Products & Solutions

AI Computing &
Storage Server
Solutions

Solution Overview

Current Opportunities &Challenges

Data Disruption Storage

With data volumes growing exponentially, storage systems are frequently failing.

Bandwidth response speed

Insufficient bandwidth slows down training progress, and IOPS bottlenecks cause computational lag.

Expansion space

The complexity of scaling up makes business expansion extremely difficult.

Our Solution

From ultra-high-performance distributed all-flash storage to massive, low-cost distributed storage, and from out-of-the-box all-in-one machines for large models to a one-stop intelligent computing management platform, Tshumart provides a full-stack AI solution that enables seamless data storage and computing power scheduling.

Core Architecture

Benchmark in Speed and Reliability

As the industry's fastest distributed all-flash file storage system, the performance of Tshumart UbiPower 18000 is remarkable: a single node bandwidth of up to 90GB/s effortlessly handles massive data throughput during AI training; 3 million IOPS per node eliminates lag in high-frequency data access. More impressively, through EC and multi-replica data redundancy technologies, it achieves multi-level reliability across nodes, components, and I/O. Even if 8 nodes fail simultaneously, no data is lost.

Solution Advantages

Fully Domestic

Hardware fully supports domestic processors and operating system platforms, ensuring a stable and reliable supply chain.

Parallel Architecture

Metadata caching operates in parallel with clients, enabling direct parallel data reads from multiple data nodes, achieving low latency and high throughput.

System Reliability

Data sharding and reconstruction algorithms reduce reconstruction time to 15 minutes per TB, while EC reconstruction ensures data redundancy remains intact.