November 2024 Summaries

DigitalOcean has announced significant enhancements to its Hatch program, specifically designed for AI and machine learning startups. The program offers exclusive benefits such as up to $100,000 in compute credits, three months of free GPU usage, and complimentary premium support. Additionally, DigitalOcean's GPU Droplets are now available at an industry-leading price of just $1.90 per hour, providing affordable access to powerful computing resources for startups. With Hatch, AI/ML startups can focus on innovation while scaling with confidence, leveraging a startup-friendly approach that includes predictable pricing, flexible resource allocation, and expert technical support. The program is designed to help founders get their companies off the ground and provide the necessary resources and community to succeed in the competitive AI/ML landscape.

DigitalOcean offers a range of GPU computing solutions to help developers and teams power their AI/ML workloads. The company's Bare Metal GPUs provide direct access to powerful hardware for high-performance AI/ML workloads, while its GPU Droplets offer scalable, on-demand deployment with virtualized instances. For those looking for an easy entry point into GPU computing, DigitalOcean's 1-Click Models powered by Hugging Face are a low-barrier solution. The company also offers fully managed Kubernetes clusters with GPU support through its GPUs for DOKS service, which is ideal for teams already using Kubernetes. Additionally, the GenAI Platform provides a quick way to get started with third-party generative AI models and agent customization, making it suitable for developers building generative AI applications like chatbots or search tools. By selecting the right option for their needs, users can leverage DigitalOcean's strengths in scalability, simplicity, and approachability to power their AI/ML workloads from development to production.

The DigitalOcean team has been using an open-source project called Cluster API to provide a managed Kubernetes service since 2019. The team's infrastructure underwent significant changes, particularly with the introduction of a new Containerized Control Plane (CCP) architecture in 2021. This change was made possible by combining Cluster API with "DigitalOcean's scrappy magic" to create a scalable and fault-tolerant managed Kubernetes service. By using Cluster API, DigitalOcean simplified the complexities of provisioning new Kubernetes clusters and providing structure for ongoing maintenance and upgrades. The team also leveraged Kubernetes in Kubernetes, which allowed them to take care of scalability and fault tolerance without additional logic or human intervention. This approach enabled flexibility for customers, making it possible to offer high availability and horizontal scaling with minimal effort. However, the team also faced challenges, such as networking complexities and the need to enforce a positive security model for multi-tenanted clusters. Ultimately, the change was successful, and DigitalOcean is excited to continue delivering more features in future, having taken a step back to develop a bold new approach.

DigitalOcean Bare Metal GPUs with NVIDIA accelerated computing offer a solution for serious AI builders who need control, power and less noise in their cloud infrastructure. These servers are purpose-built to tackle demanding AI/ML workloads, model training and custom infrastructure setups, featuring 8 NVIDIA Hopper GPUs and powerful hardware. They provide dedicated, single-tenant infrastructure with no neighbors, allowing full access to all GPUs, ideal for projects requiring direct control over hardware for optimal performance and privacy. This setup is well-suited for use cases like large-scale model training, real-time inference, and complex orchestration. With the quotes from Jacob Jackson, CEO and Founder of Supermaven, highlighting the importance of heavy-duty GPU computing power in delivering exceptional user experiences. The Bare Metal GPUs come with different configurations such as GPU Droplets which provide easy scalability and quick provisioning, but are limited by their lack of customization options compared to the Bare Metal servers.

Maintenance Mode and Restart Apps are two new features in DigitalOcean's App Platform, allowing for greater flexibility and control over applications. Maintenance Mode temporarily takes an app offline while ensuring users see a maintenance page on the home screen, ideal for tasks such as database migrations or troubleshooting. The Restart Apps feature enables developers to reboot their applications directly from the dashboard in seconds, simplifying app management and reducing unnecessary workarounds.

DigitalOcean's Virtual Private Cloud (VPC) is a private networking solution that provides isolation for resources running on the DigitalOcean public cloud on isolated private networks. It offers simplicity and ease of use, allowing users to create custom VPCs with their own address ranges without requiring prior networking knowledge. The VPC solution features a multi-tenanted fleet of bare metal hypervisors, where each production hypervisor hosts Droplets belonging to several different VPCs across many tenants. Isolation is achieved through VXLAN tunnels, which are encapsulated and forwarded using L2 constructs. The VPC control plane relies on MAC addresses for unique identification and uses OpenFlow specifications to program the OvS datapath. The system has a convergence framework that handles high throughput updates and is resilient to failures, with versioning used to track updates and ensure consistency across all targets.

At DigitalOcean, we embrace the ever-changing nature of the tech landscape and our data centers are no exception. We're thrilled to announce that NYC2, one of our first data centers launched in 2013, is once again open to new customers after temporarily closing due to capacity constraints. With expanded capacity and the launch of NVIDIA H100 GPUs at NYC2, we've ensured that our facility can support more customers, applications, and workloads than ever before. The reopening coincides with the introduction of GPU Droplets powered by these high-performance computing options, facilitating high-speed data processing for developers, startups, and innovators. Our 16 data centers are designed to be adaptable to businesses of all sizes, providing flexibility and fluidity in our infrastructure. We welcome new customers back to NYC2 and invite them to sign up or contact our sales team to learn more about how this location can support their business.

In a significant update to its cloud-native application platform, DigitalOcean has introduced foundational additions to its Kubernetes (DOKS) networking capabilities, enabling users to build and scale their applications globally. The new features aim to provide a more flexible, scalable, and secure networking foundation for cloud-native applications, addressing challenges faced by customers trying to scale beyond single regions. With the introduction of VPC-native clusters, peering, global load balancers, and internal load balancers, DigitalOcean is reimagining its DOKS networking paradigm to empower developers to build robust, resilient applications that can run globally with seamless application scalability and security.

DigitalOcean has announced the release of scalable storage for its Managed MongoDB service, allowing developers to add additional storage to their MongoDB clusters at cost-effective prices with minimal friction. This feature enables developers to minimize downtime, grow their applications as data needs change, and make their applications more flexible. Scalable storage provides independent scaling, cost efficiency, ease of use, granular billing, automatic provisioning, and expanded storage capacity for data-intensive applications built on MongoDB. The new feature is available with pricing starting at $0.20 per GiB per month, up to a maximum of 16 TiB of storage per database instance, with security features such as data encryption at rest and in transit.