The Cortex-X2 was chastised for not improving on the Cortex-power X1’s efficiency. This could explain why the Snapdragon 8 Gen 1 and Exynos 2200, as well as the Dimensity 9000, do poorly in power conversation testing. Unfortunately, the Snapdragon 8 Gen 2, Exynos 2300, and Dimensity 10000 are not likely to help problems, as all three will reportedly use a power-hungry Cortex-X3.
According to a rumour originating in Taiwan, Qualcomm, Samsung, and MediaTek have examined early Cortex-X3 samples and believe that additional tuning is necessary. The performance difference between Cortex-X2 and Cortex-A5 is not very large at the time of writing, but power draw is. The explanation for this discrepancy is that AI performance has grown by over 100% but fundamental IPC performance has remained unchanged.
The Cortex-X3 consumes 10% more power than the Cortex-X2 in its current condition when running at 3.00GHz or higher, but it is unclear whether the Cortex-X2 is running at the same frequency as the Cortex-X3. These early prototypes were reportedly put through their paces on TSMC and Samsung’s next-generation production processes. Samsung already has a 4nm node, but there’s no indication on whether the Korean company has a more developed version of the production process than TSMC.
In any event, the Snapdragon 8 Gen 2, Exynos 2300, and Dimensity 10000 will be affected by the Cortex-greater X3’s power consumption on an upgraded manufacturing process if no changes are made. Because there is no alternative, Qualcomm, MediaTek, and Samsung have had little choice but to remain with ARM’s designs thus far. Qualcomm is likely to adopt a unique design, similar to Apple’s A-series, but this could take until 2024.
Due to the development expenditures associated with transitioning to a custom design, MediaTek has little choice in the issue. The good news is that ARM’s Cortex-X3 launch is still months away, so improved versions of the processor could reduce power usage, making the Snapdragon 8 Gen 2, Exynos 2300, and Dimensity 10000 less inefficient.
also read:
