Lambda

Lambda is a cloud computing service tuned for AI developers, with on-demand and reserved NVIDIA GPU instances and clusters for training and inference. The service offers a variety of GPUs, including the new NVIDIA H100, H200 and GH200 Tensor Core GPUs, so customers can train and fine-tune big AI models.

Lambda's on-demand cloud lets customers provision GPU instances by the hour without a long-term commitment. That's good for projects that need computing resources that can wax and wane. For customers who need to use a lot of GPUs for months at a time, Lambda also offers reserved cloud options.

Among its features:

• On-Demand GPU Clusters: Scalable clusters of NVIDIA H100 Tensor Core GPUs with Quantum-2 InfiniBand.
• Multi-GPU Instances: Train and fine-tune AI models with a variety of instance sizes with 1x, 2x, 4x and 8x NVIDIA GPU options.
• Pre-Configured for ML: Ubuntu, TensorFlow, PyTorch, NVIDIA CUDA and NVIDIA cuDNN are pre-installed for fast project setup.
• One-Click Jupyter Access: Connect directly to NVIDIA GPU instances from your browser for easy project development.
• Scalable Filesystem: Create and manage storage to persist files and data with your compute, billed by usage.
• Pay-by-the-Second Billing: Pay only for the time instances are running to minimize costs.
• Simple, Transparent Pricing: No surprise costs, with pricing based on hourly usage.

On-demand instance pricing varies depending on the GPU and configuration. For example:

• 1x NVIDIA H100 PCIe: $2.49/hour
• 8x NVIDIA H100 SXM: $27.92/hour
• 1x NVIDIA A100 PCIe: $1.29/hour
• 1x NVIDIA RTX A6000: $0.80/hour

Reserved cloud pricing is available for longer-term commitments, with options for one, two and three-year terms. For example:

• NVIDIA H100: $1.89/hour/H100 (3-year term)
• NVIDIA H200: Pricing available upon request (3-year term)
• NVIDIA GH200: $5.99/hour/GH200 (3-12 month term)

Developers and researchers will benefit from Lambda's ML-first user experience, which lets them quickly provision and manage GPU instances that meet their project needs. Thousands of AI engineers and researchers already use Lambda, and the company says they appreciate its ease of use and performance.

Published on July 10, 2024