HuaKun AI infrastructure: workstation, server or cluster?
Choose between local inference, PCIe accelerators, modular platforms and clusters using memory, software fit, concurrency and a pilot.
ExploreIT & TELECOM / SOFCON AZERBAIJAN
HuaKun AI servers
PCIe accelerator servers and distinct modular inference and training platforms. SOFCONS defines the configuration around the model, accelerator memory, software stack and concurrent workload.
Prepare an enquiry
SOFCON AZERBAIJANRequirements / solution / delivery
Models and platforms
| Model | Applications | Key features |
|---|---|---|
| HuaKun AT800 (Model 3000)Manufacturer details ↗ | Kunpeng server for central inference with PCIe accelerator cards. |
|
| HuaKun AT225 Inference ServerManufacturer details ↗ | AT225: DDR4-based PCIe inference platform. |
|
| HuaKun AT225 B2Manufacturer details ↗ | AT225 B2 inference server for Atlas 300I Duo cards. |
|
| HuaKun AT525 B2Manufacturer details ↗ | AI platform combining DDR5 host memory with PCIe accelerators. |
|
| HuaKun AT9508 G3Manufacturer details ↗ | PCIe accelerator platform also listed as AT9508 G3 AI Server. |
|
| HuaKun AT3500 G3Manufacturer details ↗ | Ascend 910 modular inference server, also associated with the AT548 Inference name in official documentation. |
|
| HuaKun AT548 Inference ServerManufacturer details ↗ | AT548 inference configuration with A2G4 modules, linked to the AT3500 G3 name. |
|
| HuaKun AT800 (Model 9000) A2Manufacturer details ↗ | Modular training platform with the overseas name AT548 Training. |
|
| HuaKun AT548 Training ServerManufacturer details ↗ | A2G3 training configuration linked to the AT800 (Model 9000) A2 name. |
|
Selection guides
IT & Telecom01 02 03
From requirements to delivery
PCIe inference
AT225 and AT225 B2 are distinct models. AT225 uses DDR4; AT225 B2 supports up to six Atlas 300I Duo cards. For larger PCIe configurations, compare AT525 B2 and AT9508 G3 alongside card compatibility, host memory and networking needs.
Inference and training
Official documents link AT548 Inference with AT3500 G3 and AT548 Training with AT800 (Model 9000) A2. Select the module generation from the exact model documentation. Inference prioritizes latency and concurrent requests; training also requires data pipelines, checkpoints and collective communication.
Measure the application in a pilot
Prepare a test with the exact model revision, precision, context length and user profile. Include the accelerator runtime, operators, containers and observability tools in the configuration; agree acceptance under the real workload.
Share the model and service profile to compare PCIe and modular platforms through a relevant pilot.