TedPage: SIGGraph 2005

This year was my first chance to go to SIGGraph, something I've always wanted to do. SIGGraph is an intriguing meld between technology and art, an intersection that I find interesting and exciting. It's a complex mix of expressing data and expressing emotion. In the end, it seems to come off pretty well.

Emerging Technologies

Overall I found the emerging technologies exhibit disappointing. I expected some really crazy stuff, and found some interesting work, but very little that I would say was beyond my imagination (though I've been told I have a very active imagination). Most of the demonstrations dealt with increasing the interaction between people an computers, making the computer more responsive to the user. There were many ways to do this including everything from lighted gloves to a set of cameras using a phased-array type approach to realize when the user touched a plane to make a touch screen. None were bad, they just didn't meet my expectations for 'emerging technologies'.

The exhibit that I found the most interesting in emerging technologies was from a group in Japan. They created a device that looks similar to a set of headphones, but with ring instead of a cup. The ring encircles the ear and electrically stimulates the inner ear. This allows them to control the wearer's balance. They took two audience members, put the device on them, and asked them to walk forward in a straight line. With a control which looked like it belonged to an RC car they directed the path of the seemingly drunken puppet. His direction did shift. They also had an example of a racing car video game that was modified to work with the device and thus give the player the true feeling of acceleration. This is interesting, and has a tons of applications. Everything from training to medical purposes can be thought of. Motion canceling headphones. It is also a touch scary; when will people try and use this while driving?

Art Exhibit

The art exhibit is an interesting one, not because it covers the traditional focus of art, emotion, because it covers new techniques in art. And, in fact, I found the examples of involving the viewer more interesting in the art exhibit than the emerging technologies area. The techniques involving computers and interaction will be used to create some really impressive pieces in the future. Just as light has contributed to an art piece in the past, the viewer will be a more more central contributer in the future. But, if the viewers are creating the art, are artists becoming meta-artists? I doubt that the art community would ever define the viewers as creating the art though.

One of my favorite pieces consisted of a metal box on the wall. At first, I was very confused and in fact poked it. Then I noticed the computer next to it with a screen, which defines the 'license' for viewing the art and asks you to click 'ok'. The license is very restrictive, and I am violating it now by recording my thoughts on the piece. After agreeing a small window opens with a video playing. As soon as you are starting to watch the video the window slams shut, and the computer next to it reminds you of the license you agreed to. This still makes me chuckle.

Another piece consisted of a table covered in sand that moved a small ball, similar to a mouse ball (for those who remember the time before optical mice). It would listen to the ambient noise, and move the ball on the table depending on what was coming in. In the exhibit hall they had relaxing nature music playing and the ball was making a concentric shell-like pattern. I really wanted to put Nine Inch Nails on the speakers to see what would happen.

Stop for Lunch

Then I met Jon Phillips, Mark Bolger and John Tabor for lunch. We had a good lunch of SVG/Inkscape/Computer tech talk.

I learned that SVG Open cycles so that one year it's in Asia, then Europe and then North America. Since this year it is in Europe, next year it will be closer to home, which would be cool. Mark thinks that the best proposal so far is for Victoria Island in Canada, which is someplace that I'd love to visit anyway; but, it hasn't been decided yet. It would be cool of other Inkscape folks could start thinking about going. August 2006.

Another interesting thing was to learn more about what John Tabor's company, Trafmetrics is doing with SVG. They consult and make a program that helps with traffic planning. This includes maps, and models to determine what should be done to an intersection or street plan with growing traffic requirements. They use SVG for their graphics, editing the SVG to interact with the user's idea of where the roads should be placed. He said that he would write a case study of how SVG can be used in Civil Engineering, I can't wait to see it.

I also learned that the SVG 1.2 specification has been delayed again. Now that we've released flowed text in Inkscape, it would be nice if other SVG viewers could support it too. Err, bother.

Exhibition Floor

The exhibition floor was a fun place, with an absolute ton of people and companies. There were several companies with 3-d printers printing everything you can imagine. And, there were several companies with solutions for human motion capture. It was fun to watch the dancers dance, and then have computer models move with them. Wacom had their new mixed LCD panel tablet which was very impressive, I want one! Apparently there is a 4 month wait for them.

Tech Talk - Architecture of the Future

For those of us who couldn't afford the expensive SIGGraph passes, they had small sponsored talks called "Tech Talks". The one which I attended was sponsored by SGI and was a discussion of PC architecture issues. There were three speeches from Mike Doggett from ATI, Jim Hurley from Intel and Kevin McLaughlin from SGI.

I was late, I missed the speech from ATI.

In the speech from the architect at Intel the biggest point he made is that the bandwidth to RAM is becoming the largest bottleneck for their processors. And, really, more caching isn't solving the problems. They are limited by the board manufactures to 4-layers (surprising, as I know some consumer electronics use 6) so they can't add larger buses easily.

