Update (2010.04.30): this news is based on a one-year old news from bit-tech.net. Why X-bit labs serves us an old news ? Seems an error… the original page at X-bit labs has been removed… Not cool from a major website to publish such kind of news. This news is not wrong (and maybe still active at NV) and refreshing an old news is useless… but OpenCL PhysX seems to be a hot topic if you count the number of sites that have relayed the news 😉
According to X-bit labs, NVIDIA could eventually release a version of PhysX accelerated by OpenCL.
Keep in mind that an OpenCL version of PhysX does not mean a complete re-writing of the engine but only the GPU accelerated part which is actually based on CUDA. One consequence will be that Radeon owners will enjoy fast PhysX too. Another second consequence will concern the devices that are OpenCL capable: CPU, GPU and even NPU (Network Processing Unit like EVGA’s Xeno)…
AMD’s OpenCL implementation supports both CPU and GPU codepaths. So a system with a high end 6-core CPU and a dual-GPU graphics card will be a killer PhysX rig (of course if the OpenCL PhysX version allows the use of CPU which is the responsibility of NVIDIA). And if we are lucky, AMD will finally release an official OpenCL support like the one already done by NVIDIA.
However, Mr. Mohammad warned about possible performance issues with non-Nvidia hardware, claiming that ATI is much behind Nvidia when it comes to GPU computing.
…possible performance issues… and possible lack of features 😉
Like OpenGL, OpenCL is based on extensions for new features, and AMD will have to follow NVIDIA to keep PhysX operational…
That will be a great move on Nvidia’s behalf. This could be a piece of cheese though, so the mouse can be caught.
I have two 5850s anyway, so if need be, I could assign on of them for gpGPU OpenCL Physx, whatever.
As for Nvidia being far ahead in gpGPU, well good luck to them establishing a user base with their Fermis at these prices. And no, I don’t think that GT200 is any better in gpGPU than the Evergreens. ATI just builds what users need NOW.
Is it lookng strange only for me, that this article at X-bit Labs is referring to other article, by bit-tech, published 27 March 2009 ? Like digging 1 year old news is fun ?
Yep I didn’t checked bit-tech link… Seems indeed an old news…
Fizzx-like OpenCL for Linux also. Interesting.
But what could prevent NVidia from disabling the GPU acceleration if the device is not a Geforce just like what they are doing right now?
And it would be not surpising at all to see that radeon are 100x slower than Geforces too, because of some “strange code paths”.
So, use Bullet instead. Vendor agnostic, open source.
Unlike PhysX.
Nice… because ATI’s OpenCL implementation does not even support image textures…
I doubt the world will see a proper OpenCl PhysX realisation for AMD’s chips…
Anyway, what’s in it for nVidia? Use the resources for porting to OpenCL to enable PhysX at competitor’s hardware? I don’t believe it.
What’s in for nVidia? Greater adoption of PhysX otherwise it will be surpassed by Bullet, Havoc, or DMM in short time. Because only working on nVidia’s GPU is a developer concern and risk! No one sane would invest any time on it unless there are other motivations (like greenbags rolling from nVidia to developer company)
For now the future of Bullet is very unclear, I don’t think it will gain any significant market share in nearest 5 years…
Pingback: Thermalright Spitfire VGA cooler and VRM-R5 Heatsinks Tested With FurMark - 3D Tech News, Pixel Hacking, Data Visualization and 3D Programming - Geeks3D.com
Pingback: Anonymous