Powering open infrastructure for autonomous agents
cloud.google.com·3d ago
Cloud & Infrastructure
Google Cloud Blog·2d ago·5 min read
Introducing Z4M: storage-optimized compute for AI/ML and parallel file systems
TL;DR
Google Cloud has launched the Z4M machine series, designed for storage-optimized AI/ML workloads, offering significant improvements in data processing speed and capacity.
✦ Why It Matters
Consider migrating your AI/ML workloads to Z4M for enhanced performance and cost efficiency.
Key Takeaways
- Utilize Z4M for high-capacity AI/ML workloads.
- Leverage Cloud RDMA for reduced latency in data access.
- Optimize storage performance with Hyperdisk options.
- Scale infrastructure efficiently with Z4M's VM and bare-metal instances.
Full Summary
The Z4M machine series from Google Cloud is tailored for AI and machine learning workloads that require high storage capacity and fast data processing. It features up to 168,000 GiB of Local SSD (LSSD) storage and utilizes fifth-generation Intel Xeon Scalable processors.
The integration of 400 Gbps Cloud Remote Direct Memory Access (RDMA) networking reduces data-access latency by up to 60%, enhancing throughput for large datasets. Z4M instances can deliver up to 15.6 million random-read IOPS and 75,600 MiB/s sequential-read throughput, marking improvements of up to 70% and 110% over the previous Z3 series.
Additionally, Z4M supports Hyperdisk, allowing for optimized storage performance tailored to specific workloads. This infrastructure is particularly beneficial for applications requiring real-time data processing and high availability.