MTT KUAE

MTT KUAE is Moore Threads Full-Stack Solution for AI Data Centers. It based on MTT S5000 GPU and the dual-processor 8-GPU server MTT SGX5000. This integrated solution tackles the challenges inherent in deploying large-scale GPU computing power with efficiency and effectiveness.

Quick Deployment

Cluster construction in 30 days

Best Practices

Optimization of computing power, storage, and networking

Ready to Use

Complete suite of intuitive tools and software stack

Unmatched Performance

Distributed training for trillion-parameter models

Ready-to-Use Integrated Solution

MTT KUAE full-stack solution is built on Moore Threads' Universal GPUs and seamlessly integrates hardware and software. MTT KUAE Platform for cluster management and MTT KUAE Model Studio for accessing model services fully support MTT KUAE to achieve maximum optimization. This end-to-end solution greatly simplifies the deployment and operation of large-scale GPU computational infrastructure.

Core Features

MTT KUAE full-stack solution fully leverages the advantages of Moore Threads GPUs.

Product Portfolio

MTT KUAE Core Components

MTT KUAE Platform

Beyond standard Kubernetes cluster capabilities, MTT KUAE Platform introduces a range of innovations purpose-built for AI computing scenario.

Deeply integrates full-stack GPU computing, networking, and storage capabilities, with centralized GPU driver management to reduce deployment complexity and operational costs
Provides multi-dimensional isolation for different organizations and users through enterprise workspaces and projects
Supports GPU sharing and incorporates built-in best practices for multi-GPU-aware scheduling to improve resource utilization and maximize performance
Delivers a unified observability platform for physical nodes, storage, networking, cluster components, and workloads, accelerating issue diagnosis while reducing troubleshooting costs
Deeply integrates application and infrastructure telemetry, enabling proactive issue detection through diagnostic management and fine-grained monitoring and alerting

MTT KUAE ModelStudio

Provides an end-to-end platform for AI model development and training, with support for a wide range of mainstream open-source models.

Model Development

Launch ready-to-use development environments (VS Code & Jupyter) with a single click, preconfigured with required dependencies and mounted datasets to accelerate development
Support multiple development workspaces with persistent storage, providing isolated development environments for different projects

Model Training

Support for mainstream distributed training frameworks, with rapid fault detection and checkpoint recovery in under 10 minutes
Innovative training insights, 3D parallelism visualization for rapid identification of slow-performing nodes, and built-in operator profiling for large-scale training optimization

Key Features of MTT KUAE

Modular design of large-scale GPU computing power with flexible deployment

Optimization of the linear speedup ratio of GPU computing power

Deployment of a high-speed parameter transmission network

Deployment and scheduling of heterogeneous computing clusters

Design and deployment of a computational power service support system

Scheduling of elastic computing power for cloud-native GPU clusters

Reliability and security of computing and storage

Highly reliable automatic problem diagnosis and recovery