AI Inference Workstations
JH-W2000
The flagship desktop AI workstation: NVIDIA Blackwell GPU, 14-core Arm Neoverse, 128GB LPDDR5x unified memory and 2070 TFLOPS (FP4 sparse) with 140mm liquid cooling, in a 199.3×199.3×100.3mm chassis.
2070 TFLOPS128GB · 1TBFlagship compactBlackwell GPU · 2070 TFLOPS FP4 sparse · 128GB unified memory
Applications:LLM InferenceEnterprise Agents & Knowledge BasesPersonal AI Workstation
Specifications
| Architecture / GPU | NVIDIA Blackwell GPU |
|---|---|
| Processors | 14-core Arm Neoverse |
| CUDA Cores | 2560 |
| Tensor Cores | 96 fifth-generation Tensor Cores |
| Memory | Up to 128GB LPDDR5x unified memory |
| Storage | 1TB NVMe |
| Cooling | 140mm radiator liquid cooling |
| Networking | 1× 2.5GbE; up to 4× 25GbE (QSFP28) |
| AI Performance | 2070 TFLOPS (FP4 sparse) |
| I/O Ports | 1× full-function USB4.0 20G, 3× USB3.2 Type-A, 4× USB2.0 Type-A; 1× HDMI 2.1, 1× DP1.4a (HBR3, MST), 1× USB4.0 |
| Power Supply | External 300W / 20V DC adapter |
| Dimensions (L×W×H) | 199.3 × 199.3 × 100.3 mm |
* Configurations are indicative; final specs follow the project BOM, delivery specification and measured results.
Special Functions
140mm radiator liquid coolingSupported
Full-function USB4.0 20GSupported
Up to 4× 25GbE (QSFP28)Supported
LPDDR5x unified memorySupported
Product Features
Four compute tiers from 50 TOPS to 2070 TFLOPS (FP4 sparse)
Domestic silicon and NVIDIA Blackwell tracks
Local AI inference — private data stays on-premises
Flagship 128GB LPDDR5x unified memory and 25GbE networking
Desktop liquid cooling (140mm radiator on the flagship)
Interested in JH-W2000?
Contact us for the full datasheet, sample testing, and volume pricing.
Contact Us