Core Insights - Qualcomm is leveraging its experience in low-power CPU design from the mobile sector to enter the cloud computing market with a focus on cost-effective and energy-efficient server chips [1][2] - The Falkor CPU architecture is designed to meet performance benchmarks while maintaining low power consumption and silicon area requirements [4][7] - The Centriq 2400 series chips integrate up to 48 Falkor cores on a 398 mm² die, with a thermal design power (TDP) of 120 watts, indicating a power consumption of less than 2.5 watts per core [7][8] Performance and Architecture - Falkor is a 4-wide aarch64 core that runs on a 64-bit Arm instruction set, inheriting features from Qualcomm's previous mobile cores [4] - The architecture includes a unique L0 and L1 instruction cache setup, providing higher instruction cache capacity compared to contemporary cores [9][13] - Falkor's performance in SPEC CPU2017 tests shows a significant lead over Arm's Cortex A72, with a 21.6% advantage in integer operations and a 53.4% lead in floating-point operations [74][76] Memory and Cache Design - The L2 cache of Falkor is 512 KB with 8-way set associative design, serving as an intermediate cache between L1 and the on-chip network [49] - The Centriq's L3 cache capacity totals 60 MB across 12 slices, with a latency of 40.9 ns, which is relatively high compared to competitors [61][65] - Falkor's memory subsystem is designed to handle high bandwidth, with a theoretical memory bandwidth of 128 GB/s [69] System Architecture - The Centriq architecture does not support multi-socket configurations, limiting it to a maximum of 48 cores, which is a strategic decision to focus on mainstream cloud applications [71][72] - The interconnect topology utilizes a segmented ring design, allowing for efficient data transfer between cores and caches [56][58] Future Outlook - Qualcomm aims to re-enter the server market with new CPU offerings, building on the lessons learned from the Centriq experience [90][91] - The company is exploring partnerships for AI solutions, indicating a strategic pivot towards integrating AI capabilities into its server offerings [91]