Google Claims AI Chips 30 Times Faster Than Current CPUs And GPUs

Computer chips are getting faster all the time, but there were moments a while back when they were not fast enough for Google. As a result, the tech giant had to create its own hardware that can handle the kind of processing power and speed that it needed. This is where its artificial intelligence chip called the Tensor Processing Unit (TPU) comes into the picture, and according to the company, it’s 15 to 30 times faster than most CPUs and GPUs.

Several years ago, Google was in a bind. Its voice recognition services were exploding and it didn’t have the infrastructure to actually keep supporting it. The company could either spend more money by creating more data centers or find a way to provide that processing power to meet the demands of machine learning requests. Google chose the latter, PC World reports.

The result of its efforts is the TPU, which is basically a processor meant to increase the communication and translation within deep neural networks. Showing off exactly what its chip can do, Google recently published a report that compares the TPU to standard CPUs and GPUs used in data centers.

Based on the numbers provided by the company, it seems the TPU exceeds most conventional chips by a mile and a half. Of course, as Tech Crunch points out, these are numbers that Google is giving out. There are no third-party comparisons to draw any kind of unbiased conclusions from, so it’s worth taking the company’s claims with a grain of salt.

It’s also worth pointing out that these numbers pertain to machine learning and not to conventional computing tasks, like what most people deal with when playing games or working with demanding software. As such, while the TPU might blow its predecessors out of the water, its uses are quite limited.