Google Cloud Next announced new instance types and accelerators, focusing on AI accelerators like custom Arm-based Axion chips and Nvidia-built accelerators. The upcoming high-performance Nvidia HGX B200 and GB200 NBL72 servers will support AI and HPC workloads from early 2025, with the latter liquid-cooled.
Google introduced the A3 Mega instance, jointly developed with Nvidia, to cater to developers needing more power for training large language models (LLMs). They also unveiled the A3 confidential instance, enhancing data security during AI workloads by encrypting data transfers between Intel's CPU and Nvidia H100 GPU without requiring code changes.
Google launched Cloud TPU v5p processors at Google Cloud Next, marking a significant advancement in homegrown AI accelerators. These processors offer a substantial increase in floating point operations per second and memory bandwidth speed.
[
add
]
[
|
|
...
]