Anyscale Unveils New Products and Enhances its Leading Unified AI Platform at Ray Summit 2024

Expanded offerings empower organizations to build and scale AI strategies more efficiently in the rapidly evolving market


SAN FRANCISCO, Oct. 01, 2024 (GLOBE NEWSWIRE) -- Anyscale, the unified AI platform and the company behind the leading open-source AI Compute Engine Ray, today kicked off its annual user conference, Ray Summit. The three-day event, held annually in San Francisco, brings together leading experts from OpenAI, a16z, Meta, and Runway, among others, to discuss the latest innovations in the development of AI at scale. Anyscale announced a slew of significant platform advancements, including a foundational shift in architecture that makes Ray GPU-native; RayTurbo, an optimized version of Ray for peak performance; the general availability release of Ray Data, a library for processing unstructured data at scale; new direct integrations connecting the Anyscale platform to Kubernetes for AI/ML development; and tools for enterprise governance and observability, all wrapped in a new user experience that reduces complexity at scale. Together, these upgrades cement Ray and Anyscale’s standing as the go-to distributed compute solution for AI workloads.

The growing sophistication of AI development – with large models, multimodal data, and multi-model orchestration – make the need to scale seamlessly and efficiently more critical than ever. Poorly managed AI workloads result in wasted resources, increased costs and delayed production, limiting the full potential of AI.

“As more and more organizations integrate AI into day-to-day operations, we’ve seen Anyscale usage soar 44x over the past year, and Ray is now orchestrating more than 1 million clusters per month,” said Robert Nishihara, Co-founder of Anyscale. “With the introduction of RayTurbo and Ray’s GPU-native architecture, we are offering the only holistic compute engine on the market serving AI developers.”

Today’s announcements include:

  • Ray is now GPU-native: Ray’s dynamic and flexible structure make it ideal for AI/ML workloads, but those traits also result in higher overhead for smaller tasks across GPUs. By innovating on the default execution through static task graphs, Ray is now truly GPU-native, setting a new standard in speed, low latency, and high performance in inter-GPU communication for workloads like distributed training and model serving.
  • RayTurbo: Available exclusively on the Anyscale platform, the new RayTurbo engine is a supercharged version of Ray that delivers significant performance improvements, maximizing resource utilization and minimizing additional costs. Performance improvements include up to 4.5X faster data processing, up to 90% lower costs through reliable spot instance support, up to 5X faster node launching and scaling, and 60% higher QPS serving when compared to open-source Ray.
  • Support for unstructured data: With the general availability of Ray Data, Anyscale’s powerful solution for unstructured data workloads and distributed data processing, all users will have access to newly supported data formats including data lakehouse formats like Hudi, Iceberg, and Delta Lake.
  • Anyscale Operator for Kubernetes: Launching in partnership with Amazon EKS, Google GKE, Azure AKS, and OCI Kubernetes Engine (OKE), the Anyscale Operator for Kubernetes allows all users deploying Ray through KubeRay to seamlessly use Anyscale and RayTurbo. The Anyscale Operator runs as part of customers' existing Kubernetes clusters, letting customers run their tooling, their governance systems, and their integrations alongside other workloads.
  • Tools for enterprise governance and observability: Anyscale now provides platform engineers and IT leaders with enterprise governance and observability at both the infrastructure and application level. This includes functionality to manage compute resource usage via quota management, workload prioritization across teams, and deep observability to enable developers to optimize AI/ML workflows.
  • Ease to scale for real AI benefits: Anyscale is simplifying the onboarding process to reduce complexity via an upgraded user experience which includes:
    • Log Viewer: Persistent log storage, improved search/filter functions, and cloud export notifications for seamless debugging and monitoring.
    • Streamlined Dependency Management: Simplified installation process, consolidated information, and optimized container images for a faster, easier setup.
    • Serverless Mode: Auto-selection of compute nodes helps automate infrastructure management, enabling developers to focus on applications.

With Ray’s rapid growth and widespread adoption by companies like Netflix, OpenAI, Airbnb, and others, Anyscale continues to break new ground in enabling organizations to scale their AI and ML workloads with unprecedented efficiency.

At Ray Summit, over 1,500 attendees will hear from Anyscale customers and Ray users from Meta, Uber, Canva, Pinterest, Spotify, and many more, learning how they are leveraging Ray to power their AI strategies. The opening day of the conference offers keynotes from Anyscale co-founders, Robert Nishihara and Ion Stoica, as well as Anastasis Germanidis, Co-founder & CTO, Runway, and legendary venture capitalist Marc Andreessen, Co-founder & general partner, Andreessen Horowitz. On the following day, keynote speakers will include Kevin Weil, CPO, OpenAI; John Bicket: CTO & Co-founder, Samsara; Brandon Leonardo, the Co-founder of Instacart; and Stephen MacKinnon, vice president of applied machine learning for Recursion.

You can read more about Anyscale’s platform advancements on the blog www.anyscale.com/blog.

About Anyscale

Anyscale provides a Unified AI Platform powered by Ray, enabling organizations to scale their AI and ML workloads efficiently across any infrastructure. Anyscale's platform powers everything from training large language models to real-time data processing, helping teams of all sizes build, deploy, and manage AI workloads effortlessly.

For more information, visit www.anyscale.com.

Media Contact:
press@anyscale.com