Great software shines even brighter with great hardware underneath it.

As soon as Google unveiled its Tensor Processing Unit (TPU) within this year's Google I/O conference with Mountain View, California, it finally ticked due to this editor in particular that machine learning would be the future of computing hardware.

Certainly, the TPU is only a component of the firm's mission to thrust machine learning – the process that powers chat bots, Siri etcetera – forward. (It's also the chip that defeated the earth Go champion recently. ) Google has TensorFlow, its open source stockpile of machine intelligence software.

In addition to sure, the chips that we find in this laptops and smartphones will keep get faster and more extremely versatile. But, it seems as if we've got already seen the extent on the computing experiences that these processors provide, if only limited by this devices they power.

Now, is it doesn't TPU, a meticulous amalgamation of silicon built tailored for one purpose, and other specialized processors both equally already here (like Apple's M9 co-processor) in order to come, that stands to push this advancement of mankind's processing power – and in return our device's capabilities – further and faster than any other time.

So, we wanted to read more about this new kind of processor, how it's different exactly, the best way powerful it is and how ıt had been made. While Google Distinguished Hardware Electrical engineer Norm Jouppi wouldn't disclose much around the chip's construction (it's apparently except special to Google), he enlightened us over email regarding just the thing the TPU is capable of and its potential in the future of machine learning.

TechRadar: Precisely what is the chip exactly?

Norm Jouppi: [The] Tensor Finalizing Unit (TPU) is our primary custom accelerator ASIC [application-specific integrated circuit] intended for machine learning [ML], and it fits from the same footprint as a disk drive. It is customized to give good performance and power efficiency when managing TensorFlow.

Great software shines even brighter with great hardware underneath it.

What makes the TPU totally different from your standard processor specifically?

TPUs usually are customized for machine learning apps using TensorFlow. Note that we keep use CPUs [central processing units] and GPUs [graphics processing units] intended for ML.

How does the processor operate any differently from usual CPUs?

Our custom TPU is unique in that it uses a lot fewer computational bits. It only fires up the bits which you will want, when you need them. This will give more operations per second, while using the same amount of silicon.