Any Chance for Physx to OpenCL wrapper?

H-street

Supreme [H]ardness
Joined
Feb 25, 2005
Messages
4,538
I was thinking yesterday that back in the 3dfx days.. the Voodoo Cards only supported openGl via the MiniGL wrappers.. (converted the OpenGL calls to 3dfx Glide Calls)

got me thinking, why can't something like this for PhysX? a PhsyxMiniGL library that converts the PhysX calls to OpenCL or some other open standard?

i know that PhysX is Nvidia proprietary but it was fun to think about anyways..
 
A while back, there was a person who had PhysX running on ATI cards.. and Nvidia promptly shut him up.

It is possible and I have been entertaining the idea for a while.
 
got me thinking, why can't something like this for PhysX? a PhsyxMiniGL library that converts the PhysX calls to OpenCL or some other open standard?

i know that PhysX is Nvidia proprietary but it was fun to think about anyways..

There's no reason why it wouldn't be possible, technically. The problem is Nvidia would hever allow that, because then AMD would get the same benefits Nvidia does and so they'd lose their ability to differentiate their GPUs... if thats differentiation at all, that's another matter (considering virtually no relevant games use Physx and most have great results with now Intel-owned Havok).
 
Actually a lot of newer Games use PhysX like Alice2, Batman2, BorderLands2, and the brand new this week Metro2:LastLight. The latter is the most hardcore-requirement game to come out, next to BF3

And the older: Cryostasis, Dark Void, Darkest of Days, Mafia2. Metro2033, MirrorsEdge, Adv.Warfighter2, UT3

As long as you have an old Geforce as a secondary dedicated PPU it will run fine, just has to be a 8000+ with at least 256MB. I'm pretty sure Havok hasn't been used in games for years.

Hitman5 and BulletStorm use CPU PhysX and Hawken/PlanetSide2/WarFrame all use PhysX3(only v2 can be hybridized, btw)

cyclone3d: Regen, the owner of CLAIMED to have something like that, but 200 beta testers never got it, so no one knows if it ever existed. He even claimed NVidia helped him, in order to compete with Havok. But even though he promised an update he NEVER posted again about it and the thread dried up in 2008.

I suspect it was fake, a ploy to get members and hits, which it did, and then when hybrid became possible in 2010, it conveniently became the download HQ for it, since it was already the foremost community about the subject. Also after that patch stopped working in 2011, the new "Hybridiz" patch also made that forum its home in 2012.
 
Last edited:
OP: the job is much more complex than an OpenCL wrapper. GPU PhysX is implemented in CUDA, which compiles down to a very low level API used by the driver, via cudartxx_xx_xx.dll. It would be a very complex undertaking to take the compiled code backwards, patch it for the differences between CUDA and OpenCL, then recompile it back to a form used by the OpenCL runtime. IOW, it ain't happening.

A while back, there was a person who lied about having PhysX running on ATI cards.. and Nvidia promptly did nothing.
ftfy

Regeneration is just an attention whore.
 
If you want decent physics support for AMD cards, then start harassing game devs to write their physics code in OpenCL. Intel gpus, Amd gpus, and even Nvidia gpus can run opencl now. Amd just doesn't have the profitability to be able to subsidize inclusion of this in games like Nvidia can.
 
Intel gpus, Amd gpus, and even Nvidia gpus can run opencl now.
I always get a laugh out of those types of comments. There seems to be mythology that OpenCL is some kind kryptonite to Nvidia. :p

To the point, before going over a little history, is that Nvidia has had support for OpenCL before both Intel and AMD. Nvidia had the first pre-release version, first release version and was shipping it included with drivers long (almost a year) before AMD started doing the same, and of course long before Intel finally had an OpenCL compatible IGP (Ivy Bridge, 2012). Before that, AMD required users to download and install the SDK in order to get OpenCL support, and Intel's OpenCL SDK targeted the CPU only.

It's worth going over the history a bit. Ian Buck, the Stanford grad student who created the BrookGPU streaming language, was hired by Nvidia and he helped develop CUDA. OpenCL is largely based on CUDA (it's relatively simple to port CUDA to OpenCL, but there are a few significant differences... others are simply different names for the exact same functionality), and was helped through the standardization process by Apple (I think because if Nvidia had done it, there would have been much resistance).

OpenCL is not a negative to Nvidia, and there are a few factors why CUDA isn't going anywhere (and also why OpenCL isn't displacing CUDA in many segments). Just tryin' to keep the post length reasonable.
 
Nvidia sure acts like OpenCL is a poison pill though. Seems that Nvidia is trying to pay more F2P game developers early in their development to do exclusive PhysX effects. So when you play on AMD cards you end up thinking this game sure needs some TLC. There isn't much graphically going on. Which nullifies the Nvidia money because you end up playing and spending cash shop money elsewhere. Case in point the game Warframe. If I didn't personally receive a response from the developers that they were going to include more effects for AMD users then I would have stopped playing and spending money long ago.

Just my opinion of course. :)
 
It's in Nvidia's interest to protect their investment in CUDA, but their OpenCL performance issues are self imposed in the drivers. If more games used OpenCL for physics, Nvidia might get nudged into unlocking the potential of their consumer level cards instead of nerfing it.
 
Back
Top