Skip to content

Cnuas Datasheets

Product-style datasheets for the Cnuas Virtual AI HPC Infrastructure platform and its components. Each datasheet summarises the emulated device's features, functional specifications, interfaces, and ordering information in a single reference page.

Emulated hardware

Cnuas is a 100% software-emulated virtual AI and HPC infrastructure. Every "device" below is a QEMU/userspace model, not physical silicon. Specification values describe the emulated device as presented to guest software (drivers, verbs, runtimes), so a virtual GPU or NIC behaves like the real part it models. Performance figures are functional/relative, not absolute hardware throughput.

Datasheet library

Datasheet Part Description
Cnuas Platform Platform Full "datacenter in a box": switches, NICs, GPUs, fabric, host requirements
HiSwitch hicain-vswitchd 10-port virtual top-of-rack RoCEv2 + InfiniBand switch with in-switch Subnet Manager
HiNIC hicain-vnic RoCEv2 + native InfiniBand RDMA virtual NIC (kernel + verbs provider)
HiGPU higpu SIMT virtual GPU accelerator with HiIR ISA and CUDA-style runtime
HiLink higpu-link-switchd GPU-to-GPU fabric interconnect (NVLink-class model)

For feature availability and planned work across all components, see the Product Roadmap.

How to read these datasheets

Each datasheet follows a consistent layout:

  1. Overview, one-paragraph description and key features.
  2. Functional specifications, tabulated parameters (ports, protocols, limits).
  3. Interfaces, how the device is presented and controlled.
  4. Software support, drivers, runtimes, tools, and guest-side stack.
  5. Ordering / integration information, package name, versions, dependencies.
  6. Revision history.

These datasheets are summaries. For architecture, protocol walk-throughs, and implementation detail see the Design and Reference sections.