Revelations at GPU Technology Conference in San Jose

At the Nvidia GPU Technology Conference in San Jose, CEO Jen-Hsun Huang announced four new Nvidia products and concepts. 

The, maybe, 4,000 attendees at the keynote speech heard about and saw the new Titan X GPU card/chip. It has an 8 B transistor processor, an amazing 7 Tflop performance, a 12 Gbyte frame buffer, and will cost only $999. Available soon.

Neural networks and the concept of deep learning for voice recognition and video recognition is a key area for Nvida and its chips, cards, and boxes. 

The DIGITS DevBox GPU system has four GPU cards and is intended for researchers. Expected to ship in May and provide 28 TFlops performance while using only 1200 w. Cost will be $15k.

Jen-Hsun then talked about the Nvidia roadmap. Of particular interest was the Pascal future generation GPU. Due sometime in 2016, it will use 3D memory, a new high-speed connect, and will yield 2X floating point and 3X memory bandwidth. Over all, for neural networks and deep learning , expect the chip to be 10x faster than the Maxwell GPU.

The last new item is the Drive PX – a platform for self-driving automobile research. This system has 12 camera inputs and will cost $10k. ADAS work is a big research area for Nvidia. The speech recognition aspect of deep learning is one of the key automotive areas.