When AMD will support Havok physics acceleration in GPU?

[suryasans]

AMD and Intel each other is competitor. Since there are trouble relation after Intel fined by EU, I don't hear anything about when Havok physics will be implemented using ATI Radeon as accelerator. Havok is owned by Intel and will be optimized using Intel CPU or accelerated with integrated Intel GPU. I think physics acceleration in the game will be implemented in many games in the future. I hope AMD will find their own solutions.

[charlie]

They demo'd it at GDC in march, and it worked. It usually takes months to go from demo to product, so maybe early 2010. The demo at GDC was only worked on for about a month, so it was pretty rough. Give it time, but not much more time. If it isn't working in ~6 months, get worried.

-Charlie

[265586888]

AMD's implementation is Havok Physics @ OpenCL.

Havok Cloth on OpenCL: http://www.youtube.com/watch?v=xfrM973spw0

[suryasans]

Originally Posted by 265586888

AMD's implementation is Havok Physics @ OpenCL.

Havok Cloth on OpenCL: http://www.youtube.com/watch?v=xfrM973spw0

That's why Intel did not like their valuable Physics API used with AMD's GPU. They want AMD implemented it with their own CPU not GPU because currently they did not have the discreete products.

[265586888]

Originally Posted by suryasans

That's why Intel did not like their valuable Physics API used with AMD's GPU. They want AMD implemented it with their own CPU not GPU because currently they did not have the discreete products.

You should read the OpenCL article in Wikipedia, OpenCL can run on CPUs and GPUs.

AMD's GPU implementation of OpenCL will be incorporated into later versions of ATI Stream SDK, currently only x86 implementation is in beta testing.

[charlie]

Originally Posted by suryasans

That's why Intel did not like their valuable Physics API used with AMD's GPU. They want AMD implemented it with their own CPU not GPU because currently they did not have the discreete products.

Ummm, no. Intel (Havok) was fully involved in this, they want their API to be all over the place, and a standard, and are doing the right thing to get it widely adopted.

Compare this to Physx which is only a standard in NV press releases.

-Charlie

[rich wargo]

But, Charlie, don't you remember? When NVidia develops something, it just naturally becomes a universal standard, you know, like the law of gravitational attraction.

It comes with being conceited congenital arseholes.

[265586888]

Originally Posted by rich wargo

But, Charlie, don't you remember? When NVidia develops something, it just naturally becomes a universal standard, you know, like the law of gravitational attraction.

It comes with being conceited congenital arseholes.

NVIDIA dumps money to make them de facto standards, that's different from the open standard side of things for NVIDIA.

[Robert Varga]

I hope PhysX will die coz it's not open. OpenGL is open. OpenCL has only one different char so we'll see.

That ATI Stream... I heard so much about it, but... When will be out something really BIG? Something in some game, or in some application (hey but someway usable).

[265586888]

Originally Posted by Robert Varga

I hope PhysX will die coz it's not open. OpenGL is open. OpenCL has only one different char so we'll see.

That ATI Stream... I heard so much about it, but... When will be out something really BIG? Something in some game, or in some application (hey but someway usable).

Originally, ATI Stream is "Brook+ with CAL" *, and that will be half-dead when OpenCL and DirectX 11 Compute Shader made debuts in parallel processing.

Where in gameplay, Havok GPU physcis will be written in OpenCL (cloth and destruction were demonstrated) and Compute Shader will allow proprietary gameplay physics engines to write in DirectX code.

Just take a look at VR-Zone's report, "Designed for DirectCompute 5.0 and OpenCL", this line is crucial. With no mention to the original solution, only implying through the line "ATI Stream Technology".


Meaning that the "ATI Stream Technology" includes the following:

• "support for DirectCompute 5.0 (i.e. DirectX 11/Compute Shader 5.0) and OpenCL 1.0" and
• the stuff that already exists: "Brook+ with CAL" *

Guess which one has a higher priority in the list?

Why would Brook+ be half-dead when OpenCL and DirectCompute debuted?

One is hard to code and hard to optimize, you have to access to low-level code, and as Tim Sweeney pointed out in "the end of GPU roadmap" presentation (slide 71 of 74):



According to that presentation, that was only meant to say CUDA only, which developers have no need to optimize the code.
When count in the low-level optimizations using AMD's CAL solution, you may get 100x the "money, time and pain" to do it, consider it's only an common scenario.
And the number will go up exponentially when the complexity of the algorithm goes up linearly.

Two is AMD don't want to put too much resources in promoting these technologies, we just put away the "this may harm its CPU business" argument for the moment because it's just speculations and currently AMD believes in "platformance", fact: AMD simply don't have the extra money for large-scale promotions. So the promotion to Brook+ will remain more or less the same, probably even less.

*Originally, ATI Stream technology is a solution with:

• Brook+ as the high-level programming language; plus
• low-level optimizations via the Compute Abstract Layer (CAL) using
  1. AMD Intermediate Language (IL) and
  2. machine code with reference to GPU Instruction Set (ISA)