Mthreads provides a domestically produced AI compute platform that combines proprietary full‑function GPUs with integrated software tools for training, inference, rendering, and video processing. Their product line includes server‑grade accelerators, AI modules, and GPU virtualization solutions, enabling Chinese enterprises and cloud providers to build and manage large GPU farms without relying on foreign components.
Funding
Funding not disclosed
BGFounders
Product
Problem
Enterprises and developers in China face limited access to domestically produced, high‑performance GPU hardware and integrated software stacks needed for large‑scale AI model training, inference, high‑resolution rendering, and video processing. Existing solutions often rely on foreign components, leading to supply chain constraints, higher costs, and compatibility challenges with local ecosystems.
Solution
Mthreads delivers a complete, end‑to‑end AI compute platform built around its proprietary full‑function GPU architecture. The portfolio includes server‑grade accelerators (e.g., MTT SGX5000), AI‑focused modules (MTT E300), and a range of GPU‑based products for rendering, professional visualization, and digital office use. Complementary software—MUSA SDK, Moore Perf System, and AI training/inference suites—provides drivers, performance monitoring, and deployment tools. Cloud‑native capabilities such as GPU virtualization (vGPU) and the KUAE cloud‑native suite enable flexible scaling from on‑premise clusters to virtual desktop infrastructure. Together, hardware and software allow customers to build, operate, and manage large GPU farms for AI, rendering, and video workloads without relying on foreign components.
Target Audience
Primary customers are Chinese enterprises, cloud service providers, and OEMs that require domestic GPU hardware for AI model training/inference, high‑end rendering, video processing, and virtual desktop infrastructure.
Features
- Proprietary full‑function GPU chips delivering high compute density for AI training (up to trillion‑parameter models) and inference.
- Integrated server (MTT SGX5000) and AI module (MTT E300) designs that simplify deployment and reduce system‑level integration effort.
- MUSA SDK and Moore Perf System provide unified drivers, performance profiling, and deployment pipelines for AI and graphics workloads.
- GPU virtualization (vGPU) and KUAE cloud‑native suite support multi‑tenant cloud desktops and scalable GPU sharing across workloads.
- Specialized product lines for high‑performance rendering (MTT S80/S70), professional visual acceleration (MTT X300), and digital office (MTT S30/S10).
- End‑to‑end AI software stacks (training and inference suites) with optimized kernels for domestic AI frameworks.
- Compatibility with major graphics APIs and support for 3D reconstruction, GIS, and industrial design applications.