“Deep learning:” it’s a phrase that must be music to the ears of every first grade teacher. It’s also an emerging area of machine learning research, one that engineers are exploring with a new super-efficient graphical processing unit (GPU) designed for artificial intelligence.
MIT researchers recently unveiled a chip that is 10 times as efficient as existing GPUS. It can be used in a number of different things such as vehicles, appliances, manufacturing equipment, and even livestock. The goal is to enable these devices to adapt to their surroundings using artificial-intelligence algorithms.
“Deep learning is useful for many applications, such as object recognition, speech, face detection,” said MIT Professor Vivienne Sze in a statement. “Right now, the networks are pretty complex and are mostly run on high-power GPUs. You can imagine that if you can bring that functionality to your cell phone or embedded devices, you could still operate even if you don’t have a Wi-Fi connection.”
Dubbed Eyeriss, the new chip has 168 cores. That’s roughly the same amount as existing mobile GPUs have, but Eyeriss’s efficiency comes from its ability to compress data and allocate tasks to multiple cores. That allowed engineers to eke out the last bit of performance from each core before it has to tie up more memory, and thus use more power