Accelerating Edge Intelligence
High-performance APUs and logic processors built for massive data throughput, vector search, and real-time AI inference.
Advantages of APU: Compute-in-Memory
Overcome the "AI bottleneck" by processing directly where data is stored. Achieve millisecond-level vector search latency.
3-second time to first token
Deliver industry-leading time to first token for edge multimodal large language model inference, without relying on cloud latency.
30W subsystem power consumption
Leverage a compute-in-memory architecture to eliminate the data movement bottleneck and achieve large-scale AI acceleration at a fraction of the power consumption of traditional GPU clusters.
Billion-scale vector search
Optimized for high-throughput similarity search and retrieval-augmented generation, instantly query billions of data points.
AI Computing Products
Leda-E PCIe Accelerator Card
Designed for edge inference and workstation integration. Brings the massively parallel processing power of the Gemini-II APU to local environments.
KEY SPECIFICATIONS
Form Factor: PCIe Gen4 x16- Compute Cores: 1x Gemini-II APU
- Power Consumption: < 50W (subsystem)
- Memory: High-Bandwidth 3D SRAM
TARGET APPLICATIONS
- Edge AI Inference Server
- High-Frequency Algorithmic Trading
- Localized LLM Deployment for Secure Environments
- Real-Time Video Analytics
Leda-S 1U Edge Server
High-density compute node for enterprise data centers. Accommodates up to 16 APUs in a standard rack-mount chassis.
KEY SPECIFICATIONS
- Form Factor: 1U Rack-Mount
- Compute Cores: Up to 16x APU
- Host Processor: Dual-Socket AMD EPYC™
- Networking: Dual-Port 100GbE QSFP28
TARGET APPLICATIONS
- Enterprise Vector Database Acceleration
- Retrieval-Augmented Generation (RAG) Pipelines
- Smart City and Intelligent Traffic Management
- Telecom Edge High-Density Compute Node
Leda-E 2U Enterprise Server
Delivers maximum APU density for the most demanding workloads, built for mission-critical data centers.
KEY SPECIFICATIONS
- Form Factor: 2U Rack-Mount
- Compute Core: 1-8x Leda-E PCIe Cards
- Host Processor: Dual Intel® Xeon® Scalable Processors
- Power Supply: Redundant Titanium-grade (1+1)
TARGET APPLICATIONS
- Large-scale LLM Fine-tuning and Training
- Billion-scale Similarity Search Engine
- Bioinformatics and Molecular Search
- Synthetic Aperture Radar (SAR) Processing
Secure Your Supply Chain Today
Due to market fluctuations, our chip agency services are priced per project.