Google Cloud’s first Arm-based CPU for the data centre, a host of new compute and storage services that dramatically improve generative AI performance, a security-centric Chrome offering, and a flurry of enterprise-focused Workspace updates that take the fight to Microsoft 365.
Also, AI in everything, including Gemini and Vertex AI in data warehouse BigQuery (with fine tuning) in public preview, for "seamless preparation and analysis of multimodal data such as documents, audio and video files." (nb: Vector search came to Big Query in preview in February.)
Those were among the updates set to get serious airtime at Google Cloud NEXT in Las Vegas this week. The Stack will share more considered analysis about some of the news coming through in coming days, along with interviews with executives and customers – but here’s an early sample from a blockbuster set of press releases, GitHub repositories and blogs...
Google Workspace
“Unlike traditional email and productivity solutions, Gmail and Workspace were built from the very beginning on a cloud-native architecture, rooted in zero-trust principles, and augmented with AI-powered threat defenses.”
So said Google pointedly in the wake of the CSRB’s blistering indictment of Microsoft’s security, which noted pointedly that Redmond had “designed its consumer MSA identity infrastructure more than 20 years ago.”)
Workspace, Google’s suite of collaboration and productivity applications, has approximately 10 million paying users. That makes it a minnow compared to the 300 million+ paid seats Office 365 boasted back in 2022.
It could be more of a threat to Microsoft.
A series of new features unveiled today may make it one. They include a new $10/user AI Security add-on that will let Workspace admins “automatically classify and protect sensitive files and data using privacy-preserving AI models and Data Loss Prevention [DLP] controls trained for their organization” – a Google spokesperson told The Stack that “we’re extending DLP controls and classification labels to Gmail in beta.”
Pressed for detail, they told us that these will include:
- “Simplified deployment and data protection policies management with rules for Gmail, Drive and Google Chat unified into the same area.
- “Advanced detection policies... wide selection of predefined detectors for global and regional information types, custom detectors" [Regex and word lists] that can be pointed at header, subject, body...
- “Granular configuration of policies with scoping sender audiences (at domain, OU, and group levels) and recipient audiences (internal, external, both).
- “Actions with various levels of restriction such as block delivery of message, quarantine message for review, and log event for future audit.
- “Tools for incident management and investigation such as Alert Center and Security Investigation Tool with detailed logs and dashboards.”
Also coming soon: “Experimental support for post-quantum cryptography (PQC) in client-side encryption [with partners] Thales and Fortanix…”
Google Vids?
A new generative AI service called Google Vids baked into Google Workspace may get more headlines. That’s a “video, writing, production, and editing assistant” that will work in-browser and sit alongside Docs, Sheets, and Slides from June. Less of a serious competitor for Premier Pro and more a templating assistant that “pieces together your first draft with suggested scenes from stock videos, images, and background music.” (The Stack has clarified that users can also upload their own video, not just use stock...)
Other Workspace updates today:
Chat: Increased member capacity of up to 500,000 in Spaces for those bigger enterprise customers. Also new: GA messaging interoperability with Slack and Teams through Google-funded Mio, and various AI integrations and enhancements across Docs, Sheets etc.
Storage and compute
NVIDIA CEO Jensen Huang anticipates over $1 trillion in data center spending over the next four years as infrastructure is heavily upgraded for more generative AI-centric workloads. This isn’t just a case of plumbing in more GPUs – Google Cloud is showcasing some real innovations here.
It boasted “significant enhancements at every layer of our AI Hypercomputer architecture [including] performance-optimized hardware, open software and frameworks…”
Top of the list and hot off the press:
- New Tensor Processing Units: Google Cloud’s new TPU accelerator for AI and ML, the v5p is GA. A single v5p “pod” contains 8,960 chips that run in unison, with 2x higher FLOPS and 3X more high-bandwidth memory on a per chip basis than its v4 TPUs. Application developer Lightricks said it had seen a “2.5X speed-up in training our text-to-image and text-to-video models compared to TPU v4.”
- Google’s first custom Arm-based CPU for the data centre, “Axion” – already used internally to power BigTable, Spanner, BigQuery, Pub/Sub, Google Earth Engine, and the YouTube Ads platform in Google Cloud. We’ve seen big claims on speed and efficiency ("30% better performance than the fastest general-purpose Arm-based instances available in the cloud today; 60% better energy efficiency"). Stay tuned for a separate piece. Google Cloud is promising Axion instances “later” this year.
Various other promises of faster, cheaper compute also abound. But it’s storage and caching where GCP’s R&D work really shines. (Important for generative AI because it is a HUGE bottleneck for most models.)
A standout is the preview release of Hyperdisk, a block storage service optimised for AI inference/serving workloads that Google Cloud says “accelerates model load times up to 12X compared to common alternatives”, with read-only, multi-attach, and thin provisioning.
Hyperdisk lets uses spin up 2,500 instances to access the same volume and delivers up to 1.2 TiB/s of aggregate throughput per volume: “Over 100X greater performance than Microsoft Azure Ultra SSD and Amazon EBS io2 BlockExpress” – in short its volumes are heavily optimised and managed network storage devices located independently from VMs, so users can detach or move Hyperdisk volumes to keep data, even after deleting VMs.
“Hyperdisk performance is decoupled from size, so you can dynamically update the performance, resize your existing Hyperdisk volumes or add more Hyperdisk volumes to a VM to meet your performance and storage space requirements” Google boasts, although there are some limitations…
Other storage/caching updates:
- Cloud Storage FUSE (GA) is a file-based interface for Google Cloud Storage (GCS) that gives file access to “high-performance, low-cost cloud storage”. New today: GCS FUSE caching that Google claims “ improves training throughput by 2.9X and improves serving performance for one of our own foundation models by 2.2X.”
- Parallelstore, a high performance parallel filesystem optimised for AI/ML and HPC workloads, also has new caching capabilities that “enable up to 3.9X faster training times and up to 3.7X higher training throughput, compared to native ML framework data loaders.”
- Filestore (GA): A network file system-based approach that allows all GPUs and TPUs within a cluster to “simultaneously access the same data, which improves training times by up to 56%, to optimize the performance of your AI workloads…”
Chrome Enterprise Premium?
Chrome Enterprise Premium is a turbocharged version ofChrome Enterprise with new....
- "Enterprise controls enforce policies, manage software updates and extensions to align with enterprise policies, and support RDP, SCP, SSH and other TCP protocols;
- "Security insights and reporting support event reporting, device reporting, and forensic capabilities for enterprise-wide visibility;
- Context-aware access controls can be scaled for web applications, can help enforce continuous Zero Trust access to SaaS and web-based apps with context-aware access control, and can mitigate data exfiltration risks for sanctioned and unsanctioned applications; and
- "Threat and data protection delivers content inspection and data loss prevention, anti-malware, and anti-phishing using frontline intelligence and AI, dynamic URL filtering, and site categorization.
Yes, we agree, this sounds rather good too.
More details and pricing in a standalone piece soon.
