Google Cloud has launched new A4 virtual machines in preview powered by NVIDIA's Blackwell B200 GPUs to meet the growing demands of advanced AI workloads. The A4 VM features eight interconnected Blackwell GPUs with a 2.25x increase in peak compute and HBM capacity compared to A3 High VMs. Key features include enhanced networking, GKE integration, Vertex AI accessibility, open software optimization, a hypercompute cluster, and flexible consumption models.
Thomas Kurian, CEO of Google Cloud, announced the launch on X. The A4 VMs use Google's Titanium ML network adapter and NVIDIA ConnectX-7 NICs for high GPU-to-GPU traffic. The Jupiter network fabric supports large-scale GPU scaling. Native integration with GKE facilitates a robust AI platform. Google is collaborating with NVIDIA to optimize JAX and XLA.
A new hypercompute cluster system simplifies the deployment and management of large-scale AI workloads. Flexible consumption models offer optimized AI workload consumption. Sai Ruhul highlighted analyst estimates that Blackwell GPUs could be 10-100x faster for large transformer model workloads. Naeem Aslam tweeted that Google's integration could enhance computational power and boost NVIDIA's position. This release gives developers access to the latest NVIDIA Blackwell GPUs within Google Cloud's infrastructure for improved AI application performance.