I just got back from Nvidia’s 2010 GPU Technology Conference in San Jose California. I had an amazing trip, and am thankful that I got to go, as it was my first visit to California as well as my first trade show attendance.
[Side Note: The afternoon before the conference, I took Caltrain up to the amazing city of San Fransisco to visit a friend. He toured me around the city streets in beautiful California weather. There is so much to see and do (compared to what’s around GaTech in Atlanta);
I had a wonderful time!]
At the conference, I was in awe of all the surrounding conference speakers and presenters walking around, and got to talk to a few of them, including some Nvidia folks. There seemed to be quite a turnout of companies on the show floor, and plenty of neat research posters scattered about. I spent most of my time either helping man the AccelerEyes booth, or attending the tech talks.
The setup for the talks was sort of frustrating, since so many I wanted to go listen to were scheduled at the same time. About fifteen different rooms hosted talks simultaneously, six different hours of the day. There was no way to attend to all (or even half) of the presentations I wanted to see. Nonetheless, I still attended many cool talks (I can’t wait for CUDA+OpenCV), and the keynotes were pretty interesting as well (especially Adobe’s plenoptic lens demo).
The neatest part of the conference was getting the opportunity (last minute, not registered at all) to actually speak about some of the matrix multiplication stuff I learned over the summer at AccelerEyes! I was super excited and thankful, as it was an incredible opportunity for getting a little bit of exposure and recognition.
Beside the fact I never travel much, this trip was by far the best one I have ever had. I’m definitely more motivated to take advantage of any opportunities to travel, as it was well worth all the school I missed (and now need to make up)! On that note though, unfortunately, I have come back with a bit more apathy towards making good grades in school, versus doing actual productive things while working at a company such as AccelerEyes. Oh well, this hard work will eventually (hopefully) pay off, as it did for me to go to this year’s GTC!
UPDATE – Random Info Brain Dump: Nvidia announced their CUDA roadmap ; they also announced their partnership with ARM+Tegra ; they really want mobile Tegra to take off, but it doesn’t support CUDA, just OpenGL ; the OpenCV maintainers Willow Garage are working with Nvidia crew to “extend” (not port) OpenCV with CUDA ; Vasily and I were not the only ones saying people should go for multiple outputs of data per CUDA thread, and an Nvidia fellow asked me if they should re-think their ‘CUDA Best Practices‘ guide that focus on the simple one thread per data element output – to which I said they should at least provide examples of doing it both ways, because it depends on the algorithm, but in general you see performance improvement with more work per thread ; there was also an announcement of CUDA-x86 from PGI (CUDA C compiler targeting x86 microprocessors) ; Microsoft demoed IE9 Dx11 GPU acceleration ; I saw a couple of neat 3D TVs on the show floor ; Mental Images gave a cool interactive ray tracing demo in the keynote ; Wow, I did leave out a bunch from the original post, sorry…