Sugon unveils a new generation of general-purpose high-performance computing platform, achieving AVX-512 instruction set compatibility for the first time in China.
Sugon has officially launched its new generation of general-purpose high-performance computing (HPC) platform, marking a significant breakthrough in the domestic computing ecosystem. A major highlight of this release is that it is the first domestically developed general-purpose computing platform to natively support the AVX-512 instruction set while being fully adapted to the x86 software ecosystem. Because AVX-512 is the mainstream global standard for HPC, this native compatibility allows existing professional software in fields like meteorological simulation, materials calculation, and biomedical research to run directly without code modification or recompilation. This drastically reduces migration time and transformation costs for institutions transitioning to domestic computing power.
In terms of hardware performance, the platform is powered by a domestically produced 128-core general-purpose CPU capable of delivering 10 TFLOPS of FP64 double-precision computing power per processor. Official tests indicate that compared to its predecessor, the platform nearly doubles its HPL double-precision floating-point and STREAM memory access performance, while overall scientific and industrial simulation applications see an average improvement of 80% to 100%, matching the specifications of leading international flagship products. To support massive parallel computing clusters with tens of thousands of cards, the system employs a full-stack collaborative optimization solution. This includes BurstBuffer caching, SocketDirect communication, ParaStor F9000 all-flash distributed storage, and a self-developed scaleFabric high-speed interconnect switch, achieving an ultra-low end-to-end communication latency of just 0.93 microseconds.
To accommodate diverse data center requirements, Sugon has introduced three standardized computing node architectures featuring different cooling technologies. The lineup features standard 2U air-cooled servers for small to medium research clusters, alongside high-density cold-plate liquid-cooled cabinets that lower Power Usage Effectiveness (PUE) to 1.08 for medium industrial centers. For ultra-high-density national supercomputing hubs, the platform offers third-generation immersion phase-change liquid-cooled cabinets with a PUE as low as 1.04. Designed for high-precision general-purpose computing, the platform targets critical sectors such as geophysical exploration, industrial design simulation, integrated circuit R&D, life sciences, and weather forecasting. With complete delivery capabilities ranging from single nodes to large-scale clusters, the platform is currently available for deployment by domestic research institutes, manufacturing enterprises, and local computing platforms.
.In terms of hardware performance, the platform is powered by a domestically produced 128-core general-purpose CPU capable of delivering 10 TFLOPS of FP64 double-precision computing power per processor. Official tests indicate that compared to its predecessor, the platform nearly doubles its HPL double-precision floating-point and STREAM memory access performance, while overall scientific and industrial simulation applications see an average improvement of 80% to 100%, matching the specifications of leading international flagship products. To support massive parallel computing clusters with tens of thousands of cards, the system employs a full-stack collaborative optimization solution. This includes BurstBuffer caching, SocketDirect communication, ParaStor F9000 all-flash distributed storage, and a self-developed scaleFabric high-speed interconnect switch, achieving an ultra-low end-to-end communication latency of just 0.93 microseconds.
To accommodate diverse data center requirements, Sugon has introduced three standardized computing node architectures featuring different cooling technologies. The lineup features standard 2U air-cooled servers for small to medium research clusters, alongside high-density cold-plate liquid-cooled cabinets that lower Power Usage Effectiveness (PUE) to 1.08 for medium industrial centers. For ultra-high-density national supercomputing hubs, the platform offers third-generation immersion phase-change liquid-cooled cabinets with a PUE as low as 1.04. Designed for high-precision general-purpose computing, the platform targets critical sectors such as geophysical exploration, industrial design simulation, integrated circuit R&D, life sciences, and weather forecasting. With complete delivery capabilities ranging from single nodes to large-scale clusters, the platform is currently available for deployment by domestic research institutes, manufacturing enterprises, and local computing platforms.