They are working (in research) on the ability to sandwich two wafers together to create a compound chip with the RAM placed directly on the CPU. This way they can use the entire surface as a bus, a short one at that. They use through hole vias to accomplish the connection, which surprised me considerably. The pictures he showed (which looked like photographs more than diagrams) made the walls of the vias look very sharp. I'm curious how they are making them as acid usually makes pitted, or wider holes in the silicon. Either their coaxing the path somehow, or are not using a liquid to do the cutting.

He also mentioned the fact that they are looking beyond dual core to larger numbers of cores on a chip. He mentioned applications like ray tracing as an example of a highly parallel task that would benefit from this. With Intel's Hyperthreading, this got me curious about why they were going the multi-core route. So, afterwords (see I've got more social skills than most of you think) I asked about it, and I learned a lot.

The biggest thing that I learned was that HyperThreading is not SMT. Basically when the CPU boots up in HyperThread mode it evenly divides many of the resources (reordering buffers, renaming registers, etc.) evenly between the two threads. It is, in effect, making almost two CPUs instead of one with multiple threads. There is less intelligence about which to schedule, and less dynamic optimization, so largely single threaded applications can get a significant speed hit. But, if the two threads are independent, and not competing (i.e. not both floating point intensive) the performance increase can be about 25%. But, HyperThreading was not successful much like the PentiumPro was not successful, Windows couldn't handle it. So, it seems unlikely that Intel will go that route, and instead put more CPUs on die. Bummer, I (and the theoretical research) like SMT better.

SGI came off as a company in transition, at least in the public perception arena. The SGI representative talked a lot about them becoming a data visualization company instead of a graphics company. They've always done data visualization, but people seem to have known them for graphics. One of the big things that they're working on now is providing ways for large machines to do the data crunching, and have your desktop do the presentation (or even your cell phone really). High speed connections and culling data should provide an efficient way to do simulations.

He also talked about how much data is being used today. Today, for many applications, several terabytes of data is not uncommon. SGI has sold over forty machines with more than one terabyte of RAM this year already. He continued to hit the bus bandwidth point that was made by Intel previously, mentioning (of course) that SGI machines do that quite well.

An interesting part of the questions was when someone asked if Intel was expecting graphics to be done in the CPU, or if ATI expected general processing to be done in the GPU. Both of the representatives looked at each other and said that they weren't, but it was a funny moment for neither of them would admit that they'll probably be at each other's throats in a few years.

BOF: Cinepaint

Cinepaint is an interesting project which initially started as part of The GIMP (more later). Now it is used in the movie industry, mostly removing dust on the scanned in negatives. It has also been used for some effects. Geeks love this. They love the idea that free software is being used for creating movies that their friends and relatives see. I would say that its history isn't that clean.

Robin Rowe (the current maintainer of Cinepaint) got involved by writing an article for Linux Journal. In it, he talked about an open source tool that is being used in the movie industry but how it really didn't have a webpage or a home. People e-mailed him to get the source code. They sent him patches. And then, they called him the project leader and he started making releases. A very organic ascent to a leadership role in the project.

I found Robin's version of the Film GIMP history interesting, I'm unsure of how much is true or not, but it is interesting. Basically, the movie industry was looking for a tool that could handle the tasks they used Photoshop on Unix for when Adobe announced it was end of life. They looked at buying one, developing someone else's, and decided that it would be easier to fund the development of an open source project. They hired all the core GIMP hackers, and asked them to add in 16-bit support, and they were happy. The movie industry always thought that this would be folded into the main version of GIMP, but instead the GIMP community orphaned the project. The movie studios took support internal, and would have stayed that way had Robin not started accepting patches.

Again, I don't know how much is true or not, but it is a different perspective. I'd love to hear from someone who was there -- but I don't even know who that is! Seems a touch odd, but sometimes reality is strange.

Today, Robin describes the GIMP community as hostile towards him and Cinepaint. While I could understand their disappointment with GIMP not getting the press, I also understand that Robin himself can be difficult to deal with. I'm not willing to look through the GIMP-devel archives to determine who is right here, but none the less, there is basically no ties between the communities today.

It was also interesting to hear Robin talk about his interaction with the various studios. They have Cinepaint, and they have the source code. When they need a feature they hack it in. In some cases, they clean up the patches and send them to him. It seems kinda like Open Source was supposed to be, everyone working together for the common good. The movie industry would have been the last place I'd have looked for that.

I would describe the future of Cinepaint as fuzzy. Robin has decided to totally rewrite the software from the ground up using FLTK and the BSD license. There is a university in Glasgow that has a research team helping with this, but it seems that there is very little visibility into their work. And, as with most rewrites, there will be extreme API breakage. I imagine that there will be a position open to maintain the older version of Cinepaint at some point.

Conclusion

SIGGraph was fun. Sorry the write up got so long, I hope you made it to the end!