Fujitsu Details 150 Armv9 Core Monaka CPU for AI and Datacenters

Servers
(Image credit: Shutterstock)

Fujitsu is working on its next generation processor for artificial intelligence (AI), high-performance computing (HPC), and datacenter applications, reports Monoist. The Monaka processor promises to offer formidable performance using its 150 enhanced Armv9 cores and the ability to use accelerators. Being one of the first 2nm datacenter CPUs, Monaka set to arrive in the fiscal year 2027, which starts on April 1, 2026, and ends on March 31, 2027.

The Fujitsu Monaka processor is set to pack approximately 150 cores based on the Armv9-A instruction set architecture with scalable vector extensions 2 (SVE2). The company does not define particular vector length design implementation, which can vary between 128 bits and 2048 bits (since the A64FX supports up to 512-bit vectors, we presume the new one will support similar or larger vectors). Those cores will be spread over multiple Core dies that will be accompanied by SRAM dies and I/O dies. The latter will support DDR5 memory and PCIe 6.0 connectivity with CXL 3.0 on top to connect various accelerators and extenders. The Core dies are said to be made on TSMC's 2nm fabrication process to maximize performance and transistor density as well as reduce power consumption. 

Monaka is set to use a 3D chiplet design, a disaggregated architecture that is set to enable scalability of the architecture. This is very similar to how AMD does it with its Ryzen and EPYC CPUs. Fujitsu does not provide any details right now. Interestingly, the company never mentions HBM memory in its slides, so it looks like the company plans to use DDR5 with Monaka, possibly in its MR-DIMM and MCR-DIMM implementations.

Although incorporation of things like SVE2 clearly points to supercomputers and HPC, Fujitsu itself does not call the CPU as a successor for its A64FX. In fact, post-Fugaku supercomputer is aimed to be launched around 2030, so it will likely incorporate a processor that will use technologies cultivated in Monaka, but not Monaka itself. Therefore, Monaka is built to compete for emerging datacenter market rather than to address the world's highest-performing supercomputers.

Energy efficiency stands as one of the key features of Monaka. Fujitsu sets an ambitious goal of doubling Monaka's energy efficiency in comparison to its competitors (as if the company knows how good will processors from competitors be in 2026 ~ 2027) and to use air cooling for these processors.

Fujitsu's Monaka will be used for a broad range of commercial AI, HPC, and datacenter deployments. Therefore, Monaka is set to have robust security mechanisms such as Armv9-A's CCA (Confidential Computing Architecture), which promises advanced security capabilities, such as enhanced workload isolation.

Fujitsu's Arm-based A64FX processor enabled the company to build a supercomputer that was the world's fastest supercomputer for two straight years until it fell to the second spot in June, 2022. Fujitsu positions Monaka processor to play a significant role in the company's future strategies, which includes AI and datacenters.

Anton Shilov
Contributing Writer

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

  • bit_user
    Fujitsu sets an ambitious goal of doubling Monaka's energy efficiency in comparison to its competitors (as if the company knows how good will processors from competitors be in 2026 ~ 2027)
    I think the statement is just meant to show the emphasis they're placing on efficiency. Perhaps, to some degree, it could be based on an assessment of x86 and tradeoffs made by Intel and AMD to optimize PPA & cost.

    However, if it's going to provide market-leading efficiency, then it will definitely have some form of in-package memory. In fact, by 2027 or so, CPUs without it might no longer be competitive in the datacenter market.
    Reply