By Greg Gao

Veiglo(威固信息), a Chinese startup dedicated to solid-state storage and computing integration, recently launched a series of heterogeneous intelligent computing products RevHCU, a high-performance general-purpose parallel computing accelerator card, compatible with a variety of mainstream servers and software ecosystems. It can be widely used in AI training, high-performance computing, and other scenarios.

Heterogeneous computing is becoming the engine of the new intelligent era. The boom of 5G, the Internet of Things, artificial intelligence, cloud computing, the Internet of Everything, and intelligent perception generate data at high speed, and the global data volume is ushering in an explosive growth period. It means that the traditional single computing architecture is difficult to meet the information processing requirements. The diversity of application scenarios and data types in the intelligent era demands multiple computing power architectures. As a result, heterogeneous computing is expected to rise rapidly.

Veiglo was founded in 2013 based in Shanghai. With its in-depth research on solid-state storage technology, it expands big data, cloud computing, and artificial intelligence. It aims to provide hardware and software solutions that integrate intelligent storage and computing in the era of big data.

According to Veiglo, RevHCU series products have three major advantages: first, it covers a wide range of application and supports more than 200 artificial intelligence and general computing applications, and the number is still increasing; second, the development and test environment are easy to migrate, and it is compatible with various mainstream frameworks and CUDA programming; third, its performance is predictable and comparable to the hardware and software architecture of mainstream GPGPU products in the industry.

Specifically, RevHCU is a high-performance general-purpose parallel computing accelerator card that integrates high-performance co-processing unit chips, power modules, and other components. It is a parallel computing solution that achieves supercomputing performance, large-capacity memory, and convenience, said the company.

The key features of RevHCU include: The maximum computing power can reach 36TFLOPS@FP32, 147 TFLOPS@fp16/BF16, 295 TOPS@INT8; Comes with 32GB high-performance HBM2 memory; Compatible with mainstream general computing technology ecosystem.