The Maverick-2 Intelligent Compute Accelerator is a game-changing solution poised to redefine high-performance computing. By combining intelligent software-defined hardware with real-time optimization, it allows dynamic adaptability to the specific requirements of HPC and AI workloads. This novel architecture ventures beyond traditional CPU and GPU limitations, offering superior real-time performance and operational efficiency without the need for time-consuming code rewrites.
Incorporating a user-friendly developer toolchain, Maverick-2 supports various programming languages such as C/C++, Fortran, OpenMP, and Kokkos, while planning to extend support for CUDA and HIP/ROCm. This compatibility reduces porting efforts and boosts innovation, allowing developers to focus on producing science and insights faster. In addition, this hardware is designed with a more than fourfold performance-per-watt advantage over traditional GPUs, supported by TSMC's 5nm process technology.
Maverick-2 also sets itself apart with its unique ability to optimize hardware autonomously. Equipped with high-bandwidth HBM3E memory and PCIe Gen 5 interfaces, it's built for high-efficiency performance in both single-die and dual-die configurations, meeting modern data center demands while enabling sustainable innovation via reduced power consumption. This flexibility makes it an alluring option for researchers and organizations seeking future-proofed solutions.
