The OrionX AI accelerator resource pool solution has achieved a revolutionary breakthrough in GPU resource pooling technology: it extends GPU resource pooling capabilities to the entire data center, decouples AI applications from GPU server hardware, and realizes dynamic scaling and flexible scheduling of vGPU resources.
The OrionX AI accelerator resource pool solution 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:
"Rounding to zero" GPU resource sharing. Users can share a physical GPU through multiple virtual machines or containers. Fine-grained segmentation allows each virtual machine and container to independently set and control the vGPU resources used by each virtual machine and container.
"Unify zeros" GPU resource aggregation allows users to provide GPU resources on multiple physical servers to a single virtual machine or container for use. Multi-machine multi-cards become single-machine multi-cards, simplifying deployment, and user AI application code does not need to be modified.
"Pickup items from the air" remote GPU resources usage, users can run GPU applications on a server without a GPU. Using remote GPUs on other servers via the network, they can achieve more than 95% of the local GPU performance on a 25G RDMA network.
"Respond as needed" Software-defined GPU resources that scale dynamically as needed. The GPU resources provided by OrionX are all based on software definitions, and the vGPU resources used by users can be flexibly dynamically scaled across the entire resource pool.