Fujitsu’s new “MONAKA” processor specifications revealed

Fujitsu, a venerable stalwart in the realm of supercomputing, collaborated with Arm to develop the SVE instruction set, tailored primarily for HPC workloads. Originally conceived as an extension to Armv8, it subsequently integrated into the Armv9 standard specification. The globally renowned supercomputer, “Fugaku”, which once topped global rankings, is powered by Fujitsu’s A64FX – the inaugural processor to utilize the SVE instruction set within an Arm architecture.

Fujitsu’s forthcoming Arm processor for AI and HPC applications, codenamed “MONAKA”, is under development, poised to supersede the existing A64FX. As per reports from Monoist, “MONAKA” is equipped with approximately 150 enhanced Armv9 cores, possessing the capability to employ accelerators, thereby promising formidable performance. It incorporates the SVE2 instruction set, which features variable-length vector registers, ranging from a minimum of 128 bits to an expansive 2048 bits.

Insights suggest that “MONAKA” will adopt a 3D chip design, reminiscent of a modular architecture that facilitates scalability, bearing parallels with AMD’s Ryzen/EPYC processors. This denotes that the enhanced Armv9 cores will be dispersed across multiple chip modules, accompanied by SRAM chips and I/O chips. It supports DDR5 memory and PCIe 6.0 standard connectivity, harnessing the potential of CXL 3.0 to link a myriad of accelerators and expansion modules. However, clarity remains elusive regarding the incorporation of HBM-type memory. Additionally, “MONAKA” boasts a robust security framework, including support for the Arm Confidential Compute Architecture (Arm CCA), offering a plethora of sophisticated security features.

Destined to be among the pioneer data center processors manufactured using the 2nm process, “MONAKA” is anticipated to be unveiled in the fiscal year 2027 (spanning April 1, 2026, to March 31, 2027). It’s slated to propel the “Fugaku NEXT” supercomputer system set to debut in 2030, promising enhanced scalability and energy efficiency. In juxtaposition with contemporary solutions, the “Fugaku NEXT” supercomputer system is projected to lead with a 1.7-fold advantage in application performance and double the performance per watt.