According to recent rumors, Google may involve AMD in the development of the 10th generation TPU.
Google may choose AMD participates in the development of the next ten generations of TPUaccording to recent rumors. If this hypothesis is confirmed, it would be an especially interesting collaboration considering that AMD will be participating in another company’s customized AI acceleration project.
this tensor processing unitDeveloped internally by Google and now several generations old, it is an important part of the company’s infrastructure for artificial intelligence workloads. So Google has a Long experience in designing these wafersbut may be interested in involving AMD to meet specific architectural needs.
Google, AMD, and the build of the 10th Gen TPU
According to the reconstruction, Google and some of its customers are evaluating TPU equipped with a Greater general processing powerby integrating CPU cores directly into the same accelerator package.
This type of configuration may be particularly beneficial Reinforcement Learning and Agent Applicationsareas where tensor operations must be supported by a large amount of traditional computing capabilities. There are already signs that Google is taking this approach more and more seriously.
Google Pixel 11: The series is updated with Tensor G6 and Gemini Intelligence
In 8i TPU-based systems, one Axion CPU is used for every two accelerators under various reconstructions, whereas in previous generations the ratio was one Xeon processor for every four TPUs.
In some cases, we will also consider a configuration with one CPU per accelerator. Placing CPUs and accelerators in the same package reduces the distance between different processing units, resulting in different benefits.
What impact will AMD’s cooperation with Google have?
AMD will have skills that are uniquely suited to the project, thanks to experience gained with the Instinct MI300A, a data center product that CPU x86, GPU and HBM memory within the same package. Therefore, this technology could also provide Google with useful skills for possible future TPUs.
However, it’s unclear what AMD’s specific contribution will be. If the rumors are true, the most important aspect may be the evolution of TPU into architecture More CPU component orienteddesigned to better manage reinforcement learning, reinforcement processing, and new systems based on artificial intelligence agents.