GTC 2014 demonstrated that we have now entered the “Battery Powered Supercomputing for the Masses” era. I had the opportunity to experience a Jetson TK1 board running ubuntu 13.04 at the hands-on lab. First impressions were very positive with a snappy response to the Ubuntu window system..
The GTC hands-on labs are oriented for techies and not the press. They provide a very pragmatic, applied approach to introducing attendees to the NVIDIA technology.
The Tegra K1 processor brings the main power unit of top supercomputers to the mobile space; a Kepler-based GPU. The ability to program this GPU using the CUDA platform is going to revolutionize the amazing space of mobile processing applications; from face recognition to machine learning in autonomous robots. In this hands-on lab, we’ll learn how to access the the developer board with TK1, as well as use the OpenCV library and CUDA-enabled C/C++ to accelerate a computer vision task.
It is amazing that the NVIDIA K1 delivers 360 GF/s in a five watt TDP ( Thermal Design Power ) package. To put this in perspective, a single NVIDIA K1 SoC (System on a Chip) is 6-times more powerful than the CM-2 supercomputer released in 1987. Due to high cost, access to the CM-2 was generally limited to the supercomputing elite at the US National Labs and commercial organizations.
Much has been written about the Jetson TK1 at this point. Following are two informative articles:
- Parallel ForAll, “Jetson TK1: Mobile Embedded Supercomputer Takes CUDA Everywhere“
- AnandTech, “Nvidia Announces Jetson TK1“
For those who wish the quick summary:
- A definite “buy to try”.
- Jetson TK1 is available to pre-order today for $192. In the United States, it is available from the NVIDIA website, as well as newegg.com and Micro Center.
- Good news for European customers, the TK1 will be available for purchase at the same time it becomes available in the US!
- Enter to get one free!
- The top 50 Tegra® K1 CUDA® Vision Challenge applicants will be awarded a free Jetson TK1 DevKit and access to technical support documents and assets. Make your submission at this link.
My projection is that a K1 powered Raspberry Pi based systems will be manufactured like popcorn and that the price will drop precipitously.
Until then, I cannot wait to try the chapter 12 real-time video training code from my book, “CUDA Application Design and Development“.
Full CUDA-4.x and CUDA-5.x source code can be downloaded at the Elsevier “CUDA Application Design and Development” site.
Final thoughts: NVIDIA providing each registered GTC attendee a free Shield looks like a shrewd move as it gets everyone thinking about how to work with the K1.
To get a sense of the capability of the current Tegra powered devices, check out the Android tutorial, “How Powerful is Your Nexus 7?”
Leave a Reply