AI compute
AI Compute Appliance
Ready-to-use local AI compute. Hardware, drivers and inference stack come preinstalled, and your data stays in-house. Run it in your own server room or host it in a HyperSource data center.
Why an appliance
Data never leaves your premises
Models and data run locally, which suits industries with strict data security requirements.
Ready out of the box
Drivers and inference stack are preinstalled at the factory, so you can deploy models on arrival.
Predictable cost
For long-term use, total cost is more predictable than renting cloud GPUs by the hour.
Flexible deployment
Place it on your premises, or host it in HyperSource's Beijing, Guangzhou, Dongguan or Guiyang data centers with managed operations.
Model comparison
The datasheet is the authoritative source for full specifications.
| Spec | Pro 6000 | Pro 6000D |
|---|---|---|
| GPU | NVIDIA RTX PRO 6000 (Blackwell architecture) | NVIDIA RTX PRO 6000D (Blackwell architecture, China edition) |
| Memory per GPU | 96GB GDDR7 ECC | 84GB GDDR7 ECC |
| Low-precision inference | FP4 / FP8 supported | FP4 / FP8 supported |
| Number of GPUs | From 1 GPU, multi-GPU available | From 1 GPU, multi-GPU available |
| System configuration | CPU, memory, storage and networking configured to your needs; see the quotation | CPU, memory, storage and networking configured to your needs; see the quotation |
| Preinstalled software | AI Inference Software + TokenHub | AI Inference Software + TokenHub |
| Best for | LLM inference, LoRA fine-tuning, running multiple models in parallel | LLM inference, LoRA fine-tuning, procurement through domestic channels |
Full specifications and datasheets are available from our consultants on request.
Preinstalled software
- GPU drivers and runtime environment
- AI Inference Software: model deployment, inference serving, scheduling and monitoring
- TokenHub: unified access to local and public models, RAG knowledge base, account and usage management
- Optional preinstalled open-source LLMs (specify when ordering)
Use cases
Pair the appliance with our AI application development service and get working business applications delivered.
Two deployment options
On your premises
Install the appliance in your own server room or office. We provide on-site installation and setup.
Hosted in a HyperSource data center
Host the appliance in a HyperSource data center (Guiyang offers 15–40kW high-power cabinets) with stable power, bandwidth and 24/7 operations.
Warranty and support
Each system comes with a hardware warranty and remote technical support. The warranty period and on-site service scope are as specified in the contract. Preinstalled software includes version upgrades and technical support. Appliances hosted in HyperSource data centers are inspected and serviced by our data center engineers.
FAQ
Pro 6000 or Pro 6000D: which should I choose?
Both use the same GPU architecture and ship with the same software. The main differences are memory and supply channel: the Pro 6000 has 96GB per GPU, while the Pro 6000D is the China edition with 84GB per GPU. Choose the Pro 6000 for maximum memory per GPU, or the Pro 6000D if domestic procurement and lead times matter more. Our consultants will recommend one based on your model size and budget.
Which LLMs can it run?
All mainstream open-source LLMs can be deployed. The model size and concurrency you can run depend on the memory configuration. Our consultants will assess your needs, or you can test it at our data center.
What is the lead time?
Lead time depends on the configuration and stock at the time of order. Our consultants will confirm it with the quotation.
Can I try it first?
Yes. Request a trial or test it at our data center. Submit your requirements and a consultant will arrange it.
Related
Request a quote or trial
A consultant will reply during business hours (China Standard Time) with site, configuration and pricing recommendations.
- leo@hs.com.cn
- 24/7 technical support