SiPearl, a processor designer supported by the European Processor Initiative, will soon begin shipping its first Rhea processor for high-performance computing workloads. But the company is already developing its successor, currently called Rhea-2, which is expected to launch in 2026 in an exascale supercomputer.
SiPearl’s Rhea-1 data center-grade system-on-chip contains 72 off-the-shelf Arm Neoverse V1 cores, is designed for HPC and uses mesh networking. The CPU has a hybrid memory subsystem that supports HBM2E and DDR5 memory for high memory bandwidth and decent memory capacity, and supports PCIe interconnect based on the CXL protocol. The CPU was designed by contract chip designers and manufactured by TSMC using its N6 (6-nanometer) process technology.
The original Rhea was largely a product designed to prove that European company SiPearl could deliver data center-class processors. The CPU now powers Jupiter, Europe’s first exascale system, which uses nodes powered by four Rhea CPUs and NVIDIA’s H200 AI and HPC GPUs. Given that Rhea is SiPearl’s first processor, the project can be said to be fruitful.
With its 2ND In the new generation of Rhea processors, SiPearl must develop more competitive products. This may be why Rhea-2 will be implemented using dual chiplets. Such a design will enable SiPearl to pack more processing cores, thereby providing higher performance. Of course, it remains to be seen how many cores SiPearl plans to integrate into the Rhea 2, but at least the CPU company will be taking the same design approach as AMD and Intel.
Considering the timing of SiPearl’s Rhea 2 release and the company’s natural desire to maintain software compatibility with Rhea 1, it’s reasonable to expect that the processor will feature Arm’s Neoverse V3 cores as its second processor. Arm’s Neoverse V3 offers a considerable boost compared to Neoverse V2 (and V1) and can scale to up to 128 cores per socket, which should be comparable for HPC applications in 2025-2026 good.
While SiPearl will continue to develop CPUs, it remains to be seen whether EPI can deliver AI and HPC accelerators that are competitive with accelerators from NVIDIA, AMD and Intel.