During the Google Cloud Next ’24 event in Las Vegas, Google Cloud (GCP) introduced a series of new features, most notably the Arm-based custom Axion chip and AI accelerators.

In terms of AI acceleration, Google will work with Nvidia to integrate the Nvidia Blackwell platform expected to launch in 2025, including the super-fast HGX B200 GPU and NBL72 GB200, the GPU ideal for performance-demanding AI tasks. as high as training large language models (LLM).

For current LLM training needs, Google introduces A3 Mega – a combination of standard H100 GPU and new networking, providing double the bandwidth per GPU. Another variant is A3 Confidential which prioritizes the security of sensitive data, encrypting data end-to-end during transmission between Intel CPUs.

Google also introduced Cloud TPU v5p, the most powerful AI-accelerated processor developed by the company, with twice the floating-point computing performance and three times the memory bandwidth compared to the previous generation. previous system.

Besides chips, Google also offers AI-optimized storage options like Hyperdisk ML, which can improve model loading times by up to 3.7 times.

Additionally, Google will also introduce new general-purpose C4 and N4 instances, powered by Intel's 4th and 5th generation Xeon processors, along with C3 bare metal machines, bare metal versions. X4 memory optimized and Z3 virtual machine optimized for storage.

The innovations at Google Cloud Next ’24 demonstrate Google Cloud’s commitment to providing customers with the most powerful, flexible and efficient cloud computing solutions.