drouge

Quote by spinportal:
"Can CUDA be run on 3x00 and 4x00 GPU series? This would be the more generic path to take since all the applications talking to the CUDA API layer could be handled by an 3rd party driver, much like the STREAM SDK. If the PHYSX driver talks to this bastardized CUDA API middleware driver, its game over. No need to write a specific PHYSX driver for AMD GPUs. This could remove any legal entanglements since PHYSX is licensed differently than CUDA. If displacing a CUDA API with a third-party hack is also a licensing issue, then another mess. Bad enough getting CUDA to translate to the R600/R700 drivers. Most emulation projects are damn close, but none ever get to 100%. Too much special sauce and industry secrets. Even if such a AMD/CUDA driver is written, it will be use at your own peril / risk and prone to GPFs; nothing more than a technological showcase. Havok API is out there in much more software titles; just imagine the day AMD will wake up and have its own driver to take over the calls from the CPU to its GPU. Look at the consoles (Wii & XBOX360) with an ATI/AMD GPU and the game library. Look at the future physics with DX11. The market players have to be ready. If Physx was amazing, why hasn't nVidia gotten the PS3 / Sony touting it? Sony blabs loudly about vaporware to sell promises to be delivered. Physx is just not ready for prime-time, and will be an interesting side-note in history."

Thanks spinportal, this make's for some good discussion. I found it very informative. I agree that physx is just a stepping stone to finding a better solution. And would also like to see AMD publish something along the lines of a Havok driver. They like dropping bombs. And it takes away from thier market share when they keep so tight lipped about it. I see in alot of other forums of people buying Nvidia cards just because they support physx. With AMD's lackluster in driver's, it's no wonder.

n3omatrix

This is something that has crossed my mind as well and if Cuda and CTM were so close in their infancy how hard is it to have Physx ported without the Cuda layer? PS3 and the 360 do make use of Physx and Havok currently physx titles gears of war ,UT3.etc. ,although i am not familiar with how it is run on the 360(my guess would be third core or the unique ATI chipset) , I know that the PS3 runs it using the Cell processors. Physx was being used on the consoles before the purchase of Ageia.

Nice Rig drouge

drouge

Thanks n3omatrix, your sig says 2 pentium D's? On a p5k?

n3omatrix

It's a Pentium D dual core that is white hot with a thermaltake heat pipe and 6600 ballbearing fan ,noisy as hell ,but does the job ,till nov. well i go for a new one.....i bought cheap and clocked the crap out of it....runs all the games excellent with the settings cranked....

Suiseiseki

What ever hapened to AMD's Close to metal?

Wasnt that ment to be like Nvidias CUDA Software?

__________________

servili007

yea, i wonder what became of close to metal.... held so much hope and knowing ATI it wouldve been much more invisibly run than CUDA (CPU usage in cuda of an nvidia running physx is about 4x-20x that of ageia)

n3omatrix

CTM,Brooke, and so forth are still around.......and have a few projects on them ,they are just not in your face as much as CUDA is....

drouge

Yes, but that was over a year ago. I also read that is was more for people writing compilers I guess nothing ever panned out for them. Havok seems like a more likely choice today.

n3omatrix

Yeah ,well from the way Intel and AMD are talking it seems that Havok is going to be a single core dedicated engine instead of a GPU one with plans to add havok fx on GPU (probably to coinside with Larabee) as needed at a later date.

spinportal

There are a few fronts each company is taking, but with NDAs and tight lips, I can merely glean from the research available.

Intel - Larabee and Lucid's Hydra (multi-GPU) tied with Havok / Havok FX. Will AMD benefit from this? Kill off nVidia's SLI completely? As it is, intel chipsets handle CrossFire and snub nVidia SLI.
Microsoft - DX11 might open Physx and Havok APIs directly into the Shader programming layer. Now Physx would need to be re-written and re-routed without CUDA. Will nVidia want to do this? Will intel / AMD allow Havok to be ported? The Xbox360 (funny how apple goes intel and ms went powerpc...) is purported to use DX, but DX11 will not be making it on that console, so no traction there.
Sony - PS3 / Cell / custom nVidia (g80? "RSX") gfx is all well and good. We see Epic's unreal engine using Physx, but for PS3, it must be targeting the Cell SPEs, not the nVidia chip. I wonder how the F@H client for PS3 Cell would compete with PS3 nVidia (g80). Sony loves to toot its horn on how much the PS3 has contributed to F@H so far. How much a thrashing would a GTX280 give a PS3 at F@H (with its limited memory depth / queue)? Also Havok API targets the PS3, but does it make efficient use of Cell PPUs? Is Physx maxed out on the PS3 hardware?
Apple - makes use of ATi/AMD and nVidia gfx, so for them the holy grail would be to have a common API, or OpenCL/OpenGL, and toss Brooke+, CUDA, CTM, whatever. However, MS has invested so heavily in DX to defeat OpenGL, can these two OS makers see eye to eye on a middle ground to let OpenCL work with OpenGL and DX? Doubtful. Then havok and physx APIs would then have to target the OpenCL API (the new middleman for the powerful gfx h/w). See how all this software maintenance is beginning to mushroom?
The synergies are best with intel / apple / (havok layered on) OpenCL / larabee / lucid mix. DX11 shaders are a long shot for any physics API. Nvidia has CUDA; it doesn't need DX11 to do physics. Its in Apple's best interest that AMD and nVidia do OpenCL, instead of letting nVidia port CUDA to OSX with AMD holding the bag and trying to play catch up. AMD is locked in bed with intel / havok. AMD could serve up CUDA, but intel would raise an eyebrow - so most likely lean toward OpenCL. intel and AMD are thinking the same concept of smaller GPUs defeating a single monolithic chip. If lucid's hydra does what it claims, XFire is moot, as is the yet unleashed AMD sideband connector, so AMD is dependant intel's tech once again. OpenCL has the potential to be the "CUDA" API of the cross-OS / multi-vendor world (linux, OSX, windows). It would be in intel's best interest to port Havok API to call OpenCL to leverage and boost the existing software library on every OS. Physx is just not that entrenched, and nVidia appears to be distancing itself with its bluster. Will nVidia put CUDA on OSX (since linux support is there)? The potential for consumer end-products are finally being realized, but the 800 pound gorilla will most likely team up with the schmooz job. Havok / Physx are game dev libraries, and I doubt they want to transition teams to OpenGL or CUDA programming. nVidia is behind the ball with SLI lock-out on mainboards, power hungry chips, and no fusion / larabee vision. AMD is behind the ball in that it's still playing lil bro to intel on the CUDA / OpenCL vs. Brooke / Steam software front and Havok, and checkmated on Lucid vs. XFire. If AMD h/w were to support CUDA, refine XFire even more, activate the sideband advantage, and get havok native gfx acceleration and then getting DX11 shaders to implement havok API calls and also support OpenCL, then there is some longer term viability. CUDA / Physx is not going to be able to maintain traction by the time apple w/ opencl & havok and ms w/ DX11 (or OpenCL) & havok (and possible Linux and game consoles) can leapfrog them.