RISC-V

RISC-V 22nm Process, 222 Square Millimeters Processor Attains 4096 Core

From the past two years in the market of CPU manufacturing, two famous names are in the limelight. RISC-V and ARM are the two popular CPU architectures presently. The RISC-V processor can be customized entirely due to its open-source. Several domestic companies like Alibaba use the processor by RISC-V. In 2020, at the HotChips conference, a startup company named Manticore disclosed its RISC-V architecture processor design. The RISC-V architecture processor design by Manticore was code-named by ‘Ariane,’ and it has approximately 4096 cores.

RISC-V

RISC-V processor adopts the renowned Multi-Chipset design concept, where each chip gets integrated and packaged with four small chips. Then the small chips get divided into four quadrants, and each of the quadrants also gets to be continuously divided. In this processor, there are 32 clusters where each of them contains 8 cores. Therefore the processor has 8x32x4x4= 4096 cores. Apart from the 4096 cores, the RISC-V processor’s every chipset includes 27 MB L2 cache, 8 GB HBM2 memory, and PCIe x 16 controllers. Between various clusters, the direct bandwidth is approximately 64TB/s. Between the chiplets, there are high-bandwidth serial interconnection bus, short distance, and multiple channels.

RISC-V

Amazingly, this type of processor plans to use GlobalFoundries that manufactures 22nm processes. However, it is not highly advanced but offers several advantages such as low cost, high yield capability. ^The overall chip area of this processor is 17.9 x 14.9 = 222 square mm. It is made by following the concept of AMD Ryzen APU, manufactured by GlobalFoundries. The only difference is AMD Ryzen 3000 APU has 8 CPUs and 11 GPU.

When it comes to performance, the peak floating-point performance of every chipset can be exceeded to 4TFlops. Though the index of power consumption is not yet revealed. Manticore also mentioned that the Ariane is ideal for mainly floating-point computing application scenarios such as machine learning, scientific computing, and data analysis.