Accelerate your AI/ML workloads with DigitalOcean's high-performance computing offerings
AI and machine learning tasks need the right infrastructure to run smoothly and help you build the tech products that will define your industry. Access GPU-powered high-performance computing at DigitalOcean.
Accelerate your software applications with on-demand HPC resources
High-performance computing (HPC) is a computing system that is designed to support data-intensive, complex calculations at high speeds. These setups are often massive configurations of compute servers, storage, and networking infrastructure. They also require orchestration software and a message passing interface (MPI) to send instructions between computing nodes.
Over time, HPC has evolved into a few different form factors:
-
On-premises: All the computing hardware and infrastructure is contained onsite via a data center or purpose-built HPC.
-
HPC Cloud: Running high-performance workloads in the cloud, which makes it easy to scale high-intensity workloads on-demand via cloud resources.
-
HPC as a Service: Access to pre-configured HPC resources through a managed cloud provider or hyperscaler, such as DigitalOcean.
All of these options enable you to access HPC power and gain benefits such as increased processing speed, high-volume data ingestion, scalability and flexibility, long-term cost efficiency, and increased accuracy for complex dataset applications.
Using GPUs for HPC
Because CPUs don’t have capabilities for parallel processing, they often create bottlenecks when trying to support compute-intensive use cases. Graphics processing units (GPUs), on the other hand, can support multiple simultaneous workloads and are equipped with more memory and storage. This is why GPUs are the default processing unit for HPC applications. You can have two types of HPC configurations that use GPUs: cluster and distributed.
A cluster setup supports parallel computing with a collection of servers connected through high-speed networking. This setup minimizes latency between nodes, centralizes management, and provides high availability.
A distributed HPC configuration connects multiple computers within the same network. This option increases overall scalability, increases output via concurrency, and decreases fault tolerance.
Main HPC use cases
HPC is primarily used for use cases that require lots of parallel data processing and need hardware that has immense amounts of memory and storage for large datasets. This includes:
AI and machine learning
HPC provides the massive computational resources, data ingestion, and processing power for complex AI applications. This enables efficient model training, large-scale data handling, faster inference speeds, multi-modal AI compute, and scalability for parallel processing across distributed systems.
Research
Running climate models, quantum mechanics, genomics, and physics simulations requires intensive computational capabilities and specialized hardware to support complex data sets. HPC facilitates big data analysis, handles model complexity, enables physical process simulation, and supports collaboration between research institutions working on computationally demanding projects.
Simulation
HPC enables parallel computation to create realistic simulations by breaking complex models into smaller parts that can be processed simultaneously. This approach supports time compression for faster results, boosts simulation model realism, and enables interactive design workflows that would be challenging with traditional computing resources.
Benefits of DigitalOcean GPUs for HPC
Easy to use
With DigitalOcean GPU Droplets and Bare Metal GPUs, you can easily access the computing power you need. GPU Droplets are available with just a few clicks in our New York and Toronto data centers. Our Bare Metal GPU hardware is deployed in our New York and Amsterdam regions.
Cost effective
Transparent pricing is a core value at DigitalOcean, and our GPU offerings are no exception. View our pricing guides for GPU Droplets and Bare Metal GPUs.
Performant
Backed by DigitalOcean’s enterprise-grade SLAs, we maintain high availability for your HPC workloads and provide the processing speeds, memory capacity, and overall efficiency that you need for your AI workloads.
Connected ecosystem
DigitalOcean’s GPU products integrate with our wider product ecosystem, including DigitalOcean Kubernetes (DOKS), App Platform, Spaces object storage, and more, giving you access to a complete solution for AI/ML workloads.
Learn more about DigitalOcean AI/ML infrastructure
You can harness HPC power with DigitalOcean’s GPU Droplets and Bare Metal GPUs. Our GPU Droplets offer easy, preconfigured computing power you can spin up with just a few clicks. Or you can reserve Bare Metal GPUs for more customizable AI application deployments.
NVIDIA H100
Optimize costs and run flexible workloads on powerful virtualized GPU hardware.
-
Specifically designed to support HPC and large language model training workloads.
-
Can also support computer vision, speech AI, RAG, and conversational AI applications.
-
Available in New York, US, Atlanta, US, and Toronto, Canada, data centers.
Coming soon: NVIDIA H200
Designed for generative AI and high-performance computing workloads.
- More energy efficient and provides a lower cost of ownership, and nearly double the memory capacity than the NVIDIA H100.
AMD MI325X
Brings high memory capacity to support models with hundreds of billions of parameters.
-
Supports large model training, fine-tuning, inference, and HPC use cases
-
Reduces the need for model splitting across multiple GPUs
-
Has 256GB of HBM3E memory and 6.0TB/s of bandwidth.
AMD MI300X
Delivers leadership performance for accelerated HPC applications and the newly exploding demands of generative AI.
-
Large memory capacity supports models with hundreds of billions of parameters.
-
Optimized for large-scale AI and HPC workloads such as training foundational models, AI research, and model fine-tuning.
-
Available starting at $1.99/hr per GPU
Bare Metal GPUs
You can have full control over your HPC deployment with standalone machines or a multi-node cluster setup to build out your applications.
-
Available with the NVIDIA H100, NVIDIA H200, or AMD MI300X GPUs.
-
Supports AI/ML model training, model fine-tuning, model inference, and custom AI and HPC workloads.
-
No additional bandwidth charges for cost-effective Bare Metal GPU deployments and 1-2 day provisioning.
HPC FAQs
How does HPC use GPUs?
GPUs serve as the processing hardware for HPC systems, allowing users to run data-intensive parallel workloads for use cases like AI, research, and simulation. Traditional CPUs cannot provide adequate processing power for HPC applications.
How do GPU Droplets differ from Bare Metal GPUs for HPC?
GPU Droplets provide users with pre-configured high-performance computing power with minimal setup, streamlined management, and intuitive deployment. Bare Metal GPUs are standalone machines in DigitalOcean data centers. They allow users to create completely custom deployments with full control over the hardware and software.
Can I scale my HPC GPU workloads on DigitalOcean?
Your product selection will determine how you scale your HPC workloads with DigitalOcean. With GPU Droplets, you can add more and scale up to 8-GPU configurations as needed. Bare Metal GPU configurations require time to reserve and provision, but you can expand the number you use over time.
Which AI and ML frameworks are supported on DigitalOcean’s HPC GPU infrastructure?
Our 1-Click Marketplace supports a variety of AI/ML frameworks for GPU Droplets. Our infrastructure can also run frameworks such as PyTorch, TensorFlow, OpenCV, Keras, Scikit-learn, Hugging Face, and OpenAI.
Can I run GPU clusters for distributed HPC computing?
Yes, you can run multiple GPU clusters within one HPC cluster for distributed computing applications. However, it does require technical expertise and software to coordinate between GPU clusters.
How does DigitalOcean’s HPC GPU solution compare to hyperscalers?
Compared to other hyperscalers like AWS, Google Cloud, and Microsoft Azure, DigitalOcean’s HPC GPU solutions can be more cost-effective, scalable, and connected to an integrated ecosystem. Additionally, the GPU Droplet offering makes it easy to get production-ready infrastructure online in just a few clicks.