DigitalOcean Holdings has announced a collaboration with AMD, making AMD Instinct GPUs available to DigitalOcean customers through GPU Droplets optimized for artificial intelligence and high-performance computing (HPC) workloads. Initially launched with AMD Instinct MI300X GPUs, DigitalOcean plans to expand GPU Droplet offerings later in 2023 with AMD’s new MI325X GPUs, which AMD claims will set new performance standards for AI training and inference tasks.
AMD Instinct MI300X GPUs provide substantial memory capacity, allowing models with hundreds of billions of parameters to be loaded entirely into memory, thus reducing the complexity and overhead related to distributing models across multiple GPUs. These GPUs leverage AMD’s ROCm software platform, engineered for efficient development and rapid deployment of production-ready HPC and AI systems.
Previously available only as bare metal configurations requiring greater configuration efforts, these AMD GPUs are now accessible on DigitalOcean as GPU Droplets or as Kubernetes worker nodes. GPU Droplet configurations support either single-GPU or eight-GPU setups, allowing users choice and flexibility in balancing performance with operational costs. DigitalOcean reports competitive pricing for these GPUs at $1.99 per GPU per hour, alongside enterprise-grade Service Level Agreements (SLAs) and simplified one-click deployment protocols.
In addition to GPU Droplet availability, customers will be able to access the AMD Developer Cloud, a new managed AI development platform powered by DigitalOcean infrastructure. This platform supports rapid prototyping, large language model (LLM) fine-tuning, benchmarking inference workloads, and building scalable inference infrastructure without upfront hardware expenses or complex local setup.
“With access to AMD GPUs, DigitalOcean customers have an extensive portfolio of GPUs with the flexibility of the computing configuration that best suits their requirements,” said Bratin Saha, Chief Product & Technology Officer at DigitalOcean.
“At AMD, we are proud to work with DigitalOcean to provide developers with cutting-edge solutions for developer enablement and demanding workloads that require large amounts of memory,” said Negin Oliver, corporate vice president of business development, Data Center GPU Business, at AMD. “Together, AMD and DigitalOcean are committed to providing the critical innovative technologies required to support the evolving needs of growing tech businesses.”
Source: DigitalOcean Holdings
