Arm unveiled a 64-bit, real-time capable Cortex-R82 processor aimed at high-end storage controllers that offers an optional MMU to run full Linux distributions with support up to 1TB DRAM and optional Neon for ML workloads.

Arm announced its first 64-bit, Linux-capable Cortex-R MCU to enable more advanced storage controllers. The Cortex-R82 offers greater speed, lower latency, and a higher level of deterministic support than the Cortex-R8. It also provides 40 address bits that support up to 1TB memory address space compared to the previous 4GB limit.

The Cortex-R82 provides up to 2x “performance uplift, depending on the workload, compared to previous Cortex-R generations,” claims Arm. The Cortex-R82 processor provides a 21 percent performance uplift over Cortex-A55 when running SPECINT2006 benchmarks. A typical quad-core cluster of 5nm fabricated cores would operate at above 1.8GHz with 3.41 / 4.32 / 8.67 DMIPS/MHz performance.

Cortex-R82 performance compared to Cortex-R8 in Arm benchmarks
(click image to enlarge)

Back in 2013, when Arm announced its 32-bit ARMv8-R architecture for more advanced Cortex-R designs, the chipmaker announced plans to eventually design a 64-bit version of Cortex-R with a full memory management unit (MMU). This would enable full Linux distributions instead of the scaled-down uClinux, which is available on some higher-end Cortex-R and Cortex-M MCUs. When equipped with the optional MMU, which works in conjunction with the optional Memory Protection Unit (MPU) already available on Cortex-R MCUs, Cortex-R82 achieves that reality.

Arm is aiming the design at computational storage applications in which local processing and analytics of huge datasets is needed in a hurry. Cortex-R processors are primarily used in cellular modem subsystems, but are increasingly found in HDD and SSD controllers.

Local computational storage applications running Linux “opens huge opportunities across applications including IoT, ML and edge computing,” says Arm. As with local edge processing, local storage processing would reduce the cost and latency involved with sending data to the cloud while offering increased security and privacy.

Potential applications include database acceleration, video transcoding for streaming, and transportation. Arm offers the example of a modern aircraft, which generates terabytes of data evert day. The day is usually offloaded for analysis at the end of the day. By offering real-time analysis directly on the hard drive, the computation could happen in between flights to ensure up-to-date safety checks.

Cortex-R82 block diagram (left) and conceptual diagram showing how storage-focused cores can switch over to computation duty during off hours
(click images to enlarge)

The support for Linux and other Cortex-A compatible rich OSes enables more sophisticated applications for reallocating compute resources dynamically based upon changing workloads, says Arm. Linux also provides developers with tools and technologies such as Docker and Kubernetes, without requiring an intermediary CPU that could slow computation.

Cortex-R82 provides compatibility with security technologies such as Arm TrustZone, “ensuring isolation of the storage controller firmware from other Linux or real-time workloads,” says Arm. Cortex-R82 also adds optional Neon support, which will “greatly accelerate machine learning (ML) workloads.”

 
Multi-core and multi-OS support

The Cortex-R82 architecture’s support for up to octa-core configurations enables applications such as mixed RTOS and Linux systems. This would offer a simpler solution and reduced cost compared to mixing Cortex-A and Cortex-R or -M cores, says Arm.

Cortex-R82 has three Exception levels (ELs). The highest EL2 level enables a secure enclave and separation/isolation of virtual machines for OEM code and customer code. In one scenario, an MPU context running at EL2 handles context switches between MPU and MMU. It can then context at EL1 with OEM and/or OS code while leaving user code to run at the least secure EL0. When a real-time event occurs, the processor can switch from Linux to RTOS then back to Linux.

Multi-core capability could enable parking lot surveillance applications in which license plate information is collected for later used for billing, says Arm. During the day most of the cores would be used for storage, but at night, they could switch over to computational duty to process the data for billing and perform data analysis and machine learning processing.

 
Further information

No pricing or availability information was provided for the first Cortex-R82 processors. More information may be found on in Arm’s announcement and product page.