What’s new in Data & AI: September 2024
Google Cloud continues to innovate in data and AI, pushing the boundaries of analytics and artificial intelligence. Devoteam’s Google Cloud experts highlight what’s new since the previous months – focusing on advancing BigQuery’s capabilities, enhancing AI models, and optimising database solutions. Let’s explore the most significant updates.
AI Innovations with Gemini and Imagen
The Gemini 1.5 series has received notable upgrades, featuring 2 model variants:
- Gemini 1.5 Pro: A high-performance model offering multimodal capabilities and extended context windows (up to 2M tokens). Suitable for complex, large-scale tasks where high-quality outputs are essential.
- Gemini 1.5 Flash: A low-latency, cost-effective option that provides similar quality to the Pro version on most common tasks but is optimised for speed.
Further complementing the Gemini 1.5 series is Google’s latest image generation model, Imagen 3, which:
- Offers superior quality, aesthetics, and prompt adherence
- Has a faster generation speed of four images in less than 9 seconds
- Includes a digital watermarking feature for enhanced safety and compliance.
Efficiency Meets Savings: Context Caching and Grounding
Google Cloud’s context caching is a game-changer, allowing customers to drastically reduce costs by up to 75% by efficiently managing large context windows. With the ability to reuse fixed contexts for subsequent queries, enterprises can optimise performance while minimising expenses.
Additionally, grounding mechanisms with Gemini models enable more factual responses by dynamically integrating relevant context, which is crucial for maintaining accuracy in AI-powered applications.
Imagine you have a large dataset of research papers and need to answer various questions about them. By processing the dataset once and caching the key information, you can store that knowledge and quickly retrieve answers from the same cache context, saving significant time and resources.
Tristan Van Thielen
ML Tribe Lead Google Cloud Business Unit at Devoteam
Gemma 2: Elevating Open AI Models for Enterprises
Gemma 2 is a cutting-edge series of open models built on the technology of the Gemini family. It comes in two larger sizes, 9B and 27B, and can be easily deployed using Vertex AI or Google Kubernetes Engine (GKE). Pre-built starter notebooks will soon be available. The Gemma Cookbook offers practical guides for customising and fine-tuning models, and Gemma 2 is now available on Hugging Face and Vertex AI Model Garden.
BigQuery and Data Analytics: Continuous Real-Time Analytics
BigQuery’s continuous real-time analytics feature introduces unbounded SQL processing for real-time anomaly detection and dynamic data enrichment. It enables developers to build event-driven applications that respond as soon as data arrives.
- Partition Enhancements: Maximum partition count increased from 4,000 to 10,000, enabling BigQuery to handle up to 27 years of day-partitioned data
- AlloyDB Federated Queries: Now supports querying AlloyDB tables directly from BigQuery, offering a unified analytics experience across data silos.
AlloyDB: Bringing AI to PostgreSQL
AlloyDB is tailored for operational and analytical workloads. It supports vector search and embeddings using the ScaNN algorithm, delivering up to 4x faster search performance than standard PostgreSQL.
- AlloyDB Omni: Supports on-premises and hybrid deployments across different environments, maintaining PostgreSQL compatibility.
- Free Trial Cluster: Now available, with up to 1TB of storage and an 8 vCPU instance, this cluster allows businesses to experiment before committing to larger-scale implementations.
Expanded Capabilities in Cloud Storage and Cloud SQL
Google Cloud Storage has introduced the following features:
- Hierarchical Namespaces: Provides true folder-based operations for seamless file management in Cloud Storage. Ideal for AI and machine learning workloads.
Google CloudSQL has introduced the following features:
- PostgreSQL 16 support
- Point-in-time recovery
- IAM group authentication, making database management and security more robust and flexible.
Spanner, Google’s globally distributed database, has also received significant upgrades such as:
- Spanner Graph: Adds native graph database capabilities for complex relationship modelling and analysis.
- Geo-Partitioning: Improves data locality, reducing latency for globally distributed datasets. Ideal for scenarios requiring compliance with data residency requirements.
Wrapping Up: Another Step Forward
With powerful AI models, advanced analytics, and high-performance databases, Google Cloud’s latest offerings enable businesses to build intelligent, real-time applications. The focus on context management, grounding, and scalable solutions reflects Google’s commitment to equipping enterprises for next-generation digital transformation.
Watch a recap of our online session to explore the full range of updates and learn how Google Cloud sets new data and AI innovation standards.
Want to talk about how these updates benefit you?
Contact our certified experts to guide your Google Cloud Data & AI projects.