import { useState } from 'react' import { useNavigate } from 'react-router-dom' import { Button } from '@/components/ui/button' import { Badge } from '@/components/ui/badge' import { Card, CardContent } from '@/components/ui/card' import { Menu } from 'lucide-react' export function Technology() { const [activeTab, setActiveTab] = useState('Hardware') const navigate = useNavigate() return (
{/* Navigation */} {/* Header Section */}
{/* Title */}

Our Technology

Low-cost FPGA-based design, optimized for Large Language Models on-device.

{/* Tabs */}
{['Hardware', 'Software'].map((tab) => ( ))}
{/* Tab Content */}
{activeTab === 'Hardware' && (
{/* Hardware Text Content */}

Hardware Description

Huoxu is an ultra–low-budget single-board computer designed specifically for large language model (LLM) deployment. Despite its compact form factor and low power consumption, we provide:

  • Accelerated inference performance — delivering up to 15 tokens per second on a 30B model, thanks to its optimized FPGA-based AI engine.
  • High memory capacity — up to 40GB RAM, enabling hosting of LLMs up to ≥30B parameters.
  • Cost efficiency — making advanced AI workloads accessible at low cost.
{/* Hardware Specifications Card */}
Chip AMD Zynq UltraScale+ XCZU3EG
Processor
Quad-core 64-bit Arm Cortex-A53 CPU
Dual-core 64-bit Arm Cortex-R5 CPU
ARM Mali-400 GPU
FPGA Fabric 70K LUT, 360 DSP slices
Memory
8 GB 64-bit DDR4 (2400 Mbps) on CPU side
8 GB / 16 GB / 32 GB DDR4 (2133 Mbps) on FPGA side (SODIMM)
Storage
256 GB PCIe 2.0 x1 NVMe SSD
MicroSD card slot
I/O
USB 2.0 / 3.0 via Type-C
Mini DisplayPort
)} {activeTab === 'Software' && (
{/* Software Text Content */}

Software Stack

Our comprehensive software ecosystem is designed to maximize the potential of Huoxu hardware, providing seamless LLM deployment and management capabilities:

  • Custom FPGA-optimized inference engine with quantization support for efficient model deployment.
  • Edge LLM Platform — intuitive web interface for model management, chat, and system monitoring.
  • Support for popular model formats (GGUF, ONNX) and frameworks (Transformers, LLaMA.cpp).
  • REST API for seamless integration with existing applications and services.
{/* Software Features Card */}
OS Support
Ubuntu 22.04 LTS
Custom Linux Distribution
Model Formats
GGUF (LLaMA.cpp)
ONNX
PyTorch
Transformers
Inference Engine
Custom FPGA Accelerator
LLaMA.cpp Integration
Quantization Support
Management Interface
Web-based Dashboard
REST API
CLI Tools
Development Tools
Python SDK
Model Optimization Tools
Performance Profiler
Security
Secure Boot
Model Encryption
Access Control
)}
{/* Why us? Comparison Section */}

Why us?

{/* Comparison Table */}
{/* Table Header */}
Ours
Raspberry Pi 5
Jetson Orin Nano
{/* Table Rows */} {[ { label: 'Price', ours: '$199', pi: '$120', jetson: '$249' }, { label: 'RAM', ours: '24GB/40GB', pi: '16GB', jetson: '8GB' }, { label: 'CPU', ours: 'Cortex-A53', pi: 'Cortex-A72', jetson: 'Cortex-A78' }, { label: 'AI Engine', ours: 'Optimized Accelerator on FPGA', pi: 'Neon SIMD Instructions', jetson: 'Cuda/Tensor Core' }, { label: 'Power', ours: '<10W', pi: '5-12 W', jetson: '7-25' }, { label: 'LLM decode Performance', ours: '15 tokens/s', pi: '<5 tokens/s', jetson: '15 tokens/s' }, ].map((row, index) => (
{row.label}
{row.ours}
{row.pi}
{row.jetson}
))}
{/* Device Layout Section */}

The device layout plan

Device Layout Plan { console.error('Chips image failed to load') e.currentTarget.style.display = 'none' }} />

Need help? Contact us to get customized device and services

) }