This announcement focuses on accelerating the creation of artificial intelligence "thinking" and is the highest performance to date.7th generation TPU "Ironwood"Mark Lohmeyer, vice president of Google Cloud and general manager of artificial intelligence and computing infrastructure, said in an interview that although the current cloud infrastructure will be adjusted to meet the application needs of artificial intelligence technology, it will still provide appropriate infrastructure according to different market needs.
The demand for artificial intelligence will not completely change the cloud infrastructure
Mark Lohmeyer pointed out that although the 7th-generation TPU "Ironwood" design was indeed added to meet the needs of artificial intelligence computing, and it is advertised to have a computing power of 42.5 Exaflops, which is about 24 times the computing power of the world's largest supercomputer El Capitan, and the data transmission bandwidth is as high as 7.2 Tbps, which is 4.5 times higher than the previous generation "Trillium", the previously launched TPU hardware architecture is still retained to meet the needs of more cloud applications. At the same time, it is also cooperating with Intel, AMD and NVIDIA on computing chips. This time, it even announced that it will obtain NVIDIA's upcoming CPU code-named "Vera" and the next-generation GPU code-named "Rubin".
Although much of the current infrastructure is prepared for AI computing, Mark Lohmeyer emphasized that Google Cloud still retains application diversity and provides users with more choices, including providing its own TPU accelerated computing devices from the hardware infrastructure, and even choosing "Axion" CPUs built on the Arm architecture, or selecting computing chips from other partner suppliers according to needs. The Vertex AI platform also integrates many open source AI model resources, as well as the Gemini AI model and application tools provided by Google.
This means that users can build application services with greater flexibility on the Google Cloud platform, and also have greater options for deploying applications flexibly. Google itself even maintains a more open attitude, allowing users to more easily connect their services with services built on other cloud platforms, allowing users to build various application services in the Google Cloud service environment with greater peace of mind.
Therefore, compared with previous infrastructure design models, Mark Lohmeyer believes that there is actually not much difference. The main purpose is to add more options for the development of artificial intelligence computing applications, such as adding acceleration hardware such as TPU, and providing corresponding solutions for the data storage required for artificial intelligence training. However, even if the user builds a service that does not use artificial intelligence, it can still run stably in the Google Cloud environment.
Cloud infrastructure and services are not expected to be affected by tariff policies
As for the recent new tariff policy proposed by the Trump administration (Note), whether it may cause a certain impact or even prompt adjustments to the cloud infrastructure design, Mark Lohmeyer believes that it is not yet clear whether there will be an impact, so there will be no changes in the cloud infrastructure design. He also said that the infrastructure deployment of cloud services is somewhat different from the general introduction and sales of complete machines, so the actual impact is not expected to be too great.
Note:It was subsequently announced that the implementation would be suspended for 90 days, and the tariffs would be increased to 125% only on China.
