TPU Container Instance Introduction

TPU container instances integrate self-developed or cooperative TPU accelerator cards on standard container cloud platforms, providing high-performance computing power while ensuring on-demand start-stop and elastic scaling, suitable for high-density computing scenarios such as large model inference, training, and scientific computing.

In addition to TPU container instances, DLC Cloud Platform also provides higher-end computing power such as Bare Metal Service. If you need a customized solution, please Contact Us.

Features

• Pay-as-you-go: Open as you use, billed by the minute. No fees are incurred after calculation stops.
• Multi-model Support: Pre-set mainstream model environments such as Llama 3, Qwen, Mistral, Stable Diffusion, GPT-J, etc., and equipped with TensorFlow, PyTorch, JAX, and cuDNN, CUDA dependencies.
• Comprehensive Storage Solution: Provides free quota for system disks, supports high IOPS data disks, reducing overall costs.

Product Billing

Supports two methods: pay-as-you-go and subscription (not yet available), billing separately for calculation, system disk, and cloud storage. Refer to Compute Market for the latest prices.

Product Usage

Create, start, stop, and manage TPU container instances through the console. Combine with image repositories, networks, and security policies to quickly deploy inference or training tasks.