Filed under SIGGRAPH
I spent most of my day today at the Nvidia Exhibitor Tech Talks.
The first talk was by Evan Hart, a long time ATI-guy who recently jumped ship to Nvidia. The talk was called "OpenGL Extensions for Next Generation Hardware." Basically, this talk was about new features in the hardware that supports DirectX 10, or Shader Model 4.0. This new hardware adds a new level of programmability to graphics hardware....a geometry shader. A geometry shaders, for the first time, allows a graphics card to create or delete geometry on the fly without any help from the CPU. This opens up many new possibilities: hardware accelerated curves/NURBS, fur, automatic terrain tessellation, and point sprites to name a few. There is no DX10/Shader Model 4.0 hardware announced yet. I asked if current hardware would support the new features with a driver update and the answer was "You'll need to buy new hardware." So if you are looking for a graphics card...wait until you can get a DX10 board. Since DX10 will only run under Windows Vista (not XP), I would expect DX10 hardware around the time of the Vista launch early next year. Nvidia was very careful not to give a time frame for when they would release their DX10 hardware.
The next talk, GPU Physics, was by Mark Harris from Nvidia and Chris (didn't catch the last name) from Havok. They showed how a graphics card could be used to accelerate physics via the Havok physics engine running on a graphics card (called "Havok FX"). In the demos, it appears you get about a 10-100x improvement in the number of collision dections between objects when comparing what you can do on a CPU verses what you can do on a GPU. So instead of having 1,000's of objects on the screen, you get 10's of thousands. Havok can run on any platform (source is given, I believe). Currently Havok FX is limited to using a single GPU (this will change in the future). Future plans include the ability to calculate on the fly how something breaks apart. For example, you could send the graphics card a wall that is basically a cube. If you hit the wall with a bomb, future versions of Havok will be able to automatically break the wall into pieces.
The next talk was about Nvidia's FX Composer 2.0. New features include support for GLSL and OpenGL (no Linux support). FX Composer uses COLLADA as its standard file format for import/export of shader and model data. FX Composer can be used for creating, debugging, profiling, and tweaking shader code via GUI elements (for artists) or code (for programmers). Should be released at the end of 2006.
The last talk Nvidia I went to was "DirectX 10 Effects." Three demos featuring Shader Model 4.0 effects were shown: fur, cloth (that can be torn in real-time), and metaballs. After each demo, we went through how the effect was achieved and studied the shader code.
I picked up another book today (I can't stay away from the bookstores). This one is one of the only new books I saw at the show...."COLLADA, Sailing the Gulf of 3D Digital Content Creation." I am a big COLLADA supporter...I'll have more to say about it tomorrow when I go to the COLLADA BOF (Birds Of a Feather).
The last meeting I went to was for the OpenGL BOF. They had 8 presentations in less than 2 hours...which was a bit too much info in too short a time. They announced with the downfall of SGI, OpenGL would be controlled by the Khronos Group from now on. They announced Vista support of OpenGL is on the same level as Direct3D, so OpenGL will run like it does currently under XP, and not a slower emulated version as was feared last year in order to make it work with Vista's Aero Glass. It sounds like OpenGL 3.0 will be the equivalent to DirectX 10/Shader Model 4.0. It is not expected until SIGGRAPH 2007 at the earliest.
One of the talks was about how OpenGL can better address developer needs. When listening to what people wanted and what is planned by the OpenGL group, I couldn't help but think that OpenGL is trying to be what DirectX is today, but expects to get there in 2 to 3 years (documentation, SDK, code samples, features, tools, etc.). And since DirectX is a fast moving target, I see OpenGL just falling further behind. The only real benefit that OpenGL has over DirectX is that it runs on non-Microsoft platforms (Mac, Linux, Unix, etc.). If OpenGL is not going to lead anymore, maybe it should just become API compatible with DirectX and then we'd only have to learn about one API. It is actually more expensive to develop OpenGL apps than DirectX. You get plenty of really good DirectX developer tools between the DirectX SDK and Nvidia's SDK (both of which are free). If you want a debugger for OpenGL, it will cost you. The only game in town is gDEBugger for $490!
The SIGGRAPH Reception was at the Seaport World Trade Center. I hung out with a friend and former co-worker, Eric Hirschorn.
One more day to go!