After the launch of ARMv9 instruction set for the next decade in March last year, ARM launched the first generation of ARMv9-based products in May, including Cortex-X2, A710 and other CPUs, and today ARM launched Cortex-X3, A715 and A510 Refresh, a new generation of ARMv9 products. This generation of Cortex processors is geared towards smartphones, tablets and notebooks and other platforms, so there are more releases, as well as the new DSU-110 small and large core scheduling architecture, which can do up to 8+4+0 for a total of 12 cores.
First of all, look at the overall part of the CPU, Cortex-X3, A715 are Cortex-X2, A710 upgraded version, not only the peak performance increased by 25%, mixed performance and energy efficiency also increased by 20%.
It is worth mentioning that Cortex-X3, A715 have abandoned the aarch32 instruction set, that is, to give up 32 for and move to 64-bit architecture.
ARM’s A510 refresh, also known as A510 v2, not only improves performance and power efficiency, but also supports up to 12 core architectures, and is optionally compatible with the aarch32 instruction set, but is limited to a select group of options.
Overall, in the Cortex-X3, A715 and A510 v2 generation of architectures, ARM’s determination to fully shift to 64-bit is set, although there is no absolute elimination of 32-bit, leaving fewer and fewer areas for 32-bit compatibility.
Let’s take a look at the specific changes in the Cortex-X3, A715 and A510 v2 CPU architectures.
As a mega-core, Cortex-X3 is the main performance direction, and the architecture of this generation has been improved a lot. The decoder has been increased from 5 to 6 instructions per cycle, the chaotic execution window has been increased from 288 to 320, the integer ALU unit has been increased from 4 to 6, etc.
In other aspects, the L2 cache capacity of Cortex-X3 is also increased from 512KB to 1MB.
The final change is a big jump in performance, compared with the previous generation of Android flagship performance increased by 25%, compared with the laptop performance increased by 34% – the official ARM performance comparison is a bit confusing, just this generation performance improvement of 22%, 28% and other multiple claims, and IPC performance increased by 11%.
Next, the architecture of the A715 large core, which does not pursue absolute performance, but focuses more on energy efficiency, saving 20% of energy consumption than the A710 for the same performance, and improving performance by 5% over the A710 for the same power consumption.
In addition, because of the abandonment of the aarch32 instruction set, the area efficiency of the A715 core has risen significantly, requiring only 1/4 of the area of the A710. The overall performance is not greatly improved, but the energy efficiency and area efficiency have risen significantly.
The last is the architecture of the A510 small core, now is the A510 v2 compared to the 2020 A510 changes are not too big, the focus is still on optimizing energy efficiency, the same performance power consumption reduced by 5%, the frequency increased by 5%, so the previous 1.8GHz frequency can now be increased to about 1.9GHz.
There is also DSU-110, this is a new generation of core interconnect technology with more flexible core pairing, in addition to 1+3+4, there are 1+4+4, 2+2+4 and 8+4+0. The last configuration is for high-performance platforms like notebooks.
Other aspects, Cortex-X3, A715 and A510 v2 cores are also optimized for more advanced processes, such as Samsung TSMC’s 5nm and 4nm processes, and ARM also provides developers with a convenient development platform and tools VFP for better simulation testing, etc.
Finally, Cortex-X3, A715, A510 v2 and other CPU architectures will be available at the end of this year at the earliest, and not surprisingly, the Snapdragon 8 G2 at the end of the year will have to be replaced with a new architecture.
