Google Cloud debuts G4 VMs optimized for large-scale AI inference

Google Cloud has introduced its new G4 virtual machines powered by the NVIDIA RTX PRO 6000 Blackwell Server edition, making it the first cloud provider to offer this combination. This new generation is aimed at users and companies who need a lot of computing power for things like artificial intelligence, advanced graphics, game development, and large-scale simulations. G4 VMs bring together eight of NVIDIA’s latest GPUs, high-core-count AMD Turin CPUs, and dedicated Google networking hardware, resulting in major improvements over previous versions.

Compared to earlier G2 VMs, the G4 offers four times the computing and memory power, and six times the memory bandwidth. The RTX PRO 6000 Blackwell GPUs are equipped with upgraded components designed for modern AI tasks, like the new Tensor Cores and improved Transformer Engine, which handle tasks with higher precision and speed. The CPUs can support up to 384 virtual cores and over a terabyte of memory, while Google’s Titanium hardware boosts network speeds up to 400 Gbps.

These machines are designed to handle a variety of demanding jobs — from generating images and video with AI, to running real-time physics simulations, to next-level video game rendering. Major customers, including Altair, Ansys, AppLovin, Snap, WPP, and Nuro, are already planning to use G4 VMs to speed up engineering, content creation, and simulation workloads.

The G4 VM setup can be used with popular industry tools and supports 768 GB of fast GDDR7 GPU memory and 12 TB of local SSD storage. The preview is available now, with worldwide availability coming by the end of the year.

This new launch comes from Google Cloud, a major division of Google focused on cloud infrastructure, AI, and enterprise computing. The addition of the G4 VMs extends Google Cloud’s GPU offerings for demanding customers and reflects its ongoing push to lead in cloud-based AI and simulation services.

Source