In addition to domestically gradually withdrawing from the global supercomputing rankings, Japan has emerged in the TOP500 supercomputing list in recent years. Fugaku supercomputing once ranked first. Now Japan is starting to build the next generation supercomputing FugakuNEXT, with the goal of 100 times the performance of the current supercomputing.
The FugakuNEXT supercomputer is still jointly developed by Japan's RIKEN and Fujitsu, but this time it has to join the NVIDIA camp, changing the previous route of relying on self-developed floating point units. NVIDIA's GPU will become the main force in computing.
However, the specific GPU is still uncertain, because FugakuNEXT is targeting supercomputing in 2030. The GPU used will not come out until 2028 or even 2029. It is the next generation of Rubin and the next generation of the current Blackwell product.
The focus of FugakuNEXT is also AI performance.The performance of FP8 is expected to reach 600 exaFlops, striving to become the world's first Z-class supercomputer, which is the next generation of the current E-class supercomputer.The overall performance is 100 times higher than that of Fugaku, with the hardware structure achieving a 5-fold performance improvement and software optimization increasing the performance 10 to 20 times.
The CPU unit is MONAKA-X developed by Fujitsu, which is a derivative of MONAKA CPU. The latter is produced using a 2nm process, with 2 slots capable of 144 cores, supporting 12-channel DDR5 memory, PCIe 6.0 and other technologies. MONAKA-X should be stronger than this specification.
The FugakuNEXT supercomputer will complete the basic design by March 2026, the detailed design by March 2027, and finally start operation around 2030.
