OrionX

OrionX's AI accelerator resource pool solution has achieved a revolutionary breakthrough in GPU resource pooling technology: expanding GPU resource pooling capabilities to the entire data center, decoupling AI applications from GPU server hardware, and enabling dynamic scaling and flexible scheduling of vGPU resources. OrionX AI accelerator resource pool solutions can help customers easily build data center-level AI accelerator resource pools. AI applications can transparently use AI accelerators on any server in the data center without modification, improve resource utilization, and easily deploy AI applications. Through OrionX platform software, 4 major functional scenarios can be realized: GPU resource sharing is “reduced to zero”. Users can share a physical GPU through multiple virtual machines or containers, and the vGPU resources used by each virtual machine and container can be independently set and controlled. By “turning zero into a whole” GPU resource aggregation, users can provide GPU resources from multiple physical servers to a single virtual machine or container for use, turning multiple computers into multiple cards on a single machine to simplify deployment, and the user AI application code does not need to be modified. Using “AirFetch” remote GPU resources, users can run GPU applications on a server without a GPU, use remote GPUs on other servers through the network, and achieve more than 95% local GPU performance on a 25G RDMA network. Software-defined GPU resources that scale dynamically “on demand”. All GPU resources provided by OrionX are based on software definitions, and the vGPU resources used by users can be flexibly dynamically scaled across the entire resource pool.