Product positioning: Targeting the training of billions of parameter large models, this flagship training server is designed for ultra large scale AI pre training, scientific research supercomputing, and national level computing power centers. It is the core computing power engine for large model research and development.

core strengths

Top GPU computing power configuration:

Equipped with 8 x NVIDIA H100/H800 high-end GPUs, NVLink high-speed fully interconnected architecture, seamless data transmission between multiple cards, fully unleashing GPU parallel computing power, and perfectly supporting the full process training of billions and trillions of parameter large models

Super floating point computing power output:

Single machine FP16 computing power exceeds 5 PFLOPS, significantly shortening the pre training and fine-tuning cycle of large models, accelerating the development and implementation of AI models, and helping enterprises seize the opportunity of AI technology and business

High bandwidth and high-speed storage:

Equipped with up to 2TB DDR5 high-speed server memory and 8 x NVMe SSD high-speed solid-state storage, providing TB level read and write bandwidth to meet the high-speed interaction needs of large models and massive data, avoiding computing power bottlenecks

High speed cluster interconnection capability: equipped with 100GbE/2GBE high-speed optical ports, supporting RDMA and RoCE-v2 high-speed interconnection protocols, low latency and high bandwidth multi node cluster communication, supporting efficient collaboration of ultra large scale computing power clusters

Professional liquid cooling and heat dissipation guarantee:

Customized cold plate liquid cooling system, accurately removes GPU core heat, with PUE as low as 1.15, ensuring stable release of GPU performance under high load, eliminating frequency reduction, and significantly reducing energy consumption