
    Jul 18, 2007
    Messages:
    2,743
    Location:
    directly in front of you
    yes, you could write a driver that offloads the physx or even havok calculations to an nvidia gpu and plug it into even a 1x pci-e slot

    no need to write or try to load two two display drivers

    or you could do the opposite, write a driver for an ati gpu to do the physx or havok processing and use an nvidia gpu as the display card

    both gpu's are programmable as both have an interpreter like language
     
  11. stevo4

    stevo4 Member

    Joined:
    Jun 27, 2001
    Messages:
    10,629
    Location:
    Melbourne
    ^^^ I take it you are being strictly hypothetical.

    Plug it into 1x PCIe slot?
    Are there any programmable GPU’s on the market that will plug into a 1x PCIe slot?
    Or likely to be?


    Is anyone likely to do this?
    Unless ATI support this, it is most likely to break when a driver update is released.
     
  12. bobbavet

    bobbavet Member

    Joined:
    Apr 13, 2006
    Messages:
    1,635
    Location:
    East Maitland
    Yes but there is no saying Nvidi will not release a standalone card base on a 9 series that acts just like the old Ageia cards did.

    It would be a waste of Ageia purchase and a considerable market if they didn't.

    I think ATI are going to get side tracked and waste time and money with Havok(intel), when they should be focusing card developement. Im sure intel see this as a way of distracting the AMD machine. Conspiracy? :shock: :lol:
    ATI should allow physx on their cards, support people working with the CUDA sdk and denying Nvidia of standalone PhysX card sale.
     
    Last edited: Jul 8, 2008
  13. stevo4

    stevo4 Member

    Joined:
    Jun 27, 2001
    Messages:
    10,629
    Location:
    Melbourne
    Possible, but that would get the same criticism that Ageia got with the PPU, in that too many would complain that would be an extra cost.

    As the cost of GPU’s go down, its probably not worth making a dedicated card for PhysX when a cheap GPU could do the same.


    I would say that ATI went the Havok/Intel way, as that matches the way they are going, in that they intend to use their GPU as a vector coprocessor, closely tied to a CPU.


    Problem is that ATI is in no rush to implement or support CUDA in their drivers for their GPU’s.
    Something that will cost them, even if CUDA is free, and they probably wont, unless PhysX on CUDA takes off.
     
  14. bobbavet

    bobbavet Member

    Joined:
    Apr 13, 2006
    Messages:
    1,635
    Location:
    East Maitland
    I reckon the main reason for the high ageia cost was that it was a small start up venture. They had to pay premium to get the gear built on production lines used by major players.

    True it would be better to have it all on the one card as Nvidia but if you want to take a hold of dedicated ATI card owners and boost a market for physx a standalone should be considered.

    Games like to have the "NVIDIA the way it's meant to be played" splash screens and box logos. As they did with Sli support, I think Nvidia will require PhysX support in games before allowing it to be used.
     
  15. stevo4

    stevo4 Member

    Joined:
    Jun 27, 2001
    Messages:
    10,629
    Location:
    Melbourne
    I am not advocating having it run on one GPU.
    The same as what you are suggesting, could be acheived through drivers.
    At a much less of a financial risk, than manufacturing a card just to perform PhysX.
    I would expect it to be cheaper for nVidia to offer an existing GPU to do this, at a low price.

    PhysX is a lot further down the track in supporting hw acceleration and being used in games, than any alternative atm.
    I cant see nVidia doing anything atm that would hinder PhysX being used.
    The use of PhysX is free.
    As I pointed out above, nVidia are prepared to help out the guys at ngohq.com in getting PhysX running on an ATI card.
    More so than AMD.
     
  16. me2

    me2 Member

    Joined:
    Jul 18, 2007
    Messages:
    2,743
    Location:
    directly in front of you
    although its a hypothetic question, the info here is not hypothetical, except so far as the assumption of physics engine owners playing ball with other manufacturers

    all pci-e cards, be it a graphics card, sound card, a ppu card or a network card are capable of being slotted into a pci-e 1x slot and having a driver written for them to interact with the system, though they are not all as programmable to the extent that modern gpu's are, it is simply a matter of bandwidth requirments, e,g..given the ageia ppu was originally a pci card, that shouldn't be an issue

    you would be simply using the second display card as a fixed function physics calculator, not a display card

    the driver would be providing it with the ability to do dedicated physics type calculations..(note, that doesn't simply mean physx or havok..it can also mean specific calculations that all physics engines require)

    display driver updates wouldn't bork it any more than they would bork an original physx card, as it would be acting as a coprocessor to the cpu not just specifically a physx ppu or a hardware optimised havok engine (not that you couldn't program it to favour one or the other)

    though if you were to get elaborate, and try to feed the physics data directly to the display cards shaders from the secondary display card(the one acting as a ppu) as in the case of multi gpu card solutions..e.g. sli or crossfire, then you might have issues with display driver updates...but only if you installed them or nvidia/ati/intel decided to stop it through thier programming interface or physics engine license

    in essence, a gpu can be programmed to do any function a cpu can (and vice versa)...but the architecture of the cpu or the gpu respectively is more efficent at certain types of functions ...e.g. the functions used in military combat flight sims for the aerospace industry are a direct ancestor to the modern gpu's adaptation to simulteneous physics/display rendering

    it is of no coincidence that by replacing some of the fixed functions, with programmable shaders...that we can use gpu's for stream computing...of which physics calculations are merely a subset or for that matter database processing etc..

    as far as stream processing is concerned, if you configure the gpu to basically act as an x 86 cpu, then there is plenty of code that doesnt need to be completely rewritten, though having code specifically written will have speed advantages, just as having the physics code directly applied to the pixel shaders from within the gpu has its benefits, though it might also be a hinderance to server/client side physics in a multiuser environment, at least until stream computing matures, but that is not taking into account cpu architectures moving towards increased vector capabilities (which has been going on for a long time as well)

    just as thier are people geting physx working on ati cards, ther are also people working at getting havok working with nvidia cards or for that matter getting display cards to act as fixed function physics processors (both dependant and independant of any specific physics engine)

    [edit] both physx and havok engines are free to use, also, cuda is more advanced than alternatives when it comes to stream processing, but thats only because nvidia doesnt have an x86 license, the gpu is still slower in many of the required functions than a cpu
     
    Last edited: Jul 8, 2008

Share This Page

Advertisement: