DigitalOcean Holdings Inc.

11/21/2024 | Press release | Archived content

DigitalOcean Bare Metal GPUs: Dedicated GPU machines for advanced AI workloads

For serious AI builders who need more control, more power, and less noise in your cloud infrastructure, it's time to explore DigitalOcean Bare Metal GPUs with NVIDIA accelerated computing. These servers are purpose-built to tackle the most demanding AI/ML workloads, model training, and custom infrastructure setups. With 8 NVIDIA Hopper GPUs and powerful hardware, these servers offer high-performance compute capabilities and customization options designed for intense processing needs.

DigitalOcean Bare Metal GPUs with NVIDIA accelerated computing offer dedicated, single-tenant infrastructure with no neighbors, so you have full access to all your GPUs. These servers are tailored for projects that require direct control over hardware to achieve optimal performance and privacy, and are well-suited for use cases like large-scale model training, real-time inference, and complex orchestration. With dedicated hardware you get direct access to all resources, ideal for pushing the limits of what your applications can achieve.

"The most important thing for our company is to continue delivering a really exceptional user experience and continuing to make that even better. So that's going to involve research and development for the models, which requires heavy duty GPU computing power that DigitalOcean provides."

Jacob Jackson, CEO and Founder, Supermaven

  1. Model training at scale: Tackle large datasets with ease. Bare Metal GPUs allow you to fully optimize model training, offering privacy, performance, and control over the entire process.
  2. Model fine-tuning: Start with a pre-trained model and adapt it to your specific use cases by fine-tuning with specialized datasets. Bare Metal's high-performance GPUs and isolated environment make it possible to get the accuracy and performance needed for specialized applications.
  3. High-speed inference: With Hopper GPUs handling inference, your applications can achieve real-time performance for live predictions and decision-making-ideal for user-facing products that demand responsiveness.
  4. Custom use cases: Bare Metal GPUs are ideal for custom workloads that demand high configurability and dependable, dedicated infrastructure. Whether deploying Kubernetes clusters or custom orchestration setups, these servers provide the flexibility needed to build and scale complex architectures.

Our Bare Metal GPUs and GPU Droplets are designed to suit different workloads. GPU Droplets provide easy scalability and quick provisioning, perfect for teams focused on training, fine-tuning, or running inference on LLMs. In contrast, Bare Metal GPUs deliver maximum performance and control, ideal for sustained, high-throughput workloads that demand direct access to hardware resources and customization.

Bare Metal GPU Droplet H100 GPU Droplet H100x8
diGPU Memory 640 GB 80 GB 640 GB
(v)CPUs 192 20 160
CPU Type 2 Intel® Xeon® Platinum 8468 2x Intel Xeon Platinum 8458P processors 2x Intel Xeon Platinum 8458P processors
Network Cards 8x Mellanox Network Adapter; Mellanox Technologies MT2910 Family [ConnectX-7]; link speed 400 Gbps, 4x Mellanox Technologies MT2892 Family [ConnectX-6 Dx]; link speed 100Gps Nvidia/Mellanox ConnectX-6 2x100GE NIC Nvidia/Mellanox ConnectX-6 2x100GE NIC
Machine Memory 2.0 TiB 240 GiB 1920 GiB
Local Storage (Boot) 56 TB 720 GiB NVMe 2 TiB NVMe
Local Storage (Scratch) - 5 TiB NVMe 40 TiB NVMe

Bare Metal GPUs with NVIDIA accelerated computing are available in New York, USA and Amsterdam, Netherlands, with more data centers coming soon.

Ready to get started, or want to learn more about pricing? Contact our experts for more information and personalized guidance on how we can meet your specific workload requirements.