- Google unveils Ironwood, its 7th generation TPU
- Ironwood is designed for inference, the new big challenge for AI
- It offers enormous advances in power and efficiency, and even surpasses the El Capitan supercomputer
Google has revealed its most powerful AI training equipment to date as it seeks to do another step forward in inference.
Ironwood is the company’s 7th generation tensor processing unit (TPU) – Equipment feeding Google Cloud and its customers training and management of the workload of the AI.
The equipment was revealed during the company’s Google Cloud event in Las Vegas, where he wanted to highlight the great progress of efficiency, which should also mean that workloads can be more profitable.
Google Ironwood TPU
The company claims that Ironwood marks “a significant change” in the development of AI, making part of the decision of reactive AI models which simply present real-time information for users to process, to proactive models which can interpret and deduce by themselves.
This is essentially the next generation of AI computer science, estimates Google Cloud, allowing its most demanding customers to set up and establish increasing workloads.
In its high -end iron wood, can evolve up to 9,216 chips per pod, for a total of 42.5 exaflops – more than 24 times the calculation power of El Capitan, the largest global supercomputer.
Each individual chip offers a peak calculation of 4,614 tflops, what the company says is a huge leap forward in capacity and capacity – even its slightly lower configuration of “only” 256 chips.
However, the scale can become even greater, because Ironwood allows developers to use the software battery of the tracks designed by Deepmind of the company to exploit the combined calculation power of tens of thousands of Iron wood TPUs.
Ironwood also offers a major increase in the high bandwidth capacity of the bandwidth (192 GB per chip, up to 6x higher than the previous sixth generation TPU in sixth generation) and bandwidth – capable of reaching 7.2 tops, 4.5x greater than that of trillium.
“For more than a decade, TPUs have fueled the training and work service of the most demanding AI of Google, and have allowed our Clients to do the same,” noted Amin Vahdat, VP / GM, ML, Systems & Cloud AI.
“Ironwood is our most powerful, most capable and energy -efficient TPU. And it is specially designed for powerful thinking, large -scale inferential AI models.”
