Google Announces its TPU 3.0 or Tensor Processor Unit 3 – More Than 8X Powerful & Delivering More Than 100 Petaflops of ML Hardware Acceleration

Advertisements

Google announces its TPU 3.0 or Tensor Processor Unit 3 - More Than 8X Powerful & Delivering More Than 100 Petaflops of ML Hardware Acceleration

Google announced its second-generation Tensor Processor Unit last year so it was only fitting that the tech giant announced the Tensor Processor Unit 3. The company has also showcased that these chips are going to be liquid cooled and can deliver up to eight times more horsepower compared to the previous generation.

Naturally, the latest chip is going to be for machine learning purposes and to emphasize the power packed performance, Google revealed that for the first time it had to incorporate liquid cooling in its data centers.

Almost all major companies are trying to create customized AI hardware to handle machine operations. This framework for creating machine learning tools is optimized for Google’s TensorFlow. The company wants to increase the presence of Google Cloud and bring it to the same level as Amazon. With Amazon Web Services, Amazon wants to dominate the cloud infrastructure ecosystem.

Today we're announcing our third generation of TPUs. Our latest liquid-cooled TPU Pod is more than 8X more powerful than last year's, delivering more than 100 petaflops of ML hardware acceleration. #io18 pic.twitter.com/m8OH5vFw4g

— Google (@Google) May 8, 2018

Additionally, a lot of startups are also engaged in the development of a custom machine learning hardware. Most of them want to take down NVIDIA on performance or price for AI-related tasks. The TPU 2.0 was launched by Google at last year’s I/O conference, so it was not surprising to see the new iteration this year. However, a lot of issues surround the development of custom silicon. Do developers really need one when NVIDIA’S card can get the work done?

However, data sets are continually getting larger and something like GCP can make it easier and more affordable for companies to utilize their data set and become more defensible. Intel is also engaged in the development of similar products. It is focused on FPGA (field-programmable gate array) that is more flexible and modular and can accommodate the changing needs of the industry. However, not a lot of engineers can program FPGA.

Despite that, Microsoft is also focusing on FPGA and also announced Brainwave yesterday for its Azure cloud platform. At the heart of it all, Google is still an advertising giant, but it is refreshing to see that it is taking on other business segments while also being able to cater to the needs of people and other businesses.

The post Google Announces its TPU 3.0 or Tensor Processor Unit 3 – More Than 8X Powerful & Delivering More Than 100 Petaflops of ML Hardware Acceleration  appeared first on Wccftech.

Advertisements

Leave a Reply