Qualcomm Details its New Falkor CPU Core

Qualcomm has offered up a bunch of details on its new Falkor CPU core that is at the heart of the new Centriq 2400 SoC. That SoC is the world’s first 10nm server processor and will begin shipping later this year. Falkor has a pure 64-bit architecture and is fully ARMv8 compliant. Qualcomm says that its team designed th core to deliver high-performance and power efficiency to cloud datacenters.

The Falkor core duplex has a pair of custom Falkor CPUs inside, a shared L2 cache, and a shared bus interface to the Qualcomm System Bus ring. That creates a modular building block that is the foundation for the scalable 48-core Qualcomm Centriq 2400 SoC design. Qualcomm designed it to designed to deliver data center customers high-end compute performance using 4-issue, 8-dispatch heterogeneous pipeline.

Falkor is designed to optimize power per unit with variable length pipelines tuned per function to maximize throughput and minimize idle hardware. Qualcomm wrote, “Additionally, Falkors out-of-order and rename resources are sized to prevent instruction retirement from being in the performance-critical path, allowing unbridled usage of the multiple execution engines.”

The design is meant to handle large instruction footprints using split instruction cache with a single-cycle, low-power 24 KB L0 I-cache to complement the 64KB L1 I-cache. There is also a 32KB L1 D-cache with 3-cycle load-use latency. Falkor is aimed at cloud service environments and is ready for multi-tenant and virtualized workloads according to Qualcomm. The Centriq 2400 SoC architecture using this processor core will be unveiled along with product specifications in the coming months. There is no indication of what sort of timeframe that means.