Google Cloud Unveils A4 VMs with NVIDIA's Blackwell GPUs

In a significant leap forward, Google Cloud has announced the preview of its A4 VMs, powered by NVIDIA's cutting-edge Blackwell B200 GPUs. These VMs are set to revolutionize AI model training and fine-tuning with an impressive performance boost, featuring eight interconnected GPUs. They offer 2.25 times the compute power and HBM capacity compared to their predecessors, the A3 VMs.

This new cloud infrastructure is designed to enhance not only AI model training but also fine-tuning and low-latency serving. With Google's infrastructure innovations such as enhanced networking with Titanium ML adapters, Google Kubernetes Engine support, and access through Vertex AI, the A4 VMs are poised to set new benchmarks in cloud computing.

Gerard Bernabeu Altayo, Compute Lead at Hudson River Trading, expressed enthusiasm about leveraging the Hypercompute Cluster's innovations to accelerate the deployment of training their latest models, which are aimed at delivering quant-based algorithmic trading.