InspireSemi’s powerful Thunderbird “supercomputer cluster-on-a-chip” accelerated computing architecture is based on an array of thousands of modern, efficient, and powerful 64-bit CPU cores, tightly integrated with high speed memory and networking. It is an ideal and versatile platform for high-performance computing (HPC), AI/machine learning, graph analytics, and other compute-intensive applications.
Leveraging the modern open-source RISC-V CPU ISA (instruction set architecture) provides access to an established, thriving ecosystem of software, libraries, and development tools. Our developer-friendly all-CPU programming model eliminates vendor lock-in and greatly simplifies software development, QA, and maintenance, since there is no need to support multiple and often proprietary software stacks (e.g.- x86, CUDA, ROCM, SYCL, FPGAs, etc.). This also enables the combination of HPC & AI on one platform for deeper insights and bigger advancements.
Our highly optimized, high speed mesh network fabric provides extreme bandwidth and low latency communications between cores. Tightly integrated with the cores and distributed memory, it removes crucial bottlenecks from applications that depend on close cooperation between many independent threads. Arrays of hundreds of chips – up to one million processor cores – connect seamlessly via high-speed serial transceivers.
Compared to a modern datacenter GPU, Thunderbird provides all the raw computational power with far broader application to real world codes. Our advanced network fabric enables far better cooperation between cores. And our roots in extremely power- and cost-sensitive blockchain computing enable us to provide this performance with considerably higher energy efficiency than leading GPU’s.
Raw performance: 24 TFLOPS (FP64)
Energy efficiency: 50 GFLOPS/Watt (FP64)
Utilization: network reduces latency 20x
Deterministic + Predictable Performance
Value: exceptional Performance/$