Separate names with a comma.
Discussion in 'Off Topic' started by Endoplasmic, Apr 6, 2016.
A $2 Billion Chip to Accelerate Artificial Intelligence
Nvidia is pushing use of these kinds of chips and deep learning AI to implement autonomous self-driving for cars. They are selling a development platform and prototype lunchbox-sized water-cooled computers packed with Nvidia chips to run the driving system.
They showed off a demo of the image recognition of the system at CES that seemed impressive to me.
NVIDIA Boosts IQ of Self-Driving Cars With World's First In-Car Artificial Intelligence Supercomputer
Oh that's pretty cool! I was just posting since the name is the same name as the leaked next model
Computers built with it cost $130k though..
I didn't have time to look at the details on that, but I suspect that $130,000 is for computers that train the neural network.
Basically, they have teach the system how to drive much the way a person learns. It processes lots of real-world video taken from cameras mounted on cars and learns how to recognize what a motorcycle, or a dog, or truck looks like and where the lanes are on a road. I think it also gets inputs from the steering wheel, accelerator and brakes from actual drivers in order to learn how real people drive.
That learning requires lots of processing power and time. The result is a big file full of neural net "memories" storing what it learned. This is then installed in the car's "lunchbox"-size AI computer which then makes decisions based on that learning. This requires much less computer power than the learning process but it still takes a lot so they use specialized chips to make it go fast enough.
I vaguely recall that the goal was to get the lunchbox computer in the car down to around $2,000 or less.
Gotta start somewhere....