[PREVIEW] Multi-Core CPU Support in PhysX Seems to Work
Multi-core CPU in PhysX in action…
Do you remember this news where AMD claimed that multi-core CPU is disabled in PhysX ?
I’m currently updating Geeks3D’s PhysX FluidMark tool and from my last tests, multi-core CPU support in PhysX seems to be ok (that confirms what NVIDIA said in this news)… At least on my dev station with an ATI Radeon HD 5770 and an AMD X2 3800+ 
More on this topic when FluidMark will be released. Stay tuned!
Tweet
[ Subscribe to Geeks3D latest news by email ]
[...] brennt nicht nur ab, er bricht nun auch physikalisch korrekt zusammen, wie eine Steinmauer http://www.geeks3d.com/20100216/prev…ork/#more-6939 Auf der Firefox-Roadmap steht zumindest etwas von "multi process plugins" ( [...]
[...] Geeks3D is reporting that they are working on a new version of their PhysX FluidMark benchmark that will support Multi-Core CPU Support for running the PhysX calculations. In recent news AMD called out NVIDIA claiming that they have purposely disabled multi-core CPU support for PhysX. NVIDIA then released a statement claiming that this was simply not true stating “Our PhysX SDK API is designed such that thread control is done explicitly by the application developer, not by the SDK functions themselves“. So this new update to PhysX FluidMark should answer this question once and for all and allow people to do some really interesting tests and benchmarks with PhysX running on the GPU or dedicated PhysX card and then benchmark with the CPU handling all the PhysX calculations. [...]
I think AMD missed something and NV didn’t lied here, and maybe AMD didn’t missed anything but they misused some words. I think it will be better to say that PhysX implementation in current PhysX enabled titles isn’t optimized for multi-core systems
I think that PhysX does indeed support Multi-Core, but when going to NV’s TWIMTBP program and thier work to embed PhysX in games, they only optimize advanced physics details for GPU working, but will not make anything regarding multi-core… as they don’t want to create multi-path for different level of details for each number of cores !!
the question is…
when a PhysX app that uses a GPU doesn’t find a GPU, well it automatically goes to CPU processing ( and multi-core optimized ) or there must be a seperated path for each of these…
I think with most ( if not all ) PhysX implemtations in games, the PhysX API has two paths and the path for CPU acceleration is not that well used or optimized to work with multi-core CPU’s, they will need more Physics details profiles to retain the different configurations used in real world and this means more work…
Its more like the laziness of devvers to properly thread the PhysX implementation that leads to be it being far slower then what has been seen in apps like 3DMark Vantage.
but oh… wait. Vantage has 12 THREADS!