Heterogeneous Computing

Core Technologies
for Agentic Infrastructure

Heterogeneous Computing

Dramatically Expanding Chip Compatibility

Software-Hardware Co-Optimization

Unlocking Maximum Chip Performance

Self-Adaptive Agent

AI-native Infrastructure Powered by Intelligent Agents

Heterogeneous Computing

Unified orchestration across diverse chip architectures enables efficient integration and
coordination of heterogeneous compute resources.

Performance Breakthrough: BI-V150 1k-Card-Cluster MoE Model Trained Jointly by Infinigence and Iluvatar

DETAILS

Infinigence Conducts Training of MT-infini-3B Large Model on MTT KUAE k-card Intelligent Computing Cluster

DETAILS

97.6% Peak Utilization: How This k-card Heterogeneous Hybrid Training System Was Developed

DETAILS

Software-Hardware Co-Optimization

Unlocking full potential of chip application through in-depth optimization and
co-design across software and hardware.

ASP-DAC 2025 Best Paper: ViDA Sparse Accelerator Overcomes VDiT Bottleneck

DETAILS

FPGA 2025 Best Paper: First Video Generation Inference IP, HW-SW Co-Design Supercharges Compute

DETAILS

Semi-P/D: Evolution from P/D Fusion to Semi-P/D, A New LM Inference Paradigm

DETAILS

Self-Adaptive Agent

Environmental awareness, decision-making, and self-evolving capabilities enabled
using AI-powered infrastructure agent swarms

Infinigence Unveils Infra Agent Swarm: Redefining Agentic Infra & Agent Production

DETAILS

RLinf Goes Open Source: The First Large-Scale RL Framework Built for Embodied Agents

DETAILS

From "Dialogue" to "Telepathy": Cache-to-Cache, a New Paradigm for Agent Communication

DETAILS

Technical Insights Column

Tech for Infinite Possibilities

All

Software–hardware Collaboration

Heterogeneous Computing

Self-Adaptive Agent

Heterogeneous Foundation

Training

Inference

Agents

Embodied AI

AI-Native Enterprises

Intelligent Manufacturing

AIGC

Intelligent Devices

Heterogeneous Computing

Software-Hardware Co-Optimization

Self-Adaptive Agent

News & Updates

Media Coverage

Core Technologies for Agentic Infrastructure

Heterogeneous Computing

Software-Hardware Co-Optimization

Self-Adaptive Agent

Heterogeneous Computing

Performance Breakthrough: BI-V150 1k-Card-Cluster MoE Model Trained Jointly by Infinigence and Iluvatar

Infinigence Conducts Training of MT-infini-3B Large Model on MTT KUAE k-card Intelligent Computing Cluster

97.6% Peak Utilization: How This k-card Heterogeneous Hybrid Training System Was Developed

Software-Hardware Co-Optimization

ASP-DAC 2025 Best Paper: ViDA Sparse Accelerator Overcomes VDiT Bottleneck

FPGA 2025 Best Paper: First Video Generation Inference IP, HW-SW Co-Design Supercharges Compute

Semi-P/D: Evolution from P/D Fusion to Semi-P/D, A New LM Inference Paradigm

Self-Adaptive Agent

Infinigence Unveils Infra Agent Swarm: Redefining Agentic Infra & Agent Production

RLinf Goes Open Source: The First Large-Scale RL Framework Built for Embodied Agents

From "Dialogue" to "Telepathy": Cache-to-Cache, a New Paradigm for Agent Communication

Technical Insights Column

Performance Breakthrough: BI-V150 1k-Card-Cluster MoE Model Trained Jointly by Infinigence and Iluvatar

Wang Yu: Charting the Next Era of Edge-Side AI

Infinigence Conducts Training of MT-infini-3B Large Model on MTT KUAE k-card Intelligent Computing Cluster

FPGA 2025 Best Paper: First Video Generation Inference IP, HW-SW Co-Design Supercharges Compute

Performance Breakthrough: BI-V150 1k-Card-Cluster MoE Model Trained Jointly by Infinigence and Iluvatar

Infinigence Training Engine: Almost 2× Efficiency for Birentech 1k-Card Cluster

97.6% Peak Utilization: How This k-card Heterogeneous Hybrid Training System Was Developed

Wang Yu: What Are the Essential Paths to Improve LM Energy Efficiency?

Wang Yu: Computing, Cost and Ecosystem Hinder Sustainable Development of LM Industry

Infinite Computing, Accelerating the Future of AGI

Core Technologies
for Agentic Infrastructure