Google Cloud has launched the general availability of its next-generation Storage-optimized Z4M machine series, available as both virtual machines (VMs) and bare-metal instances. Designed for AI/ML training and inference workloads, the Z4M series offers up to 168,000 GiB of Local SSD (LSSD) storage and 400 Gbps of Cloud Remote Direct Memory Access (RDMA) networking, reducing data-access latency by up to 60% compared to TCP networking.
The Z4M VMs and bare-metal instances are powered by fifth-generation Intel Xeon Scalable processors and feature Titanium SSDs, which enhance storage performance and security. They deliver up to 15.6M random-read IOPS and 75,600 MiB/s sequential-read throughput, improving performance by up to 70% and 110% compared to the previous Z3 series. The series is optimized for workloads such as checkpointing, distributed key-value cache, parallel file systems, and enterprise-class databases.
Customers and partners have praised the Z4M series for its exceptional performance and scalability. Nutanix highlighted the combination of Z4M’s I/O performance and high storage density, while Silk noted up to 3.7x lower latency and 6x lower CPU usage with Cloud RDMA. Sycomp and WEKA also emphasized the series' ability to handle demanding AI and machine learning workloads.
Google Cloud’s Cluster Director allows for easy configuration, deployment, and management of large Z4M clusters. The series also supports Hyperdisk, offering scalable, high-performance network-attached storage with up to 512 TiB of capacity per instance. Z4M VMs and bare-metal instances are now available in select regions worldwide.