This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Read our privacy policy>

Atlas Intelligent Cloud Hardware Platform

Atlas is a new-generation intelligent cloud hardware platform built for emerging services such as public cloud, Artificial Intelligence (AI), High-Performance Computing (HPC), and video. It includes Huawei FusionServer G series heterogeneous servers, heterogeneous cards, and pooling orchestration software. Atlas addresses the requirements of deep learning, training, and inference scenarios. It takes the industry lead by supporting GPU physical resource disaggregation, pooling, and on-demand orchestration on the offline computing platform, helping customers improve resource utilization and move more confidently towards the AI era.

Technical highlights

Physical GPU resource pooling, on-demand orchestration:

  • Dynamically allocate GPU card resources on demand, improving resource utilization by 50%
  • Dynamically mount and dismount orchestration software on the client to and from GPUs, implementing dynamic adjustment within seconds
  • Support automatic GPU failover in case of system faults, eliminating the need for re-deployment

Efficient remote deployment with performance loss below 10%:

  • Deployed based on Time-Division Multiplexing (TDM)-like and GPUDirect RDMA technologies, remote GPU resources deliver over 90% efficiency, with low performance loss

Compatible with existing ecosystem, no need to change applications:

  • Provide interfaces that are compatible with standard framework interfaces, with no need to change the application software already running on local GPUs

Get big benefits with enterprise ICT solutions from Huawei

On-demand GPU allocation, improving resource utilization by up to 50%

Leverage a TDM-like technology to reuse GPU resources, improving resource utilization.

GPU capabilities accessible to old-version servers

Leverage the GPU resource pooling technology to provide GPU capabilities to old-version servers.

Independent evolution of GPU resources

Disaggregate GPU from other resources, and support independent GPU evolution.