Future of Rendering, Ray tracing on the GPU and TWiT’s UGM
Past and Future of Graphics RenderingLater in the day NVIDIA’s Tony Tomasi and David Kirk took stage in a technical session to discuss the evolution of the GPU. This time line below, while blurry, gives us a general sense of the direction that NVIDIA sees the GPU taking in the coming years.

This slide indicates what we MIGHT see in a GPU in 5 years, 2013. New features would include full support for C++ code, preemption, a virtual pipeline (closer to what Larrabee offers soon?), complete pointer support and adaptive workload partitioning.

What will these features bring computationally to the GPU? A completely arbitrary data flow is the most key unit here and continues the GPU on the migration towards more CPU-like features while maintaining the compute power of a GPU. The next points might seem counter intuitive – NVIDIA sees both a general purpose programming model and special purpose hardware included with the GPU ecosystem – essentially creating a more programmer friendly environment while keep the fixed function hardware where necessary to keep the performance of the GPU on the rise.
The note of hardware managed threading and pipeline is counter to what Intel will be doing with the Larrabee in the near future. Intel’s upcoming GPU is completely software managed and the pipeline is nearly 100% programmable with the exception of the texture units. Obviously NVIDIA’s goal in 2013 will be the same as it is for 2008 – completely merge the ability for the GPU to do typical graphics work and general purpose computational work.
The note of hardware managed threading and pipeline is counter to what Intel will be doing with the Larrabee in the near future. Intel’s upcoming GPU is completely software managed and the pipeline is nearly 100% programmable with the exception of the texture units. Obviously NVIDIA’s goal in 2013 will be the same as it is for 2008 – completely merge the ability for the GPU to do typical graphics work and general purpose computational work.

What kind of hardware power will a GPU in 2013 have? The numbers that Tomasi put forward are incredibly inspiring and include more than 20 TeraFlops of floating point performance, 1.0 TB/s of memory bandwidth and
even faster AA and texturing performance.
Ray tracing demo on the GPU
Next up was NVIDIA’s David Kirk and a demonstration of real-time ray tracing on the GT200 GPU.

The demo is running at 30 FPS in most areas but in some places the frame rate can drop quite a bit when close up on items that require a lot more reflective detail. The whole point of the demonstration though was to prove to those critics out there (read: Intel) that said ray tracing couldn’t be done effectively on the GPU.






These screenshots of course don’t really do the demo any justice and we’ll be on-hand for the next live demonstration with our HD camcorder for your viewing pleasure.
The TWiT UGM Shows up at NVISION
For uninitiated, the TWiT UGM is This Week in Tech‘s Ultimate Gaming Machine – a build that I had a small part in by suggesting parts and helping through the installation section. TWiT is hosted by Leo Laporte, a name that should be familiar to most of you, and the UGM project was headed by Colleen who works for him.



The basic system specs include a Core 2 QX9770 processor running at 4.4 GHz, 8GB of Corsair DDR3-1600 memory, 2 x 30GB VelociRaptor hard drives in RAID 0 and of course, 3-Way SLI with GeForce GTX 280s. The CPU and GPUs are all being water cooled with a custom setup designed by Colleen. I will have some much better photos as well as an in-depth write up on the Ultimate Gaming Machine some time next week as I plan on doing a review of the system for PC Perspective.
Final Thoughts
So far, though the keynotes and presentations at NVISION have bee a bit of a let down, the show itself is proving to be quite useful and entertaining. The LAN party was still going strong as of Tuesday night and the 3D stereoscopic gaming demonstrations that are being shown to the public are actually getting a lot of good feedback. The exhibit area is littered with NVIDIA partners and more great technology demonstrations. We got our hands on Battlefield Heroes, Call of Duty: World at War, Microsoft’s Photosynth and we even got to gawk at fancy new Maserati. What is probably most valuable to me though is the closed door meeting with NVIDIA executives pining about the future and what Intel, NVIDIA and AMD have to look forward to. You can be sure we’ll have all that information wrapped up in some editorials in the coming weeks.
We have just one more day left of the show that includes the closing keynote from Jen-Hsun Huang with the two hosts of MythBusters – apparently doing something very cool. I’ll be attempting to stream that live, so be ready at 3pm PST!