Playing a big bet in the market of AI chips, Google has introduced its two new custom chips. These chips are named TPU 8tx and TPU 8i Max. Both are part of the company’s eighth generation TPU family. This move by Google seems to be a direct attack on NVIDIA’s areas of strength. NVIDIA is currently the biggest player in the market of AI chips, but Google is giving options to customers through these chips.
First let us understand what these chips are. TPU i.e. Tensor Processing Unit are Google’s own made chips, which are specially designed to train and run AI models. Compared to normal processors, these AI tasks prove to be faster and more economical.
What is the use of the new TPU 8tx chip?
TPU 8tx has been specially designed keeping security and privacy in mind. It has the facility of data encryption at the hardware level. That means sensitive data and model information remain safe even while working. These chips work together with AMD processors.
This is extremely useful for government agencies, banks, healthcare companies or any organization that wants to train or run AI models without sending out their private data.
NVIDIA’s chips also have security features, but Google has made it stronger and simpler. It attacks a big pain point of NVIDIA – data security.
What is the use of the new TPU 8i Max chip?
The second chip, TPU 8i Max, solves the memory problem. In this, two chips have been combined into one package, which has almost doubled the memory, processing power and bandwidth.
Today’s AI agents and large models have to handle a lot of context i.e. information simultaneously. Like understanding an entire codebase or analyzing a long video. When memory is low, work becomes slow. TPU 8i Max has been created to break this ‘memory wall’.
It will be beneficial in jobs like coding assistant, mathematical reasoning and drug discovery. Even NVIDIA’s GPUs are sometimes hampered by memory constraints, and Google has hit on that very point.
How will this bet work for Google?
This step of Google is necessary because the company wants to reduce its dependence on NVIDIA. She also wants to strengthen her cloud business. Google Cloud’s earnings have already increased significantly and TPU sales are also contributing to this. Now about 100 customers are using these Google chips.
In simple language, Google saw that there were two big needs in the market – on one hand, data security and on the other hand, big tasks requiring more memory. To fulfill these needs, it has launched these two special chips. With this, companies have another strong option apart from NVIDIA. In the world of AI, the war of hardware has now intensified. These new chips of Google show that the company is presenting a big challenge not only in software but also in hardware.
