AI Inference Workstations
Junhangclaw series · 50 TOPS to 2070 TFLOPS · personal to enterprise
The Junhangclaw personal super-workstation series spans four tiers, from a 50 TOPS domestic-silicon entry model to the flagship JH-W2000 with an NVIDIA Blackwell GPU delivering 2070 TFLOPS (FP4 sparse) — built for local AI inference where private data never leaves the premises.
Product Models
Specification Comparison
| Model | Type | Interface | Config | Key Specs |
|---|---|---|---|---|
| JH-W50 Entry | 50 TOPS | Entry | 50 TOPS | Domestic silicon · 50 TOPS · personal users |
| JH-W120 Enhanced | 120 TOPS | Enhanced | 120 TOPS | 120 TOPS · small teams |
| JH-W280 Professional | 280 TOPS | Professional | 280 TOPS INT8 | MetaX domestic silicon · 280 TOPS INT8 · localization-ready |
| JH-W2000 | 2070 TFLOPS | Flagship compact | 128GB · 1TB | Blackwell GPU · 2070 TFLOPS FP4 sparse · 128GB unified memory |
| JH-W2000N Flagship | 2070 TFLOPS | Flagship expanded | 128GB · 1TB+4bay | Blackwell GPU · 2070 TFLOPS · 4× 2.5-inch bay expansion |
Specifications as of ; see each model page and the official product guide for details.
Frequently Asked Questions
Which models and compute tiers are available?
Five models: JH-W50 entry at 50 TOPS, JH-W120 enhanced at 120 TOPS, JH-W280 professional at 280 TOPS (INT8), and the flagship JH-W2000 and JH-W2000N at 2070 TFLOPS (FP4 sparse).
What is in the flagship JH-W2000?
NVIDIA Blackwell GPU, 14-core Arm Neoverse, up to 128GB LPDDR5x unified memory, 1TB NVMe, 140mm radiator liquid cooling, up to 4× 25GbE and a 300W external adapter; the JH-W2000N adds a four-bay 2.5-inch storage module.
Who are these workstations for?
JH-W50 for personal users, JH-W120 for small teams, JH-W280 (MetaX domestic silicon) for government and enterprise localization, and JH-W2000/W2000N for LLM inference, enterprise agents and knowledge bases, all for local inference where private data stays on-premises.
Other Product Lines
Need Product Selection or Sample Testing?
Our engineering team provides selection advice, compatibility validation, and customized solutions.
Contact Us