Fujitsu and AMD announced plans on Friday to develop a new, more power-efficient AI and HPC computing platform that will pair the Japanese technology provider’s next-generation CPUs with House of Zen’s Instinct accelerators.

Fujitsu’s Arm-based A64FX processors powered the number-ranked Fugaku supercomputer, until AMD’s larger Epyc and Instinct-based Frontier system dethroned the beast in early 2022.

“By combining AMD’s innovative GPU technology with Fujitsu’s energy-efficient, high-performance Fujitsu-Monaka processor, we aim to create an environment where more companies can use AI while reducing power consumption by data centers,” said Vivek Mahajan, head of company. Technology Officer at Fujitsu said this in a statement.

While we still don’t know much about how the two companies plan to merge these parts, a slide deck released alongside the announcement (PDF) suggests that Fujitsu is focusing on CPU-based HPC and AI inference and GPU-based based training.

However, we do know a thing or two about Fujitsu’s Monaka processor (PDF). Planned to be built on a combination of 2nm and 5nm process technology, the chip is very similar to AMD’s own HPC-centric Epyc-X chips and will use a chiplet architecture with different compute, SRAM and I/O dies that are stitched together on a silicon chip. intermediary. Similar to AMD’s Genoa-X, the compute and SRAM tiles will be stacked and connected using through-silicon vias (TSVs).

Fujitsu’s Monaka CPU looks a lot like Arm-take on AMD’s cache-stacked X-chips. – Click to enlarge

Monaka will include up to 144 ArmV9 cores with support for SVE2 vector extensions and confidential computing capabilities. The latter makes sense because Fujitsu has made a habit of deploying its chips not only in HPC clusters, but also in the cloud.

Compared to the 48-core A64FX, Monaka also supports dual-socket configurations, allowing denser nodes.

In terms of memory and I/O, the part is expected to feature 12 DDR5 memory channels, PCIe 6.0 and CXL 3.0. This is another notable departure from the fast HBM2 memory used on the A64FX.

Fujitsu is setting some rather lofty goals for the chip, predicting that it will offer twice the application performance and performance per watt of competing CPUs when it hits the market in 2027.

The fact that Monaka is still years away means we know even less about the Instinct accelerators planned for the collaboration. Depending on when the two start rolling out systems in 2027, AMD’s MI400 series platform, due out in 2026, could be a candidate. But aside from the fact that it will use AMD’s “next-gen architecture”, we don’t know anything about the part.

AMD said The registry they could not develop the collaboration further than what was announced. We also reached out to Fujitsu for comment on the timeline and integration plans, but had not heard back at time of publication.

That said, the idea of ​​pairing GPUs with Arm-based chips is certainly not new. For example, before Nvidia launched its Grace-Hopper Superchips, Nvidia worked with Ampere on an Arm-based GPU server. ®