On 19 February 2008 at the Game Developer’s Conference 2008, in the Microsoft Game Developer Day tutorial, Chas Boyd discussed DirectX Futures. There was some redundancy with his section in the Advanced Visual Effects tutorial from the day before, but also some interesting new discussion.
With respect to general-purpose computing using GPU’s (GPGPU), Boyd discussed the issue of language choice, enumerating some of the current available high level languages for GPGPU computing (MS Accelerator, Intel Ct, RapidMind, Acceleware) and some of the low level API’s (nVIDIA’s CUDA, AMD’s CAL/CtM). The idea here is that the jury is still out as to what is best. It currently also is unsure how to decide what the best division of labor is between dedicated (e.g., fixed function hardware) silicon vs. software on the GPU. Some things, he pointed, out, still make sense on dedicated silicon such as texture sampling and decoding with some codecs, or format conversions that can be simple on hardware but might require complex code.
Boyd discussed some of the trends in display technology, and how that might impact DirectX. For example new features such as higher pixel density, higher refresh rates, and high dynamic range display technology with better color gamut support might require new surface formats or texture formats, etc. (I’ve seen some HDR displays for example that actually are two displays in one, layered displays with each providing a certain number of “bits” in the dynamic range. DirectX support for such multilayer displays might require some specialized runtime code.) Microsoft is currently working with vendors to develop display protocols so that displays can communicate their capabilities to the DirectX runtime.
For the remainder of the talk, Boyd presented an interesting personal view of the long term future. Chas indicated his belief that in 5 or so years, all “performant” client computing will be done Direct3D style (not necessarily Direct3D, but in the same vein). What he means by this is that the Direct3D approach to using massive numbers of processor cores is perhaps the most successful use to date of massively parallel computing, and that the programming model is one that works well. (Well, of course this approach didn’t originate with Direct3D’s programmable shader pipelines…but the point is perhaps valid nonetheless.) The current GPU programmable shader model already exploits hundreds of cores, enabling massively parallel processing on consumer PC’s and laptops. The trend for the foreseeable future is that the number of cores will continue to grow, and that it is this growth from which computing power will grow. Some predictions, Boyd reported, based on the current GPU vendor schedules, expect more than 1024 cores on a GPU by 2010 and more than 8000 cores by 2015. All I can say is, “Wow.”
With so many tools available for graphics and rendering and computing innovation, I am truly looking forward to the next ten years in gaming and graphics and cinema. How about you?
Southwest Gaming Expo
