SIGGRAPH 2009 New Orleans in Review - The Year of the GPU
by Jeff Mottle
This year a hot and muggy New Orleans hosted SIGGRAPH 2009. As expected, the economy played havoc with attendance hitting only 11,000 attendees, down over 60% from 2008 and down almost 45% the last time the show left the West coast. Despite the lower numbers however, exhibitors across the board commented that it was the best SIGGRAPH in a very long time, citing more qualified attendees who were serious about doing business. From a personal stand point, this year was my busiest event in 13 years, and is one of the first years in recent history where new developments actually got me excited about the future of the industry.
Every year a sub-theme usually emerges at the event, closely following the technology development trends in the CG industry. This year, that trend was overwhelmingly focused around the GPU. From entire panels about the future of GPU computing, GPU based product announcements and new hardware developments, the speed of the GPU is by all accounts the single largest development in the CG field for many years. Most I spoke to expect the GPU to be integrated into mainstream workflows within two to three years.
A private press event hosted by Jon Peddie research gathered some of the top minds in the field of GPU research and adoption. This year's panel included chief scientists and engineers from NVIDIA, AMD and Intel, creative directors from BMW, researchers from the Aronne National Laboratory as well as leaders from mental images, Autodesk and StudioGPU. While there were varied opinions and thoughts about the future of the GPU and how it will evolve and be used, a few interesting points were made during the presentation.
Panelists from Jon Peddie Research Panl: (top to bottom, left to right) Dr. Bill Dally, Chief Scientist (NVIDIA), Eric Demers, GPG Chief Technology Officer (AMD), Gary FitzGerald, Creative Director, Visualization (BMW Group DesignworksUSA), Yoni Koenig, Chief Scientist (StudioGPU), Anwar Ghuloum, Principal Engineer (Intel), Mark Hereld, Researcher in Mathematics and Computer Science Division (Argonne National Laboratory), Rolf Herken, CEO and CTO (mental images), Ken Pimentel, Director of Visual Communication Solutions (Autodesk)
One panelist stated that the gaming industry is setting the standard for UI and interaction with visualization, specifically the overwhelming popularity of the Wii. Over the next ten years this will effectively change how we work. Yoni Koenig, Chief Scientist at StudioGPU, cited the parallelization the UI and how this will affect how we work and interact with scenes in real-time. Yoni, commented that user interfaces will allow artists to interact with their scenes without using numbers, but rather arbitrary sliders and controls that affect the look of the scene, allowing for a more reflexive interaction.
It was also said that web based application will play an important role in how we work by removing the burden to end users of keeping up with the technology curve. The internet will migrate from a file compute system to a cloud compute system.
On the GPU manufacturer front, I met with AMD and NVIDIA and will be doing an extensive reviews of their cards on production models in a number of applications. I'll be testing display performance as well as compute power. For more information on the status of this review click here.
There are quite a few software manufacturers experimenting with GPU right now, although nothing is production ready at the moment except for StudioGPU's MachStudio, which has been used privately by their studio for a number of years. MachStudio released earlier this year as a commercial application and they just announced version 1.2 several weeks ago. CGarchitect will be reviewing MachStudio later this year.
Boy Wonder - Rendered with MachStudio
Mental Images, Chaos Group and Caustic Graphics all gave attendees and press a sneak into their GPU integrations at SIGGRAPH as well. We interviewed Caustic Graphics a few months ago and while they do not have a commercial product just yet, they did show a technology demo in the Autodesk booth.
This demo showcases their hardware raytrace acceleration card being used to do real time ray tracing in the Autodesk® 3ds Max® software’s view port.
There has been a lot of speculation about whether a hardware solution that sits beside your GPU will win out over GPU only solutions, but it is still early days in the GPU race. I think a lot of their success will revolve around how much horsepower companies like NVDIA and AMD can pack onto the GPU and most importantly who software vendors will adopt. As GPU rendering requires substantial changes to current rendering application vendor code bases, it's unlikely many will adopt multiple technology integrations. To see the latest developments from Caustic Graphics head over to our forums here.
( We have just posted a new video and background information about the demo.
At this year's Chaos Group end user meeting, a packed house was treated to a demo of their new V-Ray RT application running on a consumer grade GPU. We have an exclusive recording of this demo here along with information about the demo and the status of the technology.
Mental Images is also working on iray, their GPU rendering solution. While there are no public technology demos available yet, we were shown a private demo a few weeks ago. The quality and speed were phenomenal. While the quality of most demos being shown by other GPU rendering solutions have not approached the quality of current CPU based production renders, iray most certainly has made some significant advances in this area. I'll be trying to get some screen shots and videos for the site soon, but the architectural scenes I saw rendered in only a few minutes were very compute intensive scenes with complex materials lit fully by secondary illumination.
While the resolution was not overly high, the same scene on a CPU would have taken much longer to process. In fact iray is 20-50 times faster than the CPU. The scenes I was shown were rendered on eight NVIDIA FX 5800 GPUs using Reality Server to render across two machines. Each machine had four GPUs. It is important to note that iray is not just a code change to allow mental ray to use the GPU, but rather a new GI simulation engine. With iray, artists will no longer have to worry about ray settings, quality and noise settings etc. You simply place the lights and create your materials based on the physical properties of the object you are trying to simulate. Once rendering starts, the quality converges over time. You decide when the quality is sufficient and save the final image. If you need to go back to the scene to refine the quality at a later time, you will be able to continue refining the scene where you left off. Leaving your scene unattended will not be an issue as termination parameters will allow you to set time and target quality settings.
For on the fly lighting and materials tests a new ray brush will allow you to paint the areas of importance. These areas will converge first allowing you to focus the rendering only on the parts of the scene you are refining.
The accuracy of the rendering will be at least as accurate as the current CPU version of mental ray, however the first version will not be able to perform any spectral calculations. There are some limitations of iray, namely that your scene must fit into RAM in order to process. However, the current implementation will allow you to fit approximately 40 million triangles into 4GB of GPU RAM. Iray will be in the hands of OEMs like Autodesk in the next month, but it will be up to them to decide how iray will be integrated into their applications.
This year I also spent quite a few hours meeting with SIGGRAPH organizers to talk about how we can get more architectural visualization representation at the conference. While it sounds like it will be a lot of work, I would really like to see what we as an industry can do to help make this happen. If you would be interested in helping out, please shoot me an email to firstname.lastname@example.org. We need both industry vendors and production companies to become actively involved in planning content and pledging support for architectural visualization in order for us to gain any traction within the organization.
Overall SIGGRAPH 2009 was one of the most energizing events I've been to in my 13 years as a SIGGRAPH goer. The GPU is going to revolutionize the way we work and I personally am excited to see where it goes. It's been a long time since a technology has shaken up the industry. I'll be attending NVDIA's GPU Conference this week and will report back with new developments in the field of GPU computing in a few weeks. In the mean time, check out Autodesk's Ken Pimental's blog where he summarizes the progress current GPU technology here.
See you at SIGGRAPH 2010 in Los Angeles!
If you would like to post comments or questions about this review, please visit our forum
You must be logged in to post a comment. Login here.