Fujitsu is preparing a new Arm processor with the code name “MONAKA” to replace the A64FX for the next generation of supercomputers

Fujitsu is an enterprise with extensive experience in the field of supercomputers. They collaborated with Arm to develop the SVE instruction set, which is primarily targeted toward HPC workloads. Originally an extension of Armv8, it eventually became a part of the Armv9 standard specification. The first processor architecture to use the SVE instruction set was the chip used in the former world’s number one supercomputer system, “Fugaku”, which was Fujitsu’s A64FX.

According to reports from theregister, Fujitsu’s next-generation Arm processor for high-performance computing is currently in the development stage. It will have greater computing power than the current A64FX and is known as “MONAKA”. The specific specifications for MONAKA, such as core count, frequency, and manufacturing process, are still unclear. The new Arm processor will be based on Armv9 or higher and is expected to debut in 2027. It will power the “Fugaku NEXT” supercomputer system and will be responsible for high-performance computing, artificial intelligence, and data analysis workloads. It is expected to have 1.7 times better performance and twice the power efficiency than current solutions.

The A64FX used in “Fugaku” is manufactured using TSMC’s 7nm FinFET process, containing 8.786 billion transistors and 596 signal pins. It integrates 52 cores, including 48 compute cores and 4 management cores with identical structures. All 52 cores are divided into four groups, each group containing 13 cores that share an 8MB L2 cache. The interconnect between the groups is Fujitsu’s second-generation TOFU, which is a 6D mesh/torus on-chip interconnect network. It is also equipped with 32GB HBM2 memory, 16 PCIe Gen3 channels, and a storage bandwidth of 1024GB/s. It can provide the performance of 2.7 TFLOPS@64bit and 21.6 TFLOPS@8bit.