Learn about the ARMv9 second generation Cortex core: New performance

ARM's Cortex-X3, A715 and A510 Refresh cores are based on the V9 instruction set.

According to official summary data, the Cortex-X3 improves peak performance by 22% and IPC (same-frequency efficiency) by 11%. The Cortex-A715 has a 20% increase in energy efficiency and a 5% increase in performance; The A510 Refresh reduces power consumption by 5%.

Note that each of the percentages mentioned above has a different unit, which on the one hand reflects the direction of ARM's upgrade in different cores, and also seems to show the bottleneck of its upgrade.

In terms of specifications, the Cortex-x3 decoder has increased from 5 to 6 instructions per cycle, out-of-order execution Windows from 288 to 320, integer ALU units from 4 to 6, and L2 cache capacity from 512KB to 1MB.

For A715 large core, ARM says that it does not pursue absolute performance, but pays more attention to energy efficiency, saving 20% energy consumption compared with A710 under the same performance, and improving 5% performance compared with A710 under the same power consumption. At the same time, the A715 abandoned the compatibility of 32-bit instruction sets, which greatly improved the area efficiency of its kernel, requiring only one quarter of the area of the A710.

Finally, the remade A510 small core focuses on energy efficiency optimization, with the same performance power consumption reduced by 5% and frequency increased by 5%.

The new Cortex core will be designed for smartphones, tablets and laptops, and the new DSU-110 core scheduling architecture will support 1+4+4, 2+2+4 and 8+4+0 in addition to the current mainstream 1+3+4 core configuration.

The Cortex-X3, A715 and A510 V2 cores are optimized for more advanced processes, such as Samsung TSMC's 5nm and 4nm processes, and ARM provides developers with a convenient development platform and tool VFP for better simulation testing.

Chips based on the new architecture are expected to hit the market as soon as the end of the year, with the first batch likely to be used by qualcomm's Snapdragon 8 Gen2 and Mediatek's Phecda flagship chips.